turbo machinery presentation collection

Michael J. Drosjack is a Senior Principalin the Rotating Equipment Departmentat Shell Global Solutions (US) Inc., inHouston, Texas. He is responsible forproviding technical support for rotating andreciprocating machinery to Shell and Shellaffiliated companies worldwide, as well ascommercial customers. Since joining Shellin 1975, he has had assignments onprojects involving specification, evaluation,

installation, and startup of machinery along with extensive fieldtroubleshooting, particularly in the area of vibration measurement,vibration analysis, and rotordynamics.

Dr. Drosjack received his B.S. degree (Mechanical Engineering,1970) from Carnegie-Mellon University, and his M.S. (1971) andPh.D. (1974) degrees (Mechanical Engineering) from The OhioState University. He is a member of ASME, the Vibration Institute,the Machinery Subcommittee of the Ethylene Products Committee,participates in API task forces, and has been a speaker andpanelist for NPRA. He has been a Turbomachinery SymposiumAdvisory Committee member since 1986.

John W. Felten is a Senior Staff Engineer inthe Rotating Equipment Group at ShellGlobal Solutions (US) Inc., in Houston, Texas.He is responsible for providing technicalsupport for rotating and reciprocatingequipment to Shell and Shell affiliatedcompanies worldwide, as well as commercialcustomers. He joined Shell in 1987 andhas had assignments in multiple refineriesencompassing day-to-day rotating equipment

support and troubleshooting, new projects, and revamps.Mr. Felten received his B.S. degree (Engineering with a

Mechanical Specialty, 1986) from Colorado School of Mines. Hehas presented expander-related material in both NPRA and ASME.

George Seamon is a Principal DesignEngineer for Dresser-Rand Company, inOlean, New York. For the last 19 years, hehas been responsible for the aerodynamicand mechanical design and development ofhot gas expanders for FCC and nitric acidservice. Prior to that, he spent six years onthe design of gas turbines and four years onthe design of the GHH type hot gasexpander. Before joining Dresser-Rand, Mr.

Seamon worked for 10 years with General Electric and Pratt &Whitney on heat transfer, aerodynamic, and mechanical design ofthe turbine section of jet engines.

Mr. Seamon graduated with a BSME/AE degree fromNorthwestern University (1967).

ABSTRACT

A methodology is presented to identify the most likely cause offailure for a cat cracker hot gas expander in which the “classic”failure mechanisms did not fit the evidence collected after thefailure. Because the classic root cause analysis (RCA) did notprovide a probable theory for the failure cause, a comprehensiveengineering analysis was required to positively rule out causes thatproved improbable and focus on the most likely theory.

Included is a review of the failure mechanisms considered in thesearch for the cause of a blade failure in the expander. This paperprovides a discussion of how common failure mechanisms wereeliminated leading to a theory in which catalyst deposits that builtup and eventually shed off of expander blades led to supersonicflows in a converging-diverging nozzle created by blade deposits.The cycling of the blade loading from successive buildup andshedding of multiple deposits on the same blades and the resultingchanges from subsonic to supersonic flows in the blade throats aresuspected to have created blade cracks in an unexpected locationleading to the eventual failure of this machine.

7

WHAT TO DO WHEN A “CLASSIC” RCA DOESN’T GIVE THE ANSWER—APPLY DETAILED ENGINEERING ANALYSIS

byMichael J. Drosjack

Senior Principal, Rotating Equipment Engineering

John W. FeltenSenior Staff Engineer, Rotating Equipment Engineering

Shell Global Solutions (US) Inc.

Houston, Texas

George H. SeamonPrincipal Design Engineer, Expanders

andTim R. Griffin

Senior Technical Specialist, Expanders and Axial Compressors

Dresser-Rand Company

Olean, New York

INTRODUCTION

A typical cat cracker process unit has the process flow configurationshown in Figure 1. The primary machine is the main air blower thatprovides air for both combustion and transport of the catalyst andfeed. A reaction is induced in either the riser or reactor andthe desired products separated in downstream units. Residualhydrocarbon is burned off the catalyst in the regenerator after thereaction occurs. The gas coming off the regenerator is hot (1250 to1350ºF) and at a pressure higher than the carbon monoxide (CO)boiler, the next downstream device. In many cat crackers, a hot gasexpander is used as an energy recovery device in this stream. Thelayout of a typical expander is shown in Figure 2. This machine, ineffect a one stage power turbine, has a very difficult environmentto work in. It is subjected to hot gas with trace elements andvarying amounts of extremely hard catalyst fines.

Figure 1. Typical Refinery Cat Cracker Process.

Figure 2. Typical FCC Expander Configuration.

The unit is a Dresser-Rand (Ingersoll-Rand) Model E-148.Design conditions were:

• Inlet pressure: 44.14 psia

• Inlet temperature: 1200ºF

• Flow rate: 775,000 lbs/hr

• Exhaust pressure: 15.56 psia

• Power generated: 24,400 hp

The expander is directly coupled to the main air blower. Alsoincluded in the train were a motor/generator to address variationsin the power generated or consumed and a helper steam turbine foruse in startup. Shown in Figure 3 is a photo of the train.

Figure 3. Photo of the Expander Train.

Due to severe imposed duties, hot gas expanders have a historyof being problematic devices. History indicates that operationalfailures are more common than in machines in more benignoperating environments. This installation had experienced severaloperational failures in the past and the refinery was able to react tothe failure in a practiced manner.

On June 6, 2003, a unit shutdown occurred. A high radialvibration trip was the result of a sudden jump in vibration causedby a blade failure. Before the shutdown, there were no obviousindications that this machine was in distress either from thevibration or process data monitoring. Loss of this machine causedthe fluid catalytic cracking (FCC) unit to be shut down. Repair andrestart resulted in seven-figure production losses.

As a result of the failure, an RCA was immediately initiated.Available data from the vibration and process monitoring systemswere gathered. Analysis began to identify anomalies or “tell-tale”indications that something was wrong or had been changing withthe machine or process. Once the machine was entered, visualexamination was made, measurements gathered, and photos taken.The rotor and stator were moved to the shop for additionalexamination—original equipment manufacturer (OEM) designstaff and additional technical staff were deployed to perform theanalysis. The broken blade was found in the piping system—arather unusual location. All of the blades were subjected tocomprehensive metallurgical analysis.

Because hot gas expanders have a well-documented failurehistory with numerous RCAs and engineering studies, the analystshad an expectation that they would reach a NORMAL conclusion.That is, a failure mode and cause that had been previouslyevaluated on many other failures. That expectation was not met.

Metallurgical work indicated that the failure was caused byfatigue, a typical occurrence in expander blade failures. However,the failure location on the blade and the initiation site were not

PROCEEDINGS OF THE THIRTY-FIFTH TURBOMACHINERY SYMPOSIUM • 20068

consistent with the normal loading associated with expanderfailures. In addition to the failed blade, two other blades werefound that were cracked in a similar location.

Historical vibration and process data were reviewed verycarefully. There were a few anomalies, but none that really tied toNORMAL causes. As a result, the RCA team progressed throughthe standard RCA practices and found themselves looking forloads and forces outside those considered typical. Thus, they hadto deviate from the NORMAL analysis protocols and employedmore sophisticated analytical tools.

ROOT CAUSE FAILURE ANALYSIS

Organization of RCA and Participants

The blade failure on the expander was an extremely sudden andunexpected event. Parallel to beginning the repair on the machine,an RCA team was formed to determine what had happened andwhy. As with any RCA investigation, it is vitally important to forma team with a breadth of experience and viewpoints. To this end,this RCA comprised personnel from refinery staff, the end-user,and the OEMs’ expander engineering resources. This personnelrepresented engineering disciplines such as:

• Process engineering

• Reliability engineering

• Rotating equipment engineering

• Metallurgical engineering

• Expander design

• Operations

With the basic RCA team formed, the charter and ground ruleswere set to define its function. Of vital importance to any team arecommunication and appropriate team interaction. There were thenormal cautions to ensure open dialogue between differentengineering groups and disciplines. An additional dimension inthis case was the inclusion of team members from multiplecompanies. Because there was a shared goal in understanding thefailure, communication was kept open and information freelyshared. This resulted in a very cohesive RCA team.

A refinery senior reliability engineer led the RCA team. Afterpreliminary face-to-face meetings held at the refinery to initiate theeffort, the RCA team continued its work with teleconferencing andnetworking via an Internet-based desktop sharing program. Thispermitted the team to review data and common files togetherwithout having to travel across the country to sit face-to-face.

This paper will detail the data-gathering and analytical workperformed to reach the ultimate conclusion. Included in thisanalysis are:

• Vibration and process data review

• Photos and physical examination of the machine

• Metallurgical failure analysis

• Classic blade stress analysis

• Finite element analysis

• Evaluation of classic blade failure mechanisms

• Load evaluation—stress testing of the blade

• Computational fluid dynamics (CFD) analysis and load calculationfor abnormal throat geometries

• Identification of potential abnormal loading scenarios

Vibration and Process Data Review

The FCC unit was started in November 2002, after completionof a planned maintenance outage. The startup of the unit went

smoothly as all equipment returned to service and operatedproperly. The process unit was lined out and the process conditionsfor the expander were stabilized. To illustrate the process conditionsupon the restart, Figures 4, 5, and 6 depict flue gas flow, expanderinlet pressure, and expander inlet temperature, respectively.

Figure 4. Flue Gas Flow.

Figure 5. Expander Inlet Pressure.

Figure 6. Expander Inlet Temperature.

Along with the stable process operating conditions, the expanderperformed well mechanically. Vibration levels were primarily lessthan 1.0 mil for the duration of the run (Figure 7).

Figure 7. Expander Disk End Vibration.

The vibration levels for this expander run were relatively lowcompared to previous runs as shown in Figure 8. During thescheduled maintenance outage in 2002, normal maintenanceactivities were performed (rotor change out), and there were no


9

design changes made to the expander, cat cracking unit regenerator,or third stage separator system.

Figure 8. Historical Expander Disk End Vibration Amplitudes.

About one-and-a-half months after the restart, it was noted thatthe expander inlet flow function was trending downward as shownin Figure 9. Flow function is a measure analogous to an orificecoefficient and is related to the effective opening of the expanderflow path. Mathematically, the expander flow function is calculatedby using inlet temperature, inlet pressure, and mass flow per thefollowing equation:

Figure 9. Expander Flow Function.

A downward trend in the flow function tends to indicate that theflow path area was getting smaller. This reduction in area wasassumed to be caused by deposition of catalyst fines. This expanderhad experienced deposition previously. Operations had procedures toroutinely inject walnut hulls online to “blast” deposition off. Visualinspection through the expander view ports verified that the walnuthulls were cleaning the shroud area effectively. The deposition wasreasoned to be occurring on the stator vanes and/or rotor blades.Injection of walnut hulls on a routine basis was continued during therun to try and maintain the flow path.

Without warning on the morning of June 6, 2003, the expandertripped because of high radial vibration.

Photos and Physical Examination of the Machine

The machine was made ready for inspection and maintenance.Deposition was visually confirmed as a significant amount ofdeposits were found within the expander case. In a normalshutdown of the expander, the machine is run to deinventory the catcracking unit and deposits would be blown out of the machinebefore inspection. Because the shutdown occurred suddenly andthe machine did not continue to run, the deposits that had formedin the flow path were not removed.

When the machine was opened for inspection, it was discoveredthat the rotor had experienced a blade failure as shown in Figure10. In addition to this blade failure, the stator was damaged, as wasthe shroud, diffuser, and inner exhaust case. The failed blade

remnant was not found within the expander casing. After athorough inspection, it was found in a vertical section of theexhaust duct approximately 30 feet above the expander exhaustflange. The failed blade was located on an expansion joint innerflange surface along with foreign objects. The objects, shown inFigure 11, were actually two pieces of what looked to be either awasher or some type of fabricated part. A metallurgical analysis ofthe object confirmed that it was neither a component of theexpander, nor a component of the associated piping and vessels.

Figure 10. Rotor with Failed Blade.

Figure 11. Foreign Objects Found Near Blade.

Two other blades were found to have cracks near the platform.These blades were located 12 blades and 22 blades, respectively,away from the failed blade (there are 62 blades in this rotor). Thesecracks were not visible to the naked eye until dye penetrant testswere conducted on each blade (Figure 12).

Figure 12. Dye Penetrant Inspection Results of One of the BladesFound to Be Cracked but Intact.


( )Flow Function

Mass Flow Inlet Temperature

Inlet essure=

× 0 5.

Pr

There was significant damage to the components within the flowpath of the expander. The stator was damaged with trailing edgesbroken on a majority of the vanes (Figure 13).

Figure 13. Stator with Numerous Damaged Vanes.

The other components that were damaged, i.e., shroud, diffuser,and inner exhaust case all showed indications of being struck bythe failed blade.

Failure Analysis

Upon completion of the initial inspection, the rotor with bladesand the complete stator were moved from the field for furtherinspection and analysis. The analysis team worked together toanalyze these components. Metallography work was performed bythe end-user and the OEM to ensure that all possible avenueswere explored. This also reduced the time required to completethe analysis.

The metallography work yielded the following:

• Blade 1—The failed blade: The fracture surface of blade 1was examined by the scanning electron microscope (SEM).Chemical analysis was performed on the blade’s fracturesurface before cleaning. The uncleaned blade had areas withlarge amounts of silicon, aluminum, and oxygen, in addition tothe expected elements. This was attributed to catalyst depositson the fracture. Polished metallurgical samples indicated thatsmall amounts of sulfur were present on the fracture face.

The fracture itself was confirmed to be the result of high cyclefatigue followed by ductile overload. Fatigue striations wereconfirmed in several areas on the fracture face, although largeportions of the fracture were damaged and “unreadable”because of abrasion, handling, and catalyst deposits that couldnot be removed during the cleaning process. The origin of thefracture was clearly identified. The coating was still intact,although somewhat thinner near the origin of the crack. Noevidence of any other cracking mechanism (environmentalcracking) was detected at the origin or anywhere else onthe fracture.

The fracture surface of blade 1 is shown in Figure 14. Theplatform and the base of the airfoil are visible in the figure. Thecrack initiated by fatigue on the suction side of the blade, about13/8 inches from the leading edge. The crack advanced by fatigueabout 80 percent of the way through the blade thickness, towardthe pressure side, and then arrested for sufficient time to form abluish oxide on the crack surface before beginning to advanceagain by fatigue. The crack propagated almost through theentire blade thickness by fatigue before converting to ductilerupture at which point the blade separated quickly from its base.

Figure 14. Fracture Surface of Blade 1.

The crack was examined in the unetched condition with theoptical microscope as well as the scanning electron microscope.The crack appeared relatively straight (there is a slight curvaturetoward the top of the blade) and primarily transgranular. Therewere a few secondary cracks that had initiated off the maincrack, but these had only progressed a grain or two away fromthe main crack. Deposits were found in the crack, although itcould not be classified as ìdeposit-filled.î Some of these depositscould be attributed to polishing compounds that were used, butevidence of sulfur was present in several locations along the wallof the crack as well as in some of the secondary cracks. Thetransgranular nature of the crack combined with the observationof striations on the fracture surface indicates a failure mechanismof mechanical fatigue.

• Blades 13 and 47—Additional blades found with crack nearplatform: Similar analysis was performed on blades 13 and 47. Thecrack identified through dye penetrant testing was opened andanalyzed using an SEM. The cracking mechanism was determinedto be high cycle fatigue, with the fracture origin in approximatelythe same location as the fracture origin in blade 1. No evidencewas found of any erosion or impact damage that might haveinitiated cracking.

• Stator vanes: Forty out of the 50 stator vanes were founddamaged. After inspecting the stator vanes, two vanes wereselected for further analysis. The two stator vanes selected, 7 and37, had fracture surfaces typical of the entire group. The fracturesurfaces were characterized in the SEM. Some areas on vane 7were distinguishable as brittle fracture, but most of the surfaces ofboth vanes were covered with catalyst deposits. Cleaning inacetone and in inhibited hydrogen chloride (HCL) did not removethis deposit.

• Hardness and impact testing: The blades and vanes were testedfor hardness as well as Charpy impact results. In all cases thetests yielded results that correlated with expected designproperties. There were no deviations found in blade, vane, orcoating material properties.

• Conclusions of metallography: Three rotor blades crackedbecause of high-cycle fatigue. All three cracks initiated atapproximately the same location, on the suction side between theleading edge of the blade and the point of maximum camber, justabove the platform.

All three blades came from different heats and the chemicalanalysis and hardness results indicated normal chemistry. Thisyields the conclusion that the original condition of the blades didnot play a role in the failure.


11

The fatigue crack on the severed blade was observed to betransgranular over its entire length. Secondary cracking was almostnonexistent. Although sulfur was present on the fracture surface, thepresence of sulfur is ubiquitous in expanders, and the conclusion isthat sulfur played no role in the blade failure.

Classic Blade Stress Analysis

A Goodman diagram is a classic tool used in analysis of a failureof this type. While a finite element analysis provides more detailedquantitative results, the Goodman diagrams permit a means tocompare to other expanders (and gas turbines) in the fleet of anOEM. A modified Goodman diagram is shown in Figure 15.

Figure 15. Modified Goodman Diagram Showing Airfoil Root Stresses.

The vertical axis is alternating stresses with the maximumvalue being the endurance limit of the material at operatingtemperature. One-third of that endurance limit is identified asthe allowable alternating stress when steady-state stresses arezero. The horizontal axis plots steady-state stresses; for hightemperature operation the maximum value with zero alternatingstress is 100,000 hours creep rupture capability. Eighty percentof the 100,000 hour rupture strength is considered the allowablesteady-state stress when alternating stresses are zero. A lineconnects the two points of allowable alternating and steady-statestresses; it represents the maximum design limit in the case ofcombined alternating and steady stress. The expected stresslevels for a machine are plotted on this diagram. Distance fromthe limit is representative of safety margin.

Figure 16 provides a comparison of the stress levels in this unit toothers in operation. From this plot it can be seen that the design ofthis particular expander appears to be conservative. There are manyother expanders and gas turbines operating with calculated stressesmuch closer to the allowable limits, i.e., lower safety factors.

Figure 16. Comparison of the Normal Stresses in the Failed Unit toHigher Stressed Units in Operation.

The calculated stresses displayed in these Goodman diagramsinclude simple pull over area calculations to which gas bendingforces are added. In the airfoil root these are standard classicstress calculations. The firtree area is very geometry specificand a stress concentration factor is used that is based on theradius of curvature of the firtree neck. Each lug has added stresscomponents from local bending and shear. The combinedstresses for the Goodman diagrams are evaluated at six locationsof high stress on the blades. Evaluation of past failures indicatesthat these six areas cover the highest percentage of previousfailures.

The Goodman diagram for the airfoil root is the most criticalarea of evaluation for this failure. In the airfoil, the leadingedge and trailing edge normally have the highest stress becausegas-bending loads add to the centrifugal stress at theselocations. The max camber point is about halfway around thesuction (convex) side of the airfoil. It gets its name from beingthe maximum distance from the minimum bending axis, whichis a line running through the airfoil center of gravity and almostparallel to a line that is tangent to the leading edge and trailingedge. Figure 17 shows the minimum bending axis and the maxcamber point. The effect of gas bending is to lessen thecentrifugal stresses at the max camber point.

Figure 17. Photo Showing Location of Minimum Bending Axis andMaximum Camber Point.

Evaluation of recorded failures indicates that the majority offailures that have occurred in the airfoil section are at either theleading or trailing edge. A few have been at the max camberpoint. A crack initiation in the location observed on the failedblade is unique and cannot be easily explained by classic stressanalysis or normal blade loading. That is, stresses caused bynormal loading are always going to be higher at the threelocations plotted in the Goodman diagram than they would be atthe point of crack initiation in this failure. This means the classicstress calculations would not predict a failure in the locationwhere this one occurred.

Finite Element Analysis

A finite element analysis (FEA) model was used to performa more detailed analytical examination of the stresses inthe blade under various loading conditions. The results forthe model with normal steady-state loads are shown in Figures18 and 19.


Figure 18. Steady-State Stress Analysis Results, A.

Figure 19. Steady-State Stress Analysis Results, B.

Load Evaluation—Stress Testing of Blading

The FEA model was calibrated using a bench test. A blade takenout of the unit had four straingauges installed around the perimeterof the root at the following locations:

• Leading edge

• Crack initiation site

• Max camber

• Trailing edge

Figure 20 shows these strain gauge locations. Figure 21 showsthe locations where point loads were applied to the blade tip andFigure 22 shows one of the loads being applied during the benchtest. Strains were measured with each load case and are shown inTable 1. The FEA model was run at zero rpm with point loadssimilar to the bench tests. Nodes of the finite elements closest tothe straingauge location were chosen and the calculated stressesnoted for each load case. Strain was calculated and compareddirectly to the measured strains. These values are also shown inTables 1 and 2 and demonstrate a very good correlation of themodel to the bench test. This confirms that the results from theFEA model can be taken as reliable. This provides a foundation forusing the FEA model for further investigation.

Figure 20. Photo Showing Location of Straingauges for Bench Test.

Figure 21. Locations of Load Application for Bench Test.

Figure 22. Photo of a Load Being Applied to a Blade During theBench Test.


13

Table 1. Measure Strain from Bench Test.

Table 2. Calculated Strain from FEA Model.

Classic Failure Mechanisms

On the basis of prior failure histories with FCC expanders,there is a methodology that is followed to address “classic”failure mechanisms. One of the most common mechanismsis the excitation of blade resonance frequencies by eitherstrut-passing-frequency-flow excitations or stator-driven-flowexcitations.

FCC expanders are further challenged by potential naturalfrequency interferences because, in the course of a run, naturalfrequencies of the blades can be lowered or raised by metalremoval driven by catalyst erosion.

The design of this expander incorporated uneven noseconestrut spacing to minimize the risk of a strong excitation.In addition:

• Three other sets of identical blades operated successfully in thismachine without fatigue failures (from 1997 to 2002).

• The current failure was not at the leading edge, trailing edge, ormax camber point where a first bending mode would show up ashigh stress.

• The rotor blades had no metal loss, and therefore had notchanged in natural frequency during the course of the run.

Thus the strut passing resonance excitation was eliminated fromconsideration.

A harmonic excitation caused by a variation in the statorvane throat geometry was evaluated. A Fourier analysis wasperformed using the stator throat dimensions of the originalstator to determine the potential excitation frequency andamplitude. The results of this analysis are shown in Figure 23.They show that the harmonics identified were not strongenough to have caused the failure. And if there had been astrong excitation, a blade failure would have occurred in thefirst few days of operation.

Figure 23. Fourier Analysis Results for Stator Vane ThroatVariations—Undamaged Stator.

Upon disassembly, the stator was found to have significantdamage. A theory was tested to evaluate a potential excitation fromthe damaged stator (assuming the damage had been a precursor tothe blade failure). The stator damage was surveyed in detail and theresulting throat dimensions analyzed to predict potential harmonicexcitations. The results are shown in Figure 24. Although the exitpulses from the stator are understandably stronger in this case thanfor the new stator, the pulses predicted at the blade’s naturalfrequencies were too weak to have been the cause of the failure.

Figure 24. Fourier Analysis Results for Stator Vane ThroatVariations—Damaged Stator.

A failure caused by any stator damage should have resulted inblade crack initiation at the leading edge, trailing edge, or maxcamber. Thus the stator-driven forces were eliminated as aprincipal cause of failure.

Foreign object damage to the blades also was considered as apossible failure cause, in particular because foreign material wasfound in the line downstream of the unit. However, the two otherblades with cracks did not show any evidence of impact such thatforeign object damage was ruled out as a likely cause of failure.

Blade tip rubs on shroud catalyst deposits have caused failuresin FCC expanders. This unit had experienced blade tip rub damagein the past from contact with catalyst deposits on the shroud.However, during this run there were no physical indications oftip rubs.

Initial Results of RCA—No Answer

Data, such as discussed above, were gathered and evaluated bythe RCA team. The evaluation protocol was practiced and appliedrapidly. A cause-and-effect diagram was utilized to record resultsand validate conclusions.

Data were used to validate or eliminate branches of the trees.Issues concerning the strength of the blades and vanes were


eliminated as possible causes with metallurgical data. The resultsof tests conducted to confirm mechanical strength as well ascoating properties were within the original design parameters.Analysis of the stator indicated that stator damage did notprecede a rotor blade failure. Ring testing of rotor blades forresonance confirmed that the blades’ natural frequencies had notbeen shifted by erosion into a region where they could be excitedby normal operation. Detailed inspection of the machineassembly verified that there were no installation or assemblyerrors that could have lead to the blade failure. The possibility ofthe failure being the result of a foreign object was investigatedand rejected. It was discovered that the failure originated in anarea of the blade that was not highly loaded under normaloperation. The crack that ultimately led to the failure was foundto have originated on the suction side of the airfoil between theleading edge and max camber point. As the “classic” modes offailure were eliminated, they were ìcrossed outî on the cause-and-effect diagram. Figure 25 displays a portion of the diagram asmodes of failure were eliminated.

Figure 25. Sample Portion of the Cause-and-Effect Diagram.

At the end of the “classic RCA,” the team had crossed off thepotential causes without reaching a positive conclusion. Thisfailure did not follow the rules of most of the failures within theexperience base of the analysis team.

Because the classic modes of failure were being ruled out as theanalysis continued, it became apparent that an unusual event musthave occurred to initiate the blade failure. The next steps for theroot cause analysis team were to look outside classic failure modes.The investigation needed to go deeper into possible causes. Thisrequired a focus on scenarios that simultaneously led to high bladestresses in the crack initiation sites while not causing increasedstresses at the expected blade failure locations.

SECOND PHASE OF THE RCA

Identifying Potential Loading Scenarios

The approach taken was to try to first identify a loading scenariothat would create higher stresses on the blade at this specificfailure location.

A tool was developed to facilitate evaluation of load cases thatincluded centrifugal and alternating gas-bending loads. It wascalled the “stress visualizer,” and it permitted more cases to beexplored in a short time. This approach provided the flexibility ofapplying calculated stresses from nonstandard gas loads on top ofthe more precise FEA steady-state stresses.

Figure 26 shows a comparison of stress safety factor for twodifferent load cases. As expected, in the base case (normalflows/loads) the crack initiation site had significant safety margin,while the leading and trailing edges had the lowest safety margin.However, if loads were adjusted, the margin at the crack initiationsite could be significantly reduced under potential loadingscenarios as shown in Figure 26. The pressure loads in this second

case were approximated based on the results of a CFD analysis thatincluded a convergent/divergent passage created by a blade depositthat is discussed in more detail in the following section. TheseCFD results provided the most plausible theory, and that theory isconsidered to be the most likely failure scenario.

Figure 26. Stress Visualizer Results for Two Load Cases.

Identifying Potential Loading Mechanisms

Extensive CFD analysis was done with normal airfoil shapes andvarying inlet and outlet flow conditions. None of those casesgenerated gas loadings that resulted in a predicted failure at thecrack initiation site. Theories were developed to explain how acatalyst deposit might affect flow patterns across the blades. Atfirst, consideration was given to uniform buildup on the statorvanes and/or rotor blades. This did bring about some shifts inincidence angles, stage reaction, and pressure profiles, but still wasinsufficient to cause the failures.

Typically, catalyst deposits on rotor blades resulted in higherrotor vibrations; however, in this case it was theorized that thesesame deposits could have aerodynamic effects on blade loading.Therefore, this CFD analysis was used to evaluate the impact of theactual shape and location of deposits. Figure 27, although not toscale, illustrates the size and shape of actual deposits found in otherunits. Investigations of expander deposits show that they arealways made up of catalyst less than two microns in diameter.Particles of that size approach the blade row with a similartrajectory to the flue gas molecules. However, their weight-to-dragratio is high so they do not make the 100 degree turn in the bladerow. Thus they plow through the turn and pile up on the bladepressure side. There is little buildup upstream of where the depositis depicted because that surface is shrouded by the leading edge.There is also little buildup near the trailing edge because theadjacent blade casts a “shadow” over that region. This results in ashallow mound shape that will be in position to bring about aconverging diverging (C/D) passageway upstream of the actualblade throat.

Figure 27. Sketch Showing Typical Size and Shape of DepositsFormed on Rotor Blades.


15

If blade deposits form a converging diverging (C/D) passage-way, the flow can easily reach the speed of sound in the new throatthat is upstream of the blade row discharge. In supersonic flow, adiverging passage produces the opposite effect on flow as it doesfor subsonic flow. The flow downstream of a C/D nozzle minimumarea will accelerate, if the conditions are right.

A CFD model was then built with a similar deposit shape to thatseen in other units, and the results on the first pass were dramatic.Figure 28 shows the Mach number variation for the baseline modelwith no deposit. In contrast, Figure 29 shows the same plot with atypical deposit where the exit Mach number is 1.9.

Figure 28. Mach Number Distribution without Deposit.

Figure 29. Mach Number Distribution with Deposit.

To demonstrate the effects on blade surface static pressure,Figure 30 was developed via compressible flow equations. Itcompares a blade under normal Mach number distribution to one

with a C/D passage. It can be seen that the static pressure can be aslow as 4.3 psia at Mach 1.9. This meant that the rotor blades couldbe subjected to large variations in pressure loading with variationsin catalyst deposit size and shape.

Figure 30. Static Pressure Across Blade with and without Deposit.

There is a mechanism for changing the shape of the depositsimilar to the phenomenon of snow buildup on a mountainside. Inboth cases the bond is mechanical and the buildup will continue togrow until the weight of the deposit is too much for the strength ofthe bond. In the case of a snow buildup, that force is gravity, andthe snow pile slides off in the form of an avalanche. On a rotorblade, the catalyst is in a centrifugal force field that is about 11,000times the force of gravity. Initially the bond strength of the depositcan be great enough to overcome the centrifugal load. However, asthe deposits grow in size eventually the increased centrifugal loadcauses deposits to shed.

Introduction of a Sonic Shock Theory into the RCA

Because the CFD analysis was able to predict abnormally highMach number due to blade deposits, the RCA went back throughthe process data. The intent of the re-review of the process data wasto determine if process conditions could help prove or disprove theCFD theory on supersonic flow in the blade row. Available datawould not allow for exact measurements of Mach number withinthe expander. However there were enough data to understand therelative trend in Mach number. Figure 31 illustrates the trend inMach number for the expander two months before the failure.

Figure 31. Calculated Mach Number at Blade Midspan (No Deposits).

The calculated Mach number shown in Figure 31 used availablepressure data and assumed a clean expander. Based on a cleanexpander, a threshold “redline” was plotted to represent when the


flow would have exceeded Mach 1.0 over a significant portion ofthe blade trailing edge. As shown in Figure 31 the Mach numbermakes a sudden change upward about three days prior to thefailure. This significant change just days before the failure led tothe highest Mach number recorded during the entire run ofthe machine. The data showed that the Mach number remainedsupersonic for a period of time just before the failure.

These process data supported the theory of a high Mach numberin the blade row. With the previously discussed review of flowfunction indicating deposition within the expander flowpath, theRCA was able to add process data support to the CFD modeling.

As this theory of deposition creating a C/D passage leading tohigh Mach number in the blade row progressed, the cause-and-effect trees were updated to reflect the investigation. An exampleof the cause-and-effect tree for supersonic flow within the bladerow is shown in Figure 32.

Figure 32. Sample Portion of the Cause-and-Effect DiagramAddressing Sonic Shock Theories.

CONCLUSIONS

Failure Mode

The suspected primary cause of the failure was identified ascatalyst deposition in a local rotor blade passageway producinga converging/diverging nozzle that drives flow supersonic. Thisphenomenon can produce buffeting that creates high alternatingstresses at the crack initiation site. The stresses alternate because ofunsteady shock wave formation at those times when the flowis supersonic or when flow fluctuates between subsonic andsupersonic. Unstable supersonic flow can be produced by a singlecondition or a combination of conditions. For instance, it canhappen because of wake disturbances (e.g., broken stator vanes),circumferential variations in catalyst deposits on the shroud,disturbances in the upstream flow, flow variations from day-to-night, etc., either alone or (more likely) in combination. Thus theinvestigation led to the discovery of this sonic shock theory thatcould plausibly lead to crack formation in the blades at the locationin which this failure occurred. After having developed thispotential theory, further review of the process data showing highMach numbers just prior to failure provided additional confidencein the theory’s probability.

Future Action

In order to minimize the potential for a recurrence of this failurein the future, the analysis team generated the following list ofmitigation steps:

• Monitor the expander for:• Mach number• Efficiency• Flow function• Stack emissions/opacity• Vibration

Monitor cat carryover/distribution• Continue frequent walnut hull injection• Monitor catalyst quality (particle size)

Assemble an expander reliability team that meets monthly (morefrequently if required) and reviews the results of the data generated.This team will have the authority to mandate the appropriate levelof flow path cleaning to avoid another failure of this type.

In addition, more comprehensive solution development wasundertaken as a separate effort.

Closure

This effort demonstrated that the “old reliable” classic approachesto engineering might be insufficient to solve problems. The RCAfollowed well-known, historically proven steps that were unable togenerate an answer. Instead, the study team had to employ moresophisticated engineering tools in an almost “experimental” mode topermit a positive conclusion to be reached. The result, through theuse of state-of-the-art engineering tools, was the identification of avery plausible failure mode and cause that had not been consideredin the past by the investigators.

The future is exciting.

BIBLIOGRAPHY

Balfoort, J. P., 1974, “Power Recovery Systems and Hot GasExpanders,” Proceedings of the Third TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas, pp. 1-7.

Drosjack, M. J. and Felten, J. W., 2004, “CCU ExpanderReliability: A Machinery Issue or a Process EngineeringIssue,” 2004 NPRA FCCU Conference.

O’Dea, D. M., 1980, “Reliable Operation of Flue Gas Expanders,”Chemical Engineering Progress, AIChE.

Stettenbenz, L. M., 1970, “Minimizing Erosion and Afterburn in thePower Recover Gas Expander,” ASME Paper No. 70-PET-6.

ACKNOWLEDGEMENT

The authors would like to acknowledge the dedicated personnelat the Shell companies and Dresser-Rand who contributed to thisRCA including Shell’s Art Tinney (Refinery ReliabilityEngineering), Jason Kaufman (Refinery Pressure EquipmentEngineering), Michael McGranahan (Refinery Operations), MarieMiglin (Metallurgical Engineering), and Anthony Soby (RotatingEquipment Engineering). Also to be recognized for their significantcontributions are Dresser-Rand’s Greg Holland (Aero-ThermoDesign), Jason Kopko (Aero-Thermo Design), Paul Chilcott (SolidMechanics Engineering), and Dan Claus (MetallurgicalEngineering). The authors would also like to thank the Shellcompanies and Dresser-Rand for permission to publish this work.

The technical information contained in this presentation iscontrolled for export by the US Government. Please do notforward the contents without obtaining authorization from theauthor/presenter.


17

•

Kazuhiro Takeda is presently theVibration and Control System LaboratoryResearch Manager in Mitsubishi HeavyIndustries Hiroshima Research andDevelopment Center, in Hiroshima, Japan.He is working mainly on research anddevelopment of process control systems forchemical and power plants.

Dr. Takeda received a B.S. and M.S.degree (Mechanical Engineering) fromKanazawa University in Japan, and

received a Ph.D. degree (System Engineering) from OkayamaUniversity in Japan.

Kengo Hirano is presently an Instrumentand Control Engineer working inMitsubishi Heavy Industries TurboMachinery Engineering Department plantdesign section, in Hiroshima, Japan. Heis an I&C specialist for compressorprocess control.

Mr. Hirano graduated from MiyakonojoNational College of Technology (ElectricEngineering).

ABSTRACT

Centrifugal compressors are widely used in petrochemicalindustries and natural gas fields. One of the major control problemsassociated with these compressors is pressure fluctuation. In recentyears, various control methods have been designed and applied inthe field to reduce pressure fluctuation and improve the overalloperating stability.

This paper describes recent advanced complex gas compressioncontrol techniques used for centrifugal compressors, and averification method using a newly developed control system anddynamic simulator.

One of the control techniques used for fuel gas compressors thatsupply fuel gas to gas turbine combined cycle (GTCC) powerplants is described in this paper. Two compressors supply the fuelgas to two independent gas turbines through one common header.

Each gas turbine is related to each individual compressor, whichhas to supply fuel gas for the individual gas turbine according tothe load by modulating the control valve opening. The compressoralso has to stabilize the common header pressure fluctuation bychanging the gas turbine loads. For this process, one controltechnique is proposed with feed forward (FF) control for the loadchanging of gas turbine and feedback (FB) control for commonheader pressure stabilization, as well as application of the abovetechnique in a plant operating in the field. With the FF and FBcontrol, if load shedding of one gas turbine occurs and the gasturbine load suddenly goes down, the other gas turbine cancontinue operating safely.

The compressor control performance is verified with a compressordynamic process simulator during the design and manufacturingstages prior to shipping. In particular, in order to ensure smoothstartup at site, the control panels are connected to the processsimulator and the above control technique is applied. Confirmationthat all individual functions are working properly is madebefore shipping.

INTRODUCTIONIn recent years, centrifugal compressors have been used for

various gas compression processes. In those processes, aconventional control technique, i.e., the combination of mastercontroller and antisurge controller, has proved very effective.

But other compressor types, like fuel gas compressors, requirequick response and reliable controllability. For these compressors,there is a limit to how far the controllability can be tuned usingconventional control techniques.

In this paper, an advanced control technique using feedforward control is proposed. This control technique places moreemphasis on controllability. First, details of the controltechnique are introduced. Then the controllability of the controlsystem is checked by studying the dynamic simulation results.After confirmation of results, the control system is applied inthe field, where the controllability is verified by studying fieldtest data.

FUEL GAS COMPRESSOR WITH COMMON HEADERFigure 1 shows a typical integrally geared fuel gas compressor

with inlet guide vane (IGV). This compressor has two compressionstages, one gear wheel, and one pinion gear. Each gear is mountedon each shaft. The round arrows show the rotating direction of eachshaft, and straight arrows show the gas flow direction.

25

ADVANCED CONTROL TECHNIQUE OF CENTRIFUGALCOMPRESSOR FOR COMPLEX GAS COMPRESSION PROCESSES

byKazuhiro Takeda

Research Manager, Research and Development Center

andKengo Hirano

Instrument and Control Engineer, Plant Design Section

Mitsubishi Heavy Industries, Ltd.

Hiroshima, Japan

Figure 1. Schematic Diagram for Centrifugal Compressor.

Table 1 shows the fundamental features of the fuel gas compressor.The handling gas is natural gas from the well. The compressor hastwo stages, and is driven by a three-phase induction motor. TheIGV controls the compressor flow. The seal type is a dry gas seal.

Table 1. Fundamental Feature of Compressor.

Figure 2 shows the compressor arrangement. The main motordrives the wheel gear. The gear train, consisting of the wheel gearand the pinion gear drives the two-stage compressors. The mainmotor also drives the lube oil pump mounted on another pinionshaft. The gas enters into the first stage of the compressor throughthe IGV, is compressed in the first and second stages, and then thedischarge gas goes out to the outlet section.

Figure 2. Compressor Arrangement.

Individual fuel gas compressors are usually dedicated to eachgas turbine in a GTCC plant so that both can work as a pair. Hence,

the control philosophy and the operating sequence are simple.When changing the gas turbine is required, the compressoroperation can simply be changed simultaneously by reading theheader pressure of the related gas turbines.

However, when the fuel gas is provided to multiple gas turbinesthrough a common header, the compressor control system will bedifferent. Figure 3 shows an example of a fuel gas compressorarrangement for a GTCC plant, with two gas turbines and three fuelgas compressors connected by a common header. The fuel gascompressor flow capacity is the same as that of the gas turbine. One ofthe fuel gas compressors is a standby unit. When both gas turbines areunder operation fed by fuel gas through the common header, a suddenchange in operation by one gas turbine will cause a pressurefluctuation in the common header. The above pressure fluctuation willaffect the operating performance of the other gas turbines.

Figure 3. Fuel Gas Compressor Units Arrangement for GTCCPlant Using Common Header.

In this condition, there is a limit to using the conventional controlmethod (like the master controller method for controlling thecommon header pressure in order to get effective controllability ofthe common header pressure fluctuation).

COMPRESSOR ADVANCED FLOW CONTROL

Figure 4 shows an example of a performance curve of acompressor driven by a constant speed motor with the individualinlet guide vane opened. The horizontal axis shows the flow rate,and the vertical axis shows the pressure ratio. The IGV opens from20 percent to 100 percent to supply the required flow in this case.When the IGV position is 20 percent opened (at minimum openingpoint), the antisurge valve (ASV) will open to reduce the flow ofthe gas turbine. The pressure ratio of the fuel gas compressor isalmost constant in this operating condition. Also control valvescontrol the common header pressure, so the compressor operatingpoint shifts toward the dotted line of the performance curve.

Figure 4. Compressor Performance Curve and ASV/IGV Control Band.


On the performance curve, the IGV and ASV control bands aredivided into two by a boundary line, formed by the cross section pointof the pressure ratio line (dotted line) and IGV minimum opening line(dotted/dashed line) for ASV control. One band is on the left side ofthe boundary line and the other band is on the right side.

Therefore, at first, the load signals from the related gas turbines areentered in the function generator called “feedforward load function.”

Figure 5 shows the above-proposed control system. Figure 5 (1)shows the schematic control diagram for the compressor unit, and(2) shows the load sharing control diagram.

Figure 5. Compressor Units’ Load Sharing Control Method byFF+FB/FF Selector.

This control system has two operating modes: FF+FB mode andFF mode. The compressor control panel or distributed controlsystems (DCS) selects these modes with a selector switch (SEL)mounted on the master control panel or DCS. Then the gas turbineselector (SEL) selects the gas turbine and the compressor unit. Theoutput of SEL gives a FB value for FF+FB mode, and a neutralvalue (50) for FF mode. The FF+FB mode is controlled by thepressure control value (FB) and feedforward load function (Fxload)output value (FF), whereas the FF mode is controlled directly byFF value. This control system avoids any interference between thecompressor units. The manipulating value (MV0) is calculated(Equation [1] below) in the calculation unit (CAL) as shown inFigure 5 (1) using FB value and FF value.

MV0 is limited to the range of zero percent to 100 percent. If FFvalue deviates from the actual and design conditions, FB valuescan cover the full range of FF value according to this formula. TheIGV and ASV positions are decided using the split range functiongenerator Fx1 and Fx2.

Figure 6 shows an example of feedforward load function. Thehorizontal axis shows the load signal of the fuel gas flow ofthe assigned gas turbine. The vertical axis shows the FF signal asa percentage.

Figure 6. Example of Feedforward Load Function (Fxload).

There are two boundary lines in Figure 6. One is the boundaryline (fine double-dotted/dashed line) for the minimum ASVposition, and the other is the boundary line (thick dotted/dashedline) for opening/closing the ASV and IGV as shown in Figure 4.

In Figure 6, zero percent load signal means that the gas turbinestops. When the G/T controller gives zero percent load signal, thecompressor is in its full recycle operating condition, and thedischarge pressure of the compressor should be the same as the setvalue of the header pressure. The reason for this is that if the gasturbine starts, the compressor has to supply fuel gas to the gasturbine as soon as possible. Therefore, the boundary line for theminimum ASV is used as a limit line to keep the discharge pressureof the compressor above the set value of the header pressure.

Figure 7 shows an example of the split range function for theASV (Fx1) and IGV (Fx2). The horizontal axis shows the output ofthe calculation (CAL) unit (MV0) and the vertical axis shows thepositions of the ASV (MV1) and IGV (MV2).

Figure 7. Example of Split Range Function for ASV and IGV.

Thus, using the combination of the feedforward load function,ASV function (Fx1), and IGV function (Fx2), the ASV controls thelow load from zero percent to the boundary flow, by adjusting theFF signal from minimum ASV position to 50 percent open, and theIGV controls the high load from the boundary flow to maximumflow by adjusting the FF signal from 50 percent to 100 percent.

Figure 8 shows an example of the compressor’s load sharingcontrol system by a master controller for the common headerpressure. This alternate control system can be used instead of theabove-mentioned system as shown in Figure 5. The difference


27

( )MV FF FB0 2 50= + × −

between the two is that the present control system uses the mastercontroller for the common header pressure instead of theFF+FB/FF selector shown in Figure 8 (2). In this way, the FB valueis calculated by the master controller, and is used in the CAL unitof all compressor units shown in Figure 8 (1).

Figure 8. Compressor Units’ Load Sharing Control Method byMaster Controller for Common Header Pressure.

These load sharing control systems are different from commonload sharing control system. These load sharing systems concentrateon the controllability rather than the efficiency of the compressors.Fundamentally, the required gas turbine fuel flow modulates thecontrol valves directly using feedforward load function with thiscontrol system. The FB value only adjusts the common headerpressure to the set point.

SIMULATION

In order to verify the proposed FF+FB/FF selector controlsystem, a dynamic simulation is performed. Some examples of theverification simulation are shown below in G/T parallel operatingcondition.

Case 1: G/T B Trip

• Before• G/T A at 75 percent load relates to compressor (comp) A at

75 percent load in FF+FB mode• G/T B at 55 percent load relates to comp B at 55 percent load

in FF mode

After• G/T A keeps the load close to the load condition before

G/T B Trip• G/T B load suddenly goes down to zero percent at trip and

at the same time comp B stops.

Case 2: G/T A and B Load Shedding at the Same Time

• Before• G/T A at 100 percent load relates to comp A in FF+FB mode• G/T B at 100 percent load relates to comp B in FF mode

• After• G/T A and B load suddenly goes to 30 percent (minimum

load) during load shedding

Figure 9 shows the simulation result of Case 1. These graphsshow trend data of the G/T demand flow, header pressure,compressor flow, and recycle flow. In this simulation, G/T B tripoccurs 10 seconds after simulation start. After G/T B trip, A takescare of the G/T load and adjusts the header pressure with theFF+FB function. Comp B stops and decreases the compressor flowby opening the ASV and coasting down.

Figure 9. Simulation Result for Case 1 (G/T B Trip).

From the result of this G/T B trip simulation, it is found that thecommon header pressure fluctuation is within a small range(10.66/20.29 bar [19.57/24.21 psi] from SV), and the pressurebecame stable in three minutes.

Figure 10 shows the simulation result of Case 2. In this simulation,G/T A and B load shedding occurs 10 seconds after simulationstart. Comp A and B load goes down to 30 percent with feedforward load function following G/T load, and MV0 becomessmall. The ASV opens to increase the flow, so comp A and B


•

recycle flow increases and comp A adjusts the header pressure withthe FF+FB function.

Figure 10. Simulation Result for Case 2 (G/T A and B LoadShedding at the Same Time).

From the result of this G/T A and B load shedding simulation, thecommon header pressure fluctuation is found to be within 3 bar (43.5psi) (11.86/21.10 bar [126.98/215.95 psi] from SV), and thepressure became stable in two minutes in spite of the large gas turbine(G/T) load change from 100 percent to 30 percent in a relatively shorttime. Table 2 shows a summary of these simulation results.

Table 2. Simulation Result.

VERIFICATION OF CONTROL SYSTEMUSING DYNAMIC SIMULATOR

The FF+FB/FF selection control system is applied to theoperating compressor units. The control system is installed in theprogrammable logic controller (PLC). Before this control system isapplied to the actual compressor, the verification test using thedynamic simulator is performed in order to verify the controlfunction and sequence during the factory acceptance test (FAT).

Figure 11 shows the system configuration of the test condition.The control system is programmed using a PLC program loader,and is downloaded to the PLC located in the control panel. Theoperating condition can be confirmed in the operating panel. Thesimulation model of the gas turbines, compressors, and otherauxiliaries is installed in the process simulator that is the same as

the dynamic simulator used in the previous “SIMULATION”section, and run in real time. The switching panel initiates the startpermissive condition, trip cause, etc. The PLC, process simulator,and switch panel are interconnected using hard wire cables.

Figure 11. Schematic System Configuration Diagram of SimulationTest Using Dynamic Simulator.

Figure 12 shows the simulation test condition using the dynamicsimulator to verify the PLC program. In this verification test, thedetail sequence and actual control functions can be verified. Table3 shows examples of the test item using the dynamic simulator.

Figure 12. Simulation Test Condition Using Dynamic Simulator toVerify PLC Program.

Table 3. Examples of Simulation Test Item.


29

An example of the run back test shown in Table 3, Item No. 5, isdescribed as follows.

Case 3: G/T - A and B Run Back by Comp Trip

• Before• G/T A at 75 percent load relates to comp A at 75 percent load

in FF+FB mode• G/T B at 75 percent load relates to comp B at 75 percent load

in FF mode

• After• G/T A and B load suddenly goes down to 30 percent

(minimum load) by run back when comp A trips• Compressor B supplies the total fuel to gas turbine A and B

Figure 13 shows the result of the run back test. This graphshows trend data of the G/T A and B demand flow, headerpressure, and compressor A and B flow. In this simulation test,comp A trip occurs 60 seconds after simulation test start. Aftercomp A trips, comp A flow goes down quickly. G/T A and Bload goes down to 30 percent load. Comp B flow goes down atonce, then goes up to recover the decreasing commonheader pressure.

Figure 13. Example of Simulation Test Result (Run Back Test).

From this result of the simulation test, the common headerpressure fluctuation is found to be within 3 bar (43.5 psi)(11.06/21.80 bar [115.37/226.11 psi] from SV), and thepressure became stable in two minutes. Table 4 shows a summaryof this simulation result.

Table 4. Simulation Test Result for Run Back Test.

ACTUAL CONTROL PERFORMANCEIn the following paragraph, the application of the FF+FB/FF

selector control system in a plant operating in the field isdescribed.

Figure 14 shows the load shedding test result. This graph showsthe trend data of the comp A flow, G/T A demand flow, and headerpressure. In this test, G/T A relates to comp A, and comp A is inFF+FB mode. G/T B and comp B are stopped. After G/T A loadshedding occurred, G/T Flow decreases. At the same time, comp Aflow is decreased by feedforward load function. Header pressure atfirst goes up then comes down.

Figure 14. Actual Plant Load Shedding Test Result.

From this result of the actual test, the common header pressurefluctuation is found to be within 2 bar (29 psi) (11.34/20.61 bar[119.44/28.85 psi] from SV), and the pressure became stable inthree minutes. Table 5 shows a summary of this test result.

Table 5. Load Shedding Test Result.

Furthermore the feedforward load function was verified in thefield. Figure 15 shows the verification result of the function. Thisgraph shows the original design function line, modified functionline, and actual operating points. The modified function line isbased on the actual operating points, using a linear equation.

Figure 15. Feedforward Load Function and Actual Operating Points.

The difference between the original and modified function wasthe molecular weight of the gas. The original function is set by thecompressor performance curve based on the design base molecularweight, 16.6. As for the actual gas, methane was less and ethanewas more than the design base gas composition, and the actual gasmolecular weight was 17.9 (about 108 percent from design).Because the actual gas base molecular weight became heavier thandesign base, the compressor performance has become better.Therefore the inclination of the function became small, andthe IGV control band became narrow for the feedforwardload function.

If the gas composition and molecular weight change, thefeedforward function will actually shift. But in this control system,the FB signal covers the deviation from the design. It is possible tocorrespond by this control system if the molecular weight changeis within 30 percent. This modified function is applied to the actualcompressor controller.


CONCLUSION

In this paper, a compressor load sharing control system with aFF+FB/FF selector has been proposed. This control system ismainly provided for a GTCC plant fuel gas compressor. The keyfactor in this control system is the controllability of the compres-sor. This control system uses FF+FB mode and FF mode for eachcompressor unit. The gas turbine fuel flow is input as the FF signaland the control valves are modulated by the FF signal. The FBsignal comes from the header pressure controller, and adjusts thecontrol valves.

The functions of the control system are verified using a dynamicsimulator. From the result of the simulation, it was confirmed thatthis control system has good controllability. Furthermore, thiscontrol system was verified using a simulation test, so it isconcluded that the above control system shows good controllabil-ity in a plant operating in the field.

NOMENCLATURE

ASV = Antisurge valveCAL = Calculation unitComp = CompressorFB = FeedbackFF = Feed forward

Fx = Function generatorG/T = Gas turbineGTCC = Gas turbine combined cycleIGV = Inlet guide vaneL/S = Load sheddingLSS = Low signal selectorMV = Manipulating valuePd = Discharge pressurePIC = Pressure indicating controllerPLC = Programmable logic controllerPs = Suction pressurePV = Process valueQd = Discharge flowSEL = SelectorSV = Set valueTd = Discharge temperatureTs = Suction temperatureUIC = Antisurge controller

BIBLIOGRAPHY

Shinskey, F. G., 1996, Process Control Systems—Application,Design, and Tuning, McGraw-Hill Publishing Company.


31

Robert L. (Bob) De Maria is PlantReliability Superintendent at the BeulahSynfuels Plant, in Beulah, North Dakota.He has responsibility for the reliability ofthe plant’s rotating and stationaryequipment and piping systems. He has 35years of experience with the specification,maintenance, and reliability of rotatingequipment in the petrochemical industry,which includes 20 years of experience atDakota Gasification Company.

Mr. De Maria received a B.S. degree (Mechanical Engineering,1970) from Stevens Institute of Technology. He has authoredseveral papers and magazine articles on rotating equipment.

M. Theodore (Ted) Gresh is VicePresident of Marketing and Engineering,Flexware, Inc., in Grapeville, Pennsylvania.He has been involved in the design ofhigh efficiency centrifugal compressorstaging, field-testing of compressors andsteam turbines, and troubleshooting fieldperformance problems for more than 30years. While most of this time was in theTechnical Services Department of ElliottCompany, he is presently with Flexware,

Inc., a company focused on consulting services, training seminars,and software for turbomachinery performance analysis.

Mr. Gresh received a B.S. degree (Aerospace Engineering, 1971)from the University of Pittsburgh. In addition to numerous papersand magazine articles, he has published a book on the subject ofcompressor performance, and has several patents related to turbo-machinery. He is a registered Professional Engineer in the State ofPennsylvania.

ABSTRACT

Continuous online aerodynamic performance monitoring ofturbomachinery is crucial to plant operation and maintenance and isa key part of a thorough machine reliability program. Integratingmachine real-time performance along with mechanical parameters(vibration data, thrust, bearing temperatures, and oil condition)makes for an all-encompassing condition-based maintenance

program resulting in better maintenance and reliability decisions.Case studies of machine problems and the need to extend the timebetween overhauls are discussed. The benefits that are realized fromimplementation of online performance monitoring are presented.

INTRODUCTION

Just like racing driver Jimmie Johnson strives to keep his car intop condition in order to keep out in front of the pack, every plantmanager must work hard to keep his plant in top operationalcondition. It is that little bit of added performance that is the cream,the profits that make the difference between surviving and thrivingin today’s competitive economic climate.

While a good, effective maintenance program is not free, and thecost of maintaining that program cuts into the bottom line, theadded revenue from high onstream factors more then offsets thiscost. The goal is to have a low-cost program with real benefits.Instant feedback on information with minimal intervention is key.Basing a maintenance program on equipment condition rather thanoperating time will go a long way toward saving money.

While there is always the risk of too much information, knowinghow the equipment is running is just as important as knowing plantprofits on a real-time basis. Online performance monitoring ofcritical turbomachinery equipment adds value by knowinginstantly when something goes wrong or is starting to go wrong,and the data provided helps scheduling maintenance andtroubleshooting for root cause. Getting to problems quickly cutslosses and adds to the bottom line. And, if the equipment is runningfine, then why spend the time and money to open it andinspect/replace all the wearing parts?

MAINTENANCE ANDOPERATIONAL IMPROVEMENTS

In addition to confirming you are getting what the equipmentmanufacturer promised you would get, online performancemonitoring will give you instant feedback for maintaining peakefficiency. While there are other indirect means to monitorcompressor efficiency, directly calculating efficiency is the bestmethod to assure the problem is in the compressor and notsomewhere else. Monitoring the performance can help withmaintenance scheduling such as cleaning (Figure 1) andscheduling a major turnaround. When it comes to considerations ofa rerate for increased capacity, the current and historical data willbe invaluable in making the decision on how to expand.

55

THE ROLE OF ONLINE AERODYNAMIC PERFORMANCE ANALYSIS

byRobert L. De Maria

Plant Reliability Superintendent

Dakota Gasification Company

Beulah, North Dakota

andM. Theodore Gresh

Vice President

Flexware, Inc.

Grapeville, Pennsylvania

Figure 1. Polymer Buildup in the Diffuser Passage of a Cracked GasCentrifugal Compressor. Online Monitoring Shows the Rate of Decayin Performance and Aids in Scheduling Maintenance Such as SprayWashing or Disassembly for Manual Cleaning. (Courtesy Gresh, 2001)

Plant operators are constantly making adjustments to improvethe plant processes. Knowing the effect on the compressor viainstant feedback from online performance monitoring can expeditetheir work immensely saving valuable man-hours.

Performance monitoring can help with mechanical as well asefficiency issues. Some very basic information gathered fromperformance monitoring like where the compressor is operating onthe performance curve can help prevent impeller failures or at leasehelp in troubleshooting efforts. Documenting surge events and chokeevents are vital in finding the root cause of a failure and provide thedata necessary to change procedures for preventing future failures.

Startups and other transients like trip situations can lead tosevere upsets that may not show up as a significant event on thevibration monitors but may push the compressor into deep chokeor surge, and unless online monitoring is implemented, will gounnoticed (Figure 2). Improper startup of a refrigerationcompressor can result in liquid ingestion and damage to thecompressor internals especially the impellers. Normal performancetesting will not show these phenomena since a performance test istypically conducted during stable, steady-state conditions.

Figure 2. Monitoring and Documenting Where the Compressor IsOperating on its Performance Curve Can Enhance MechanicalReliability. Damage to Equipment Is Known to Occur in the ChokeRegion as Well as in Surge. (Courtesy Gresh, 2001)

Online Performance Monitoring System

Online performance monitoring utilizes compressor operatingdata and compares the processed results to the manufacturer’sexpected performance to determine the compressor health.Pressures, temperatures, speed, and flow rates are processed todetermine operating work input, head, and efficiency (Tables 1 and2). These calculated results are then displayed on a chart (Figure 3)and compared to original equipment manufacturer (OEM)expected values. A historical data log (Figure 4) of raw data andcalculated results is maintained for use in determining maintenanceschedules and troubleshooting problems.

Table 1. Input Data for GB1701 Methanation Compressor Pulledout of the Plant Historian Software. Input Data Then MovesThrough the Spreadsheet below and Then to the CalculationEngine. Table 1 below Shows Live Compressor Information BeingReceived and (Table 2) the Calculation Results.

Table 2. Output Data for GB1701 Methanation Compressor.

Figure 3. Head, Efficiency, and Work for GB1701 MethanationCompressor. Data Shows Compressor to Be Operating as Expected.


Mechanical Reliability•Avoid Deep Choke (Stonewall)•Avoid Surge•Monitor Transients

Figure 4. Historical Data Chart for GB 1701 MethanationCompressor Showing Efficiency (Upper Curve) and DeltaEfficiency (Lower Curve).

In a typical online performance monitoring system, compressoroperating data are transferred from a plant data collection systemthat, for most new plants, includes all the necessary data requiredfor determining the compressor performance such as head, power,and efficiency. Some older plants do not collect all of this necessarydata and some pressures and temperatures will need to be added tothe existing data collection system in order to implement the onlinemonitoring system. These data are transferred via dynamic dataexchange (DDE) to the performance software calculation enginewhere the data are processed.

While some may consider it possible to monitor data bycomparing discharge pressure or pressure ratio over time, this isnot really a good procedure and can create more confusion thanhelp. A change in inlet temperature or pressure will affect thedischarge pressure and even the pressure ratio. So, it is wise tomonitor head and efficiency. When calculating head, the bestmethod is by using Benedict-Webb-Rubin (BWR) equations ofstate. This will provide the best means of comparing calculatedresults with the OEM performance curves for the compressor.

Along with head and efficiency, the work input is also calculated.By monitoring the work input, it is possible to monitor the qualityof the compressor data being collected. Work input is representativeof the energy transferred from the compressor impeller blades to thegas. Any degradation in the interstage seals, corrosion, or foulingwill show up as a loss in efficiency and head, but the work inputis fixed and remains the same regardless of these losses. Thus,monitoring work input is a good means to confirm that the inputdata are accurate.

Case History A

While online performance monitoring does not in itself enhancethe mechanical integrity of the equipment, it is another tool forobtaining additional information otherwise not available or difficultto obtain or decipher. Such information can offer additional guidanceto operators for avoiding conditions that could potentially causeequipment failure. Specifically, online performance monitoringprovides an easy method of monitoring where the compressor isoperating on its performance curve making it easy to assure thecompressor is operating within the OEM limits. If a failure doesoccur, the information available from online performancemonitoring will be valuable in the troubleshooting efforts.

In 1997, a newly installed propylene refrigeration compressor insouthern Louisiana tripped on vibration after just 20 hours ofoperation (Kushner, et al., 2000). The compressor was a two-section centrifugal with all new internals and driven by a constantspeed motor. Inspection of the compressor found that the cause ofthe high vibration was mechanical failure of the last (fifth) stageimpeller. A triangular section of the impeller back plate (Figure 5)was missing and several cracks were found near each vane tip.

Figure 5. Damage at Tip of Fifth Stage Impeller of PropyleneCompressor. (Courtesy Kushner, et al., 2000)

The second section of the compressor and specifically the laststage impeller were found to be operating well beyond the end ofthe curve, in deep choke conditions (Figure 6).

Figure 6. Head Ratio Versus Inlet Flow Ratio for Last (Fifth)Stage Impeller of Propylene Compressor. (Courtesy Kushner,et al., 2000)

While the compressor was an older two-section machine, theinstallation was new. The compressor had been rerated for the newsite with new diaphragms and new rotor. The motor was oversizedsignificantly for the application so that it could be a spare for anothercompressor. During startup, liquid propylene was sprayed into theinlet piping upstream of the knockout drums to provide gas tooperate the compressor. Unfortunately, the flowmeters, which werenew, were installed correctly, but the wrong factor was input into thecomputer resulting in a calculated flow rate that wassignificantly less than the actual flow rate. So, when the operatorsnoted the flow was looking low, they were concerned about surgingthe compressor, and increased the liquid injection to increase the gasto the compressor and bring it further away from surge. Unknowingto the operators, they were pushing the compressor deep into choke.Further adding to the problem was the knockout drum design. Thedrum was horizontal and did not have demister pads.

Since several cracks were found near each vane tip on the toe ofthe fillet welds, resonance of an impeller natural frequency was atthe top of the list of possible causes of the failure. The design ofthe impeller was a cutback design. The blade outer diameter wasless than the impeller hub and cover outer diameter. So, theresolution of the problem was to cut the hub and cover back to theblade diameter, stiffening the impeller and raising the impellernatural frequencies.

There were no metallurgical problems found with the failedimpeller. The metallurgical examination by the OEM showed thatcracking was not a result of preexisting (manufacturing) defects.Stress levels at the crack initiation site were calculated to be 15.4 ksi.

THE ROLE OF ONLINE AERODYNAMIC PERFORMANCE ANALYSIS 57

The only impeller resonance found was for plate modes at theimpeller outside diameter (OD). While these frequencies could beexcited by return vanes at the impeller inlet, they should dissipategreatly, according to the vendor, by the time flow reached theimpeller OD where the failure occurred. In the vendor’s opinion, itwould be difficult to have this high excitation unless there wereliquids involved. The diffuser did not have vanes that might excitethe impeller tip.

Flow through the compressor’s second section during this 20hour period was calculated to be 7950 icfm. This is 5 percentbeyond the end of the OEM performance curve and 15 percenthigher than the original rated flow.

Borer, et al. (1997), did research that clearly shows that flowwithin a volute and around its tongue or cutwater causes a nonuni-form static pressure field that will act as an excitation force at theimpeller OD. While this pressure gradient is present throughout thecompressor operating range, computational fluid dynamics (CFD)analysis shows that it becomes stronger in the choke region(Figures 7 and 8). Also, full load testing with straingaugesdemonstrated that overload dynamic stresses are significantlyhigher than dynamic stresses near surge.

Figure 7. CFD Profile of Pressure Gradient Around the Peripheryof a Discharge Volute for a Compressor Operating near Surge.Note How the Pressure Is Relatively Uniform.(Courtesy Borer, et al., 1997)

Figure 8. CFD Profile of Pressure Gradient Around the Peripheryof a Discharge Volute for a Compressor Operating in OverloadConditions. Note the Large Pressure Change at the Volute Cutoff.(Courtesy Borer, et al., 1997)

Based on this, prudence dictates operation within the boundariesof the OEM curve avoiding choke as well as surge.

The Coal Gasification Plant

The commercial-scale coal gasification plant in Beulah, NorthDakota, began operating in 1984 and today produces more than 54

billion standard cubic feet of natural gas annually. An abundantlignite resource underlying the rolling North Dakota plainssupplies the plant with the fuel source. The plant consumes morethan six million tons of coal each year (Figure 9).

Figure 9. The Coal Gasification Process. (Courtesy Stelter, 2001)

Coal gasification involves dismantling the molecular structure ofcoal and reassembling the resultant hydrogen and carbon asmethane. The heart of this plant is a building containing 14 gasifiers.Lignite is fed into the top of the gasifiers. Steam and oxygen are fedinto the bottom of the coal beds causing intense combustion(2200ºF). The resulting hot gases break down the molecular bondsof the coal and steam, releasing compounds of carbon, hydrogen,sulfur, nitrogen, and other substances to form a raw gas that exits thegasifiers. Ash discharges from the bottom of the gasifiers.

The raw gas is cooled and the tar, oils, phenols, ammonia, andwater are condensed from the gas stream. Other byproducts areused as boiler fuel for steam generation.

Methanation takes place by passing the cleaned gas over a nickelcatalyst causing the carbon dioxide to react with free hydrogen toform methane. Final cleanup removes traces of carbon monoxide.

Synthetic natural gas leaves the plant through a 2-foot-diameterpipeline, traveling 34 miles south. There it joins a 1249 mile inter-state natural gas pipeline system, which transports the gas topipeline companies. These companies supply thousands of homesand businesses in the United States.

In addition to natural gas, the coal gasification plant producesnumerous products from the coal gasification process that haveadded great diversity to the plant’s output. These products includeammonium sulfate and anhydrous ammonia, which are fertilizersthat supply valuable nitrogen and sulfur nutrients for agriculturalcrops. Other products include phenol for the production of resinsin the plywood industry, cresylic acid for the chemical industry,liquid nitrogen for refrigeration and oil field services, methanol forsolvents, naphtha for gasoline blend stocks, carbon dioxide forenhanced oil recovery, and krypton and xenon gases for thenation’s lighting industry.

In 2004, this plant began implementation of a continuous onlinemonitoring system for critical machinery because of the followingevents causing the company +$10,000,000 in lost profits:

• 30,000 hp condensing mechanical drive steam turbine wasopened for inspection after 15 years of operation and was found tohave severe erosion in the nozzle block area resulting in extendingthe scheduled outage

• 30,000 hp axial air compressor blade failure resulting in a 50percent reduction in production (Case History B )

• 14,000 hp centrifugal gas compressor was partially fouledcausing a 10 percent reduction in plant rates (Case History C)

Setting Up and Monitoring Compressors

While the Beulah coal gasification plant has realized for sometime the value of adding online performance monitoring to their


condition-based maintenance program, limited instrumentation hashampered progress. There are 22 critical unspared compressor trainsthat directly affect the plant revenue stream plus other sparedcompressor and expander trains. All of these machines are not nowmonitored continuously because of instrumentation deficiencies.Most of these machine trains have been in service for more then 20years. As the machine control panels and instrumentation areupdated, all of the compressor trains will eventually have continuousperformance monitoring. In this interim period, manual data aretaken from local instrumentation to calculate machine performance.

Data required for online monitoring includes: inlet anddischarge pressure and temperature at each compressor sectionnozzle, flow rate, and speed. Additionally, it is necessary to have anaccurate gas analysis and have a copy of the compressormanufacturer’s predicted performance curves. Gas data are enteredinto the computer along with the curve data for reference.

Accurate data are essential to obtaining good results with anycondition-monitoring program. Calibrating all instruments involvedwill assure accurate data and correct results. One way to confirm thedata are accurate is to monitor the compressor work input. If thework input falls on the curve, then the data are most likely accurate.If the measured work input does not fall on the work input curve,then the first thing that needs to be done is to check the accuracy ofall instruments. This includes the flowmeters, pressure andtemperature readings, and the gas analysis.

At the Beulah plant, data are pulled from the plant historiansystem to a Microsoft® Excel spreadsheet. From there the onlinemonitoring system takes the data via DDE connection and processesthe data to obtain the appropriate compressor parameters such ashead and efficiency. A single point is displayed on the performancecurve (Figure 3) and the curves are compensated for speed. Ahistorical trend tracks various data (Figure 4) versus time to showdegradation of the compressor efficiency and thus aid maintenancescheduling. Tracking the delta efficiency (the difference between thepredicted and the actual efficiency) and delta head is a very goodindication of compressor health. Tracking the delta work input (thisvalue should be zero or near zero) is a good way of tracking theoverall quality of the online performance monitoring system.

Case History B

In June 2002, an axial air compressor blade failedcatastrophically in the first row of the low pressure section (Figure10). The compressor is driven by a 30,000 hp fixed speedsynchronous motor through a parallel shaft gear increaser at anoperating speed of 4530 rpm. This compressor is one of twomachines operated in the air separation facility for oxygenproduction used in the plant gasifiers and thus crucial to plantoperation. The other machine is direct driven by a steam turbine.Both machines were placed in continuous operation in 1984. Thesemachines operate at about 140,000 scfm and a discharge pressureof 90 psia. A similar failure had also occurred in the motor driventrain in1995. Subsequent to this failure a protective coating wasinstalled on the low pressure blading to protect it from erosion andcorrosion. The impact to plant revenue of this second failureprompted an extensive investigation into the root cause.

Figure 10. Axial Air Compressor Blade Failure. The Second in theCompressor’s 22 Year Operating History.

A thrust bearing failure on the steam turbine train in 1993resulted in an upgrade to a directed lube tilt pad design in bothcompressors. In 1994 the speed was increased from 4425 rpm to4530 rpm, which increased capacity by 12 percent. There were noincidents of high vibration, or high thrust bearing or journalbearing temperatures prior to the blade failure. Analysis of thefailed blade showed it to be of the correct material and mechanicalproperties and was not the result of foreign object damage,corrosion, or erosion. Since only one minute averaged data wereavailable from the plant historian, the operating point on a dynamicbasis was not available making it difficult to determine that themachine had been operating in a choke condition. A finite elementanalysis (FEA, Figure 11) was performed to determine thestructural response for the defined load conditions. Metallurgicalexamination suggested that the failure was due to fatigueassociated with a relatively high mean stress. In addition, bladevibrations and plastic strain above the tolerance limit most likelyassisted in the fatigue failure. It was the contention of theinvestigator that the crack leading to the blade failure was initiatedfrom the concave (high pressure) side of the thin blade edge andprogressed in the transverse direction (opposite direction toairflow) along about 60 percent of the fracture surface until theblade succumbed to ductile overload (Figure 12).

Figure 11. Finite Element Analysis of Failed Axial Blade. Note theHigh Stress on the Blade Trailing Edge.

Figure 12. Failed Axial Blade. Crack Initiated on High Pressure,Trailing Edge of Blade.

After returning the machine back to service with the spare rotor,high vibration was measured on the inlet air duct with anaccelerometer at the blade pass frequency of the rotor’s first row ofblades. Vibration in g’s ranged from 1 g to more than 75 g’s.Dynamic loading data were recorded from pressure sensorsmounted in the inlet ducting and in the compressor balance line.High energy levels were detected at various times suggesting that


the blade design was marginal (Figure 13). This prompted adetailed look to see where the machine was operating on its curve.Observation of fluctuating downstream and upset conditionswere absorbed by the motor driven train while the steam driventrain was base loaded. This was causing the motor train tointermittently operate in a choke condition (Figure 14). Correctiveactions included optimization of blade geometry, higher strengthmaterial and tighter blade manufacturing tolerances, and revisedoperating procedures. The authors believe that a continuous onlineperformance monitoring system could have mitigated this situationor at least assisted in the analysis for root cause.

Figure 13. Goodman Diagram for Failed Axial Blade.

Figure 14. Axial Compressor Performance Map Showing NormalOperation. System Demands Occasionally Drop DischargePressure Resulting in Very Brief Periods of Choke Condition andEventual Blade Failure.

Case History C

Synthetic natural gas is compressed by two parallel case steamdriven compression trains after methanation for delivery into anatural gas pipeline. The trains operate at a maximum speed of13,700 rpm. Suction conditions are 250 psig at a suction tempera-ture of 110ºF. Discharge conditions are 1440 psig max and 110ºFafter final cooling before entering the pipeline. Each train has thecapability of delivering 85 mmscfd of gas. During periods of highpipeline pressures, the capacity was noticed to be less thenexpected causing a reduction in plant rates. Two problems wereidentified. The first was the suction temperature had increased to150ºF. This problem was traced back to a fouled final heatexchanger in the methanation area. The second problem showed adecrease in efficiency of the first compressor case having fivestages. An increase in 13 vibration of the first compressor case wasalso noted.

A scheduled overhaul of the compressor revealed fouling.Random flaking off of the fouling explained the higher vibration.Analysis of the deposits showed it to be predominantly iron oxide.An inspection of the carbon steel suction piping revealedsignificant corrosion on the bottom of the pipe. The first attempt tomitigate the fouling problem was to install a nonstick coating onthe rotors. This did not solve the problem because the coating waseventually abraded and the fouling reoccurred.

An analysis of the process gas showed it to be saturated.Furthermore, the piping from methanation to the suction of thecompressor is outside and not traced or insulated. Ambientconditions can range from –40ºF to 110ºF. It was realized thatparticularly in the winter, the cold piping was causing condensateto form in the bottom of the piping, which reacted with the gas toform carbonic acid. The acid formed then corroded the steel pipe.The corrosion products were transferred to the compressor by thehigh gas velocities. A suction knockout pot on the line to thecompressor has not been effective in preventing the carryover fromreaching the compressor. A glycol water removal unit is locatedbetween the first and second compressor cases to knockout thewater before exporting the gas.

Injection of chemicals to neutralize the acid formation on thecompressor suction piping was investigated and found to be veryexpensive. The current implementation plan is to steam trace andinsulate the suction piping to keep the condensate from formingand to improve the efficiency of the suction knockout pot. Figure15 shows a drop in efficiency prior to cleaning. Similarly, thecompressor efficiency returned to normal after cleaning. Figure 16shows buildup on the rotor. Continuous performance monitoring isassisting with tracking this situation.

Figure 15. Synthetic Natural Gas Compressor Showing Drop inEfficiency.

Figure 16. Synthetic Natural Gas Compressor Rotor. Note Foulingin Impellers. Monitoring Compressor Performance Helps Trackthis Condition and Aids Scheduling of Maintenance.


CONCLUSION

In this day and age of high energy prices, revenue losses causedby machine deficiencies, and downtime at the Beulah gasificationplant can exceed $600,000 per day. Over the years we have realizedthe importance of continuous machine vibration, thrust, bearingtemperature, and oil condition monitoring to operating reliable andsafe turbomachinery. Unfortunately, this does not tell us everythingthat we need to know to maximize revenue and minimize mainte-nance expenses.

By integrating machine real-time performance monitoring alongwith mechanical parameters (vibration data, bearing temperatures, andoil condition), we have realized that better-informed decisions can bemade. Online performance monitoring is now a key part of the plant’smachine reliability program. Key benefits that have been seen include:

• Knowing the machine performance immediately significantly aidsthe process of troubleshooting a machine problem and minimizingdowntime/loss production.

• Trending of machine performance allows review of operatingpoints that may have subsequently affected machine condition.

• The effects of process changes can be evaluated immediately.

• Knowing machine performance along with vibration, thrust,bearing temperature, and oil condition, scheduled maintenance andoverhauls can be extended for well designed, maintained, andoperated equipment.

• Performance monitoring provides valuable information whenjustifying an extended time between overhauls to an insurancecarrier as well as minimizing insurance premiums.

REFERENCES

Borer, C., Sorokes, J. M., McMahon, T., and Abraham, E. A.,1997, “An Assessment of the Forces Acting Upon

a Centrifugal Impeller Using Full Load, Full PressureHydrocarbon Testing,” Proceedings of the Twenty-SixthTurbomachinery Symposium, Turbomachinery Laboratory,Texas A&M University, College Station, Texas, pp. 111-121.

Gresh, M. T., 2001, Compressor Performance, Aerodynamicsfor the User, Woburn, Massachusetts: Butterworth-Heinemann.

Kushner, F., Richard, S. J., and Strickland, R. A., 2000, “CriticalReview of Compressor Impeller Vibration Parameters forFailure Prevention,” Proceedings of the Twenty-NinthTurbomachinery Symposium, Turbomachinery Laboratory,Texas A&M University, College Station, Texas, pp. 103-112.

Stelter, S., 2001, The New Synfuels Energy Pioneers, Bismarck,North Dakota: Dakota Gasification Company.

BIBLIOGRAPHY

Kushner, F., 2004, “Rotating Component Modal Analysisand Resonance Avoidance Recommendations,”Proceedings of the Thirty-Third TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas, pp. 143-162.

Kushner, F., Walker, D., and Hohlweg, W. C., 2002,“Compressor Discharge Pipe Failure Investigation with aReview of Surge, Rotating Stall, and Piping Resonance,”Proceedings of the Thirty-First Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University,College Station, Texas, pp. 49-60.


James M. (Jim) Sorokes is Manager ofDevelopment Engineering, with Dresser-Rand Company, in Olean, New York. He isinvolved in all aspects of productdevelopment for Dresser-Rand worldwide.He previously spent 28 years in theAerodynamics Group, becoming theSupervisor of Aerodynamics in 1984 andbeing promoted to Manager ofAero/Thermo Design Engineering in 2001.During Mr. Sorokes’ time in the

Aerodynamics Group, his primary responsibilities included thedevelopment, design, and analysis of all aerodynamic componentsof centrifugal compressors. His professional interests include:aerodynamic design, aeromechanical phenomenon (i.e., rotatingstall), and other aspects of large centrifugal compressors.

Mr. Sorokes graduated from St. Bonaventure University (1976).He is a member of AIAA, ASME, and the ASME TurbomachineryCommittee. He has authored or coauthored more than 30 technicalpapers and has instructed seminars and tutorials at Texas A&Mand Dresser-Rand. He currently holds two U.S. Patents and hastwo other patents pending.

Harry F. Miller is the Product Manager -Marketing of Turbo Products at Dresser-Rand, in Olean, New York. His career inturbomachinery began 31 years ago withDresser Clark, and Mr. Miller has helda variety of design engineering andmarketing positions, most recently, beingManager of Development Engineering andLeader of the DATUM Development Team.His prior work experience consists of four

years as a mechanical construction engineer for the PennsylvaniaPower & Light Company.

Mr. Miller received a BSME degree from Northeastern University,and an MBA degree from Lehigh University. His areas of expertiseinclude turbocompressor and gas turbine design and application.

Jay M. Koch is Manager of Aero/ThermoDesign Engineering, at Dresser-Rand, inOlean, New York. He has been employedthere since 1991. Prior to joining Dresser-Rand, he was employed by Allied SignalAerospace. He spent 14 years in theAerodynamics Group, before beingpromoted to his current position in 2005.During Mr. Koch’s time in theAerodynamics Group, his responsibilities

included the development, design, and analysis of all aerodynamiccomponents of centrifugal compressors. Additionally he wasresponsible for the development of software used to select andpredict centrifugal compressor performance.

Mr. Koch holds a B.S. degree (Aerospace Engineering) from IowaState University. He has authored or coauthored 10 technical papers.

ABSTRACT

This work describes the potential consequences associated withoperating a centrifugal compressor in overload. Nomenclature isoffered to explain what is meant by overload operation, andmethods that are used by original equipment manufacturers(OEMs) and end users to define overload limits are presented. Thepaper also describes the conditions that can lead to overloadoperation. Computational fluid dynamics (CFD) results are used toillustrate the forces acting on an impeller when it operates at veryhigh flow. Finally, this paper suggests considerations that should beaddressed when designing (or selecting) an impeller that could besubjected to extended overload operation.

INTRODUCTION

It is common knowledge that operating centrifugal compressors insurge can have detrimental effects on the mechanical integrity ofparts. Prolonged excursions into surge have caused damage toimpellers, bearings, seals, and other rotating components. The violentnature of surge events makes it necessary to install sophisticatedcontrol systems to prevent compressors from operating in surge.

63

THE CONSEQUENCES OF COMPRESSOR OPERATION IN OVERLOAD

byJames M. Sorokes

Manager of Development Engineering

Harry F. MillerProduct Manager - Marketing of Turbo Products

andJay M. Koch

Manager of Aero/Thermo Design Engineering


Olean, New York

While there are numerous publications in the open literatureaddressing the consequences of operating in surge, there are few, ifany, articles on the potential dangers associated with operating acompressor in overload—the high flow region of a performancemap. In the vast majority of applications, overload operation doesnot represent any cause for concern. In fact, many compressorsspend the majority of their life operating in the high flow end oftheir performance envelope. Yet, in some applications, overloadoperation can be just as damaging as operation in surge.

This paper will discuss the potential consequences of operationin overload. The primary focus will be the potential damage thatcould occur in impellers, as they are typically the most expensiveand difficult component to replace in a centrifugal compressor. Thediscussion will begin with definitions of the terms “overload” or“deep overload.” These terms mean different things to differentpeople. Therefore, it is important to understand the meaning ofthese terms in the context of this paper. The focus will next move tothe types of situations that can lead to operation in overload. This isfollowed by an overview of some “real-world” examples of damagecaused by overload operation. Computational fluid dynamics(CFD) and finite element analysis (FEA) results will then be usedto provide a clearer picture of the forces imposed on impellers whenthey operate well above their intended design flow rates.Suggestions are offered on factors that should be considered whendesigning impellers that may be subjected to prolonged overloadoperation. Application of the general guidelines outlined in thisdiscussion will in no way ensure the long-term mechanical integrityof the impellers. Given the unique factors associated with eachsituation, it is imperative that you seek the guidance and directionfrom engineering. However, proper application of these guidelinesmay increase the impellers’ resistance to the forces imposed byoverload operation that could lead to unwanted consequences.Finally, conclusions will be offered on the need to be cognizant ofthe consequences of overload operation when establishing theoperating envelope for a new compressor application.

DEFINITIONS

Unlike surge, there is no commonly accepted definitionfor overload or “operation in overload.” The compressoraeroperformance maps given in Figures 1 and 2 will be used toillustrate the various definitions. The maps show the headcoefficient and polytropic efficiency plotted against the normalizedinlet flow coefficient.

Figure 1. Definition of Overload.

Figure 2. Setting Overload Limit.

In the broadest sense, the term “overload” is used to referenceany condition under which the compressor’s inlet flow exceeds itsdesign flow rate (Figure 2). However, most compressors regularlyoperate at flow rates higher than design and most users do notconsider this operating in overload. To them, this is normaloperation. Instead, this group deems overload to be operating atany condition that exceeds the safe or recommended flow limit forthe machine. This hazardous region to the extreme right on theperformance map is often called “deep overload.”

End users typically defer to the original equipment manufacturer(OEM) to establish an overload operating limit if the OEMbelieves such a limit is warranted. This is where the difficultiesreally begin because, unlike surge, compressor OEMs, in general,do not specify an overload limit. Because compressor performancedrops off rapidly in the overload region, it is frequently assumedthat an end user will not want to operate the compressor in thatregime or will lack the driver capacity to do so. As will be seen, thiscan be a risky assumption to make.

The majority of OEMs address the issue of overload operation byputting a clause in the operator’s manual stating that off-the-mapoperation must be avoided. In the strictest interpretation, this clausecovers both the low-flow and high-flow ends of a performance map,i.e., do not operate at flow rates lower or higher than shown on themap. Still, surge control systems are installed to ensure no (orlimited) operation in surge. It is somewhat uncommon to findcontrol systems that limit overload operation. Again, this is quiteoften the case because the compressor driver cannot providesufficient power to maintain prolonged operation in overload.

On occasion, OEMs will provide an overload limit on the right-hand side of the performance map (refer to Figure 2). This limit isset based on a variety of methods that are described in thefollowing paragraphs. In fact, in rare instances, anti-overloadcontrol systems have been provided that operate based onalgorithms very similar to an anti-surge control system, except thata discharge throttle valve is utilized instead of a recycle valve.

Percent of Design Flow

One of the simplest methods for establishing the overload limit isto specify the maximum amount of flow as a percentage of the designflow. Such limits are typically set based on experience and are afunction of the machine Mach number (U2/A0), gas handled, numberof stages, etc. It is well known that a compressor that runs at highmachine Mach number or is handling high mole weight gases willhave less overall flow range than a compressor that runs at low U2/A0or handles low mole weight gases (Figure 3). Therefore, the allowablepercentage increase from design flow must vary. For example, onemight impose an overload limit of 120 percent of the compressordesign flow rate for a high mole weight machine and allow as muchas 140 percent of design flow for a low mole weight application.


Figure 3. Variation in Flow Range with Mole Weight.

Impeller Mach Number

Some OEMs limit the overload capacity based on calculatedvalues for impeller inlet relative Mach number. Calculations areperformed with increasingly larger flow rates and the performancemap is truncated (or “cut off”) when the Mach number in any stagewithin the compressor exceeds the specified level. Typically, thelimit is a calculated Mach number of 1.0, the flow rate at which theimpeller throat is choked. However, to be conservative, someOEMs set the overload limit based on an inlet relative Machnumber of 0.96 or lower.

It is noteworthy that all OEMs do not use the same methods tocalculate inlet relative Mach numbers. Further, inlet relative Machnumber can be calculated at various locations on the impeller leadingedge, i.e., at the shroud, at the mean. Therefore, agreement should bereached between the end user, contractor, and OEM regarding theexact definition of the Mach number being considered.

Another approach often employed by end users is to specify amaximum allowable inlet relative Mach number at the designcondition for any impeller in a compressor. This method assumesthat limiting the inlet relative Mach number at design will ensure asufficient amount of overload capacity. However, this method doesnot ensure that the compressor is not operated in a risky portion ofthe performance map.

Drop in Head Coefficient/Efficiency

Another popular method for establishing overload limit is basedon the rate of change of head coefficient or efficiency as a functionof flow coefficient. This method takes advantage of the rapid dropin the compressor’s performance in the overload region as seen inFigures 1, 2, and 3. Often referred to as “stonewall” (because of itsvertical nature and the fact that it limits compressor flow like astone wall), the high flow end of the map has a highly negativeslope. Some OEMs will limit the overload capacity to the flowrate corresponding to an “X” percent drop in head coefficient orefficiency for a 1 percent increase in flow. Common values of “X”fall between 10 percent and 20 percent.

An alternate form of this approach limits the overload capacityto the flow rate at which the efficiency or head coefficient isa given percentage of the design flow level, i.e., 15 percent, 20percent, or 30 percent of the design efficiency or head coefficient.Beyond this point, the performance would be so low as to beunusable in most processes.

Section Summary

Clearly, it is important that the OEM and end user have acommon understanding of how the overload limit was establishedand the potential implications of violating this limit. The latter willbe addressed in subsequent sections herein.

CONDITIONS LEADING TO OVERLOAD OPERATION

Common circumstances that lead to prolonged operation inoverload are:

• Loss of parallel compressor or compression train,

• Performance degradation (i.e., fouling) within a compressor,

• An undersized compressor (either due to changes in flowrequirements for an existing compressor or misapplication of a newmachine),

• Alternate operating conditions (i.e., summer and winter conditionsin a pipeline application),

• Unanticipated changes in gas characteristics (mole weight, etc.),

• Process upsets.

Parallel Compressors

Many facilities have duplicate compressors or trains ofcompressors and the process flow is divided equally (or nearlyequally) between these compressors or trains. Each machine in thetrain is sized such that the compressors operate near the middle oftheir performance map under normal conditions. However, if forsome reason (process upset, regular maintenance, etc.) one or moretrains are taken offline, plant operators may attempt to make up forthe missing train(s) by forcing more flow through the remainingoperational trains. This causes those remaining trains to operate atsignificantly higher flow rates, pushing them into the overloadregion of their performance maps. Assuming the drivers havesufficient power (and many do not), it would be possible tooperate in the overload region for extended periods of time. Theperformance would suffer (i.e., low efficiency) but productionrequirements would be maintained.

Excess Driver Power

It is considered normal practice by many users to rate gas turbinedrivers for the highest expected ambient air temperature expected,or at least an “average” of the historical high temperatures for thatlocation. This is done so that the compressor and hence the “processplant” may be operated at close to design output on a hot day, orthroughout the hot season. With typical aeroderivative gas turbinesrated for a 120ºF inlet, the power output will increase from 25percent to 75 percent when the ambient drops to 30ºF. It is also notuncommon for steam turbines to be rated for 110 percent of thecompressor rated power with “minimum” steam conditions, only tobe capable of providing 50 percent or more power when operated at“maximum” steam conditions.

Performance Degradation

In a multistage compressor, it is possible to drive the latter stagesinto overload if there is a reduction in the performance of theearlier stages or if there is a decrease in the mole weight of the gasbeing compressed. With regard to the former, the impellers orstages in a compressor are aerodynamically matched such thatwhen all are functioning correctly, each operates near the middle ofits respective performance map. However, if the performance ofthe early impellers or stages in the compressor begins to degrade,those impellers or stages will not provide the expected volumereduction. Therefore, any subsequent stages will be forced to“swallow” more flow than expected. If the performancedegradation is substantial, the latter stages will operate in overload.Again, assuming there is sufficient driver capability, it would bepossible for the stages to operate in overload for prolonged periods.Again, the horsepower consumption would increasesignificantly but production could be maintained.

The same scenario occurs if there is a change in the gascharacteristics such that the volume reduction or overall flow rangeof the individual stages within the compressor drops. This will

THE CONSEQUENCES OF COMPRESSOR OPERATION IN OVERLOAD 65

cause the impellers or stages to operate at successively highercapacity relative to design until the latter stages are operating at ornear choke.

Undersized Compressor/AlternateOperating Conditions/Unexpected Changes

At times, compressors are purposely or inadvertently undersizedin the selection process. For example, an end user may anticipate asizeable reduction in flow rate during the life of the compressor.This user may choose, in the beginning, to operate the compressorat the high-flow end of its map, knowing that in coming years theflow rate will be reduced and the compressor will operate nearerthe center of its map.

Occasionally, end users purchase equipment before finalizingtheir operating requirements. When the compressors are put intooperation, they discover that the inlet conditions and/or gasmixtures are not as expected or that their production requirementsexceed those originally projected. These changes can cause thecompressor to operate at flow rates that are much higher thananticipated. In short, the compressor operates in the overloadregion of its performance map.

Finally, process conditions may also change during the life of acompressor, causing the flow rate through the machine to increase.The end user may lack sufficient funding to revamp or upgrade thecompressor, choosing instead to operate existing equipment inoverload despite the low performance.

POTENTIAL CONSEQUENCESOF OVERLOAD OPERATION

Two examples are presented to illustrate the potentialconsequences of overload operation. In both cases, the compressorswere known to have operated for extended periods in the overloadregion. Note: Before proceeding further, it must be noted that endusers and OEMs are often very reluctant to provide details ondifficulties resulting from off-design operation. End users do notwant details of their operating practices made public and OEMstypically do not want any suggestion that their products areanything but perfect. For that reason, limited details will beprovided on the two sample cases.

The first example is taken from a high-pressure gas reinjectioncompressor. The compressor experienced repeated impellerfractures. The welded impeller was a low flow coefficient design,having a very short blade height relative to the impeller diameter(low b/r). Essentially, the impeller disk was fracturing at or aroundthe leading edge region of the impeller. In the worst case, the innerportion of the impeller disk separated from the outer portion.

Following an investigation by the OEM and end user, it wasreported that the impeller had been run extensively in the overloadregion of its performance map when a parallel train was takenoffline. Dynamic forces caused by the combination of highimpeller leading edge incidence, and a nonuniform pressuredistribution caused by the downstream discharge volute, weresufficient to initiate cracks in the impeller, leading to the fractures.The compressor had to be taken out of service for several days toallow installation of the spare rotor.

It should be noted that the forces that led to the fractures weresignificantly lower at the design flow condition. That is, at designflow, the dynamic forces due to incidence and the volute pressure fieldwere not sufficient to initiate the cracking. Further details on this casecan be found in Borer, et al. (1997), and Sorokes, et al. (1998).

The second example is from a compressor processing heavyhydrocarbons. The subject impeller was a high-flow coefficientdesign, implying that the leading edge was quite tall (high b/r).Blade fractures occurred near the impeller’s leading edge. In a fewof the blades, a portion of the blade broke away as indicated by thecrosshatched area in Figure 4. Again, based on an analysisconducted by the end user and the OEM, it was found that the

compressor had operated in overload for long periods. In this case,the dynamic forces because of incidence caused a portion of theimpeller blade’s leading edge to fracture and separate from theimpeller. The loss of material produced an unbalance on the rotorand the compressor had to be taken offline for repairs.

Figure 4. Schematic of Blade Leading Edge Fracture

While the mechanisms that ultimately led to the impellerfractures were different (more on this in the discussion to follow),generally speaking, the root cause for the fractures was the same—operation in overload.

DESCRIPTION OF FORCES

Although the forces resulting from operation in overload have asimilar root cause to those found in surge or stall, the true nature ofthose forces is radically different. The violent forces associatedwith surge are typically very low frequency (6 Hz or less) andresult from the flow reversal through the compressor when theimpeller or impellers can no longer overcome the downstreamstatic pressure. When in surge, the inability of the compressor toovercome such pressure is directly related to the increase inincidence or other losses in the compressor components, i.e.,impellers, diffusers, return channels, etc. That is, as the flow rate ina compressor is reduced from design toward surge, the angle atwhich the flow impinges on the bladed or vaned componentsincreases, thus increasing the incidence (or delta angle between theflow angle and the blade/vane angle) (Figure 5). At some point, theincidence angles lead to flow separation or other anomalies thatcause very high losses within all of the compressor components,making it impossible for the compressor to overcome thedownstream pressure. Because flow moves from the region ofhigher pressure to a region of lower pressure, the flow reversesdirection and goes backward through the compressor. The resultingforces on the compressor internals can be destructive.

Figure 5. Impeller Incidence.


The basic driver of the forces associated with overloadoperation is also incidence. However, the frequency of the forcemechanism tends to be high frequency. Whereas surgephenomenon has a frequency of 6 Hz or less, the forcesassociated with overload tend to be more on the order of bladepassing frequency (1 to 3 kHz), i.e., the number of impellerblades times compressor running speed. This distinction isimportant as one considers the nature of the potentially damagingforces.

As with surge, when the compressor is operated at flow ratesabove design, there is an increase in blade incidence on theimpeller. Eventually, the incidence becomes high enough that, asin surge, the flow separates from the impeller blade. In or nearsurge, this separation occurs on the suction surface; while inoverload, it occurs on the pressure surface of the blade. The flowseparation causes a reduction in the effective throat (orminimum) area in the impeller, leading to a significant increasein the flow velocity. If the velocity gets high enough, shockwaves will form and choke will occur. The pressure fieldsassociated with shock waves are highly dynamic and causeexcessive forces on the impeller blades and walls. These forcesalone could be sufficient to damage an impeller. However, whenthese forces are further exacerbated by other nonuniformitieswithin the compressor flow path, one can rapidly reachconditions that can quickly lead to impeller fractures. Two suchsituations are described in the following discussion.

Blade/Vane Interactions

It is well known that interactions occur between impellers andstationary vanes upstream (inlet guidevanes or IGVs) and/ordownstream (diffuser vanes). Numerous technical papers havebeen written on this subject including Kushner (1980), Fisherand Inoue (1981), and Eckert (1999). Summarizing, the upstreamor downstream vanes are surrounded by fields of varyingpressure through which the impellers rotate causing pressurefluctuations and, therefore, varying forces on the impeller blades.The frequency of the pressure disturbance is determined bymultiplying the rotational speed of the impeller by the numberof stationary vanes. Of course, if the upstream componentcomprises multiple vane rows, the vane wakes from all of theupstream vanes must be considered because it may be possiblefor the wakes from the first row to propagate through the secondrow and reach the impeller.

Problems arise when the frequency of the pulsations align witha natural frequency in the impeller, thereby exciting the impeller,resulting in a resonance condition. When the forces are ofsufficient magnitude, it would be possible to initiate a fracture inthe impeller.

Blade/vane interactions during overload operation are ofparticular concern because of the high level of energy in the gasstream. In overload, the higher-than-nominal flow rate causes thevelocities within the aerodynamic components to be higher thanat design. Therefore, the static pressure variation at the exit of theupstream inlet guide vanes will be less uniform than it is atdesign flow. That is, the high core flow velocity in the IGVpassages will cause regions of low static pressure, while thestatic pressure in the stagnated (or wake) region immediatelydownstream will be very high (Figure 6). If the inlet guide vanesare sufficiently close to the impeller blade leading edges (as isoften the case with full inducer-style impellers), each blade willbe subjected to this highly nonuniform static pressure field.Because the magnitude of the pressure variation is higher inoverload than at design, the forces during overload operationmay be sufficient to create problems, even though the forces atdesign are not.

Figure 6. CFD Results Showing IGV Wakes.

The situation with downstream diffuser vanes is somewhatdifferent. Rather than interacting with wakes, the excitation forcesassociated with vaned diffusers are a consequence of the pressurefields forming around the diffuser vanes because of potential floweffects (Figure 7). Again, as the impeller blades pass the diffuservanes they pass through the “lobes” in the pressure distribution.The resulting variation in pressure imposes a dynamic force on theimpeller that could excite an impeller’s natural frequency. As withthe IGV wakes, the magnitude of the pressure variation surround-ing the diffuser vanes will be greater during overload operationthan at design. This is due in large part to the increased incidenceon the diffuser vanes and probable separation of the flow from thevane pressure surface. The increased velocities resulting from thehigher flow rates also increase the kinetic energy or velocitypressure (1/2ρV2). This causes even higher loads across the impellerblades. As a result, the dynamic forces in overload will be greaterand might be sufficient to initiate a fracture whereas those at designare not.

Figure 7. CFD Results Showing Pressure Field Around LSD Vanes.


Interaction with Other Pressure Nonuniformities

The forces acting on the impeller during overload operationcan be further exacerbated by the presence of other pressurenonuniformities in the compressor flow path. A prime example ofthis is the nonuniformity caused by a discharge volute or collector.Again, several technical papers have been published that documentthis nonuniformity, including Hagelstein, et al. (1997), andSorokes and Koch (2000). Essentially, an impeller upstream of adischarge volute or collector will be operating in a nonuniformcircumferential pressure field as seen in Figure 8. The impeller canbe somewhat shielded from the effect of the volute though the useof vaned diffusers. However, the nonuniformity persists nonetheless.

Figure 8. Nonuniform Pressure Field Due to Volute.

As reported in Sorokes and Koch (2000), such nonuniformities,whether caused by a volute, compressor inlet, or other drivingmechanism, further increase the level of dynamic forces acting onthe impeller because the circumferential pressure distributiontends to become even more skewed when operating in overload.The dynamic strains shown in Figure 9 are for an impellerupstream of a volute operating at various flow rates. As can beseen, the strains are highest in the overload region of theperformance map. It should be noted that these strains wereobtained during an extensive test program to identify the rootcause of the impeller fracture.

Figure 9. Variation in Impeller Dynamic Strain with Flow.

Wake Effects

There is much conjecture as to whether it is possible for animpeller to excite itself into resonance. Although there is nodefinitive proof that this can occur, it has been theorized that theblade wakes shed by an impeller could provide the mechanism forself-excitation. Under normal operation, the wake regions aresufficiently small and of reasonably limited pressure magnitude (i.e.,the difference between the static pressure in the wake region and thestatic pressure in the core flow is small). Therefore, there is notenough force or “delta pressure” to cause any difficulties. However,as in the inlet guide vane, the difference between the static pressurein the core flow and static pressure in the secondary zone and bladewakes is higher. It is possible that an impeller blade could interactwith the wake shed by the preceding blade (Figure10), resulting inan excitation at blade passing or other frequency (i.e., number ofimpeller blades times the rotational speed).

Figure 10. Impeller Blade Wakes.

ANALYTICAL METHODS

Computational fluid dynamics can provide valuable insight into theflow physics and the detrimental forces associated with operationin overload. For example, Borer, et al. (1997), used CFD to investigatepossible excitation mechanisms in a problematic stage in ahigh-pressure reinjection compressor. Their work showed that a highstatic pressure load was occurring near the leading edge of the impellerwhile operating in overload. The static pressure distribution for theimpeller is shown in Figure 11. The root cause of the pressure load washigh negative incidence on the impeller blade caused by operating at ahigh flow rate (i.e., overload). When the impeller was operating nearerdesign flow, the pressure load was significantly smaller. In short, theanalytical results were helpful in identifying the adverse conditionswithin the impeller that contributed to the fractures.

Figure 11. Impeller Static Pressure Distribution During OverloadOperation.


CFD can also be used to develop pressure loads or net impeller bladeforces that can be applied when conducting finite element analyses toassess the structural integrity of designs. CFD can also be used todetermine how said forces vary with flow rate. Historically, CFD waslimited to providing steady-state (or time averaged) pressuredistributions that could be applied as pressure loads in FEA studies.However, in the past decade, advances in CFD have made it possible togenerate unsteady or transient pressure distributions. These unsteadypressures can be translated into a force at a frequency and imposed as aboundary condition in an impeller natural frequency analysis. Thisoffers designers the opportunity to assess possible excitation anddetermine if changes are needed to avoid a resonance problem.

High-flow coefficient impellers that may be subjected to overloadoperation are of particular concern. For clarity, high-flow coefficient isdefined herein as any impeller having a flow coefficient, φ, of 0.12 orgreater, where φ or phi is an internationally accepted, nondimensionalflow coefficient relating flow, speed, and diameter. By their nature,such impellers have very tall leading edges, that is, the length of theblade from hub to shroud at the leading edge is large. The excessivelength makes it possible for the blade to flex or flutter if it is subjectedto a sufficient dynamic force. The impeller blade leading edge is mostsusceptible to flexing because it is the tallest and often the thinnestportion of the blade. If the magnitude of the vibrations becomessufficiently large, a fracture could occur. Therefore, it is imperative thatdesigners understand the magnitude and the frequency of the forcesacting on such impellers.

The above issue is of even greater concern on unshrouded or openimpellers. Without a cover or shroud to help hold the blades in position,the blades have significantly more freedom to flex or flutter. As a result,operation near stonewall on open impellers is strongly discouraged.

As noted, operation in overload causes high levels of negativeincidence on the impeller blade leading edges. This negative incidencecauses unbalanced pressure forces between the two sides of the blades.If these forces become dynamic or unsteady because of the presence ofupstream IGVs or some other circumferential asymmetry in the flowfield, the fluctuating pressure forces could be more than sufficient toinduce alternating stresses in the impeller. Assuming the excitationforce does not change in frequency, and assuming the frequency alignswith an impeller natural frequency, resonance could occur andultimately result in the loss of mechanical integrity.

As an illustration, the pressures acting on the leading edge of ahigh-flow coefficient impeller are shown in Figures 12 and 13. Theanalytical results given in Figure 12 are for the impeller operating at itsdesign flow rate while the results in Figure 13 are for overloadoperation. In both cases, the plot on the right is for the pressure surfaceof the blade, while the plot on the left is for the suction surface. Also,the pressure scales are consistent among all of the figures.

Figure 12. Static Pressure Load on High Flow Coefficient ImpellerOperating at Design Flow.

Figure 13. Static Pressure Load on High Flow CoefficientImpeller Operating at Overload.

As can be seen, the variation in pressure loads is much higherwhen the impeller is operating in overload. One can note the widevariation in pressure on the pressure side leading edge at overloadversus the considerably lower variation at design. Since thepressure on the suction side of the leading edge is fairly consistentbetween design and overload, it is also clear that the delta pressuresuction to pressure surface is higher in overload than at design.Note further that the pressure distribution in Figure 13 bears aresemblance to the schematic of the blade leading edge fracture inFigure 4. This alone is not definitive evidence that a fracture wouldoccur. However, were the pressure forces on the impeller unsteady(which they certainly would be), and were the unsteadiness to be ata frequency that aligns with the natural frequency of the bladeleading edge (i.e., the blade leading edge mode), a blade fracturelike the one shown in Figure 4 could occur.

DESIGN OR OPERATIONAL CONSIDERATIONS

It is critically important that end users and OEMs discuss the fullrange of conditions that a compressor will face in operation. Forexample, if the process engineers or end user know that there is achance for extended operation in overload, the OEM may want toundertake more detailed analyses of the proposed equipment.These analyses may include detailed CFD and FEA analyses asdescribed above to understand the potential ramifications ofoverload operation. These analyses can be used to:

• Establish the maximum safe operating limit,

• Help the end user understand the risks of operating in that flowregime, or

• Provide guidance to the OEM regarding how to modifycomponent designs to minimize the risks by increasing therobustness of the design.

Regarding the latter, it may be possible to change the number ofvanes in adjacent stationary hardware to avoid natural frequencyinterference issues. Alternatively, the design or manufacturingtechnique used for an impeller could be changed to make it lesssusceptible to interference issues. For example, the OEM maychoose to apply more stringent criteria for welds (i.e., weld shaperequirements, more rigorous inspection techniques) or possiblyresort to single-piece impeller fabrications. Of course, a naturalfrequency in the speed range is acceptable as long as its responsedoes not exceed allowable limits imposed by the end user or OEM.In short, by having full knowledge of the potential modes ofoperation, the OEM can undertake efforts to minimize the risks.

One additional example illustrates how this happened on a recentcompression project. The client, a major oil and gas producer,needed more compression on an existing offshore platform to boost


gas going to shore. Because of the declining pressure nature of thefield, the compressors were designed with both the present andseveral future conditions in mind to allow for change-out of internalparts over the life of the field to accommodate the changingoperating conditions. Because the application required very highhorsepower drivers (>40 MW), and because the early years ofoperation had the highest inlet pressures, only a few stages(impellers) were required to meet the specified discharge pressure.This resulted in a fairly high horsepower per stage relative to theend user’s and OEM’s experience for this particular size unit. Incontrast, operation in future years at much lower inlet pressureswould require more stages (impellers) to meet the same desireddischarge pressure. Therefore, in the later years of operation, thehorsepower per stage would drop into more a comfortable range.

The end user also desired that if one unit had to be taken out ofservice, the remaining units would need to accept more inlet flow;that is, operated in overload, a flow rate 30 percent to 50 percenthigher than the design flow rate. Because the stage power was highat the normal design point, the OEM was concerned about theprospect of even higher powers being absorbed in an off-designmode. When the end user consulted the OEM about the potentialoperation at higher flows, the OEM responded with an engineeringstudy to determine the magnitude of the dynamic stress. The studyincluded an innovative combination of transient CFD and FEA.The resulting analyses indicated that the original impeller geometrywas acceptable but with relatively low safety margins. The OEMthen determined that the calculated stresses could be significantlyreduced with very minor changes to the geometry of the impeller,notably in the blade to shroud fillet weld. This substantiallyincreased the factor of safety, and providing confidence that theimpeller integrity would not be at risk due to overload operation.

If the end user had not consulted the OEM, it is uncertainwhether or not a problem would have occurred. However, byhaving an open dialog, and by being proactive in performing ananalysis, the risk of a problem was much more remote.

Of course, it may not be possible to eliminate all of the risksassociated with overload operation. Even the most advancedanalyses are but approximations of the real world. Therefore, onecannot be assured that such analyses will capture all of thepotentially damaging phenomena within the compressor flow path.Further, even the most robust designs will fail if subjected to theright excitation mechanism, i.e., one that aligns with the naturalfrequency of the impellers. Even a solid ring would fail if subjectedto the right excitation. At some point, common sense must prevail.The end user and OEM must face the reality that the safestapproach is to avoid operating in the overload region of theperformance map. If the compressor might break if you run there,do not run thereóor be prepared to undergo regular overhauls tocheck for internal damage.

End users can employ overload control systems to ensure that acompressor does not operate beyond some agreed upon maximumcapacity. These are implemented by incorporating algorithms inthe control system that limit driver operation (speed, load, etc.) orrestrict the movement of control valves to keep the compressor ina safe region on its performance map. Some argue that suchoverload controls limit production and decrease profitability.However, when this reduction is weighed against the costs and lostproduction associated with equipment failure, limiting overloadoperation does not seem to be a bad choice.

CONCLUSIONS

The most important conclusion to be derived from this work isthat, contrary to many commonly held beliefs, operation inoverload can subject a centrifugal compressor to adverse forces. Infact, in some circumstances, overload operation can be just asdetrimental to component structural integrity as surge.

Of course, it is critical that the end user and OEM come to anunderstanding on the meaning of overload operation. The term

“overload” has many connotations, so it is important that all partiesadopt a common definition. Because one person’s “overload”might well be another’s normal operation, a more rigorousdefinition must be applied when specifying compressor flow rangerequirements or discussing how the compressor is being operatedin production. It might also be possible to adopt an industrywidestandard that defines “overload” as operating a compressor atflow rates that exceed the maximum flow rates shown on theperformance map provided by the OEM. This would put the onuson both the OEM to provide an accurate prediction of overloadcapability and the end user to properly assess their need to operateat such high flow rates.

As seen herein, in some reported cases, prolonged operation inoverload can lead to impeller fractures. Overload operation canalso exacerbate structural natural frequency interference issues thatmay exist within a compressor flow path. Forces and/or pressurenonuniformities tend to be greater when operating at flow ratesmuch higher than design. Such increases are caused in large part bythe increased incidence levels on impeller blades or adjacentstationary vanes (i.e., vaned diffusers).

Advanced analytical tools such as computational fluid dynamicsor finite element analysis help quantify the magnitude of the forcesassociated with overload operation. Such analyses can also be usedto mitigate risk. Designs can be modified to reduce the potential forharmful interferences, or operating limits can be derived so as toavoid risky portions of the performance map. However, the mostsophisticated analyses and most advanced manufacturing methodscannot eliminate the risk of component failures due to overloadoperation. Common sense dictates that the most effective way toeliminate such risk is to avoid high risk operating conditions.Simply put, if there is increased risk of mechanical failure byrunning at a portion of the performance envelope, the risks must beweighed against the potential gains and the cost of maintenance orreplacement of the equipment.

In conclusion, though not receiving as much attention as surge,prolonged operation in overload can have very detrimental effects ona centrifugal compressor. End users and OEMs alike need to becognizant of the potential risks associated with operating in thisportion of the performance map. End users accept the risks associatedwith surge and, despite the extra horsepower consumed, often runtheir compressors on recycle so as to avoid surging the units. Asnoted, the forces associated with surge and overload are similar, yetthe industry has not taken steps to protect equipment from overloadoperation. Failure to recognize the risks associated with overloadoperation can have a devastating impact on compression equipment,production, profitability, and engineering careers.

NOMENCLATURE

ACFM = Actual cubic feet per minuteA0 = Sonic velocity of gas at impeller inletb = Impeller blade heightCm = Impeller meridional velocityD2 = Impeller exit diameterIGV = Inlet guide vaneLSD = Low solidity vaned diffuserN = Rotational speed in rpmQ = Inlet volumetric flow in ACFMr = Impeller radiusU1 = Impeller inlet peripheral velocityU2 = Impeller exit peripheral velocityU2/A0 = Machine Mach numberV = Gas velocityW1 = Impeller inlet relative velocity

φ =

ρ = Gas density


700Q.

Ν D23

REFERENCES

Borer, C., Sorokes, J. M., McMahon, T., and Abraham, E. A., 1997,“An Assessment of the Forces Acting Upon a CentrifugalImpeller Using Full Load, Full Pressure Hydrocarbon Testing,”Proceedings of the Twenty-Sixth Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 111-122.

Eckert, L., 1999, “High Cycle Fatigue Cracks at Radial FanImpellers Caused by Aeroelastic Self-Excited ImpellerVibrations, Part 1: Case History, Root Cause Analysis,Vibration Measurements,” Proceedings of DETC99, ASMEDesign Engineering Technical Conference.

Fisher, E. H. and Inoue, M., 1981, “A Study of Diffuser/RotorInteraction in a Centrifugal Compressor,” IMechE JournalMechanical Engineering Science, 23, (3), pp. 149-156,London, England.

Hagelstein, D., Van den Braembussche, R., Keiper, R., andRautenberg, M., 1997, “Experimental Investigation of theCircumferential Pressure Distortion in Centrifugal CompressorStages,” ASME Paper No. 97-GT-50.

Kushner, F., 1980, “Disc Vibration—Rotating Blade and StationaryVane Excitation,” ASME Journal of Mechanical Design, 102,pp. 579-584.

Sorokes, J. M. and Koch, J. M., 2000, “The Influence of LowSolidity Vaned Diffusers on the Static Pressure NonuniformityCaused by a Centrifugal Compressor Discharge Volute,”ASME Paper No. 00-GT-454.

Sorokes, J. M., Borer, C. J., and Koch, J. M., 1998, “Investigationof the Circumferential Static Pressure Nonuniformity Causedby a Centrifugal Compressor Discharge Volute,” ASME PaperNo. 98-GT-326.

ACKNOWLEDGEMENTS

The authors acknowledge Mr. Donald Schiffer, Mr. Paul Geise,and Mr. Edward Abraham for their support in the preparation ofthis paper. The authors also recognize the efforts of Mr. JasonKopko, Mr. Robert Kunselman, and Mr. Edward Thierman in thepreparation of several of the figures included herein. Finally, theauthors thank Dresser-Rand Company for allowing them theopportunity to publish this paper.


Frank Davis is an Engineering Consultant with ExxonMobil,assigned to Ras Laffan LNG Company Ltd., in Doha, Qatar. He hasmore than 30 years of experience in the field of rotating machinery.Mr. Davis specializes in project engineering for machineryapplications, bid reviews, and machinery acceptance testing. After20 years machinery engineering experience with Exxon and MobilCorporations, he was a member of the Project Task Force managingthe engineering and construction of four LNG plants in the State ofQatar over the past 10 years. He has developed specifications,completed bid reviews, and followed manufacturing, testing, andparticipated in the startup of more than 40 machinery trains including65 MW gas turbine driven refrigeration compressors and gasexpander/compressor units.

Mr. Davis received a B.S. degree (Mechanical Engineering, 1966)from New Jersey Institute of Technology and is a registeredProfessional Engineer in the State of New Jersey.

Reza Agahi is a Consultant, in Irvine,California. He recently retired after 33 yearswith GE Rotoflow where he last served asDirector of Marketing, Sales, andCommercial Operations. Dr. Agahi hastaught in universities in Southern Californiaand has authored more than 40 articles andpapers in system engineering and turboma-chinery applications. He is the inventor andcoinventor of several GE Rotoflow Patents.

Dr. Agahi received B.S. and M.S. degrees (MechanicalEngineering, 1968) from Tehran University, and an M.S. degree(1974) and Ph. D. degree (Operations Research and SystemsEngineering, 1977) from the University of Southern California.

Randy Chih-Chien Wu is a Senior Engineer with GE Oil & GasOperations LLC-North America, in Rancho Dominguez, California.He began his career with Rotoflow Corporation (now part of GE Oil& Gas) as Thermodynamic Design and Test Engineer for expanderand compressor products. In his 19 years of working experience, hehas been involved in various disciplines in designing and developingof expander and compressor products. His current responsibilitiesare focusing on risk review and NPI activities as well as test andcommission operation of expander and compressor product.

Mr. Wu received a B.S. degree (Physics and MechanicalEngineering, 1976) from Chung Yuan University and an M.S. degree(Mechanical Engineering, 1983) from the University of Nebraska atLincoln. He is a member of ASME.

INTRODUCTIONThe application of the turboexpander in natural gas processing

and the petrochemical industry had its beginning at a small gasplant in Southwest Texas where Dr. Judson S. Swearingen installedthe first natural gas turboexpander (Swearingen, 1999).

Turboexpander technology has developed considerably in thelast 40 years. For example:

• Advances in fluid dynamics theory and computational fluiddynamics have made it possible to design a turboexpander withhigh isentropic efficiency and performance predictability;

• Progress in rotordynamics evaluation and modern finite elementanalysis capabilities have resulted in more reliable turbomachinery.

• Increase in demand and economies of scale have resulted innatural gas processing and petrochemical plants becoming largerand larger (Figure 1) (Agahi, 2003).

81

FULL LOAD, FULL SPEED TEST OFTURBOEXPANDER-COMPRESSOR WITH ACTIVE MAGNETIC BEARINGS

byFrank Davis

Engineering Consultant

Doha, Qatar

Reza AgahiConsultant

Irvine, California

andRandy Wu

Senior Engineer

GE Oil & Gas

Rancho Dominguez, California

Figure 1. Turboexpander Flow Development Since 1960.

Three large size turboexpander compressor (TEC) units with activemagnetic bearings (AMB) were installed in a gas plant in the early1990s (Agahi, et al., 1994). Because of some difficulties with theoperation of the inlet guide vanes (IGV) at the commissioning stagecaused by hydrate formation, oil and gas companies and engineering-procurement companies (EPCs) were somewhat skeptical about AMBtechnology and its reliability. Furthermore, with more offshoreplatforms using turboexpander compressor units, ease of commissioningand reliability were among the major concerns of end users.

In order to address these concerns, the end users of turboexpandercompressor units began to demand testing of the mechanical andcontrol designs of custom designed equipment, and full load, fullspeed testing (FLFST) of turboexpander compressor units. The firsttests were carried out in the late 1990s (Bergmann, et al., 1996).These FLFSTs used a hydrocarbon gas mixture that was intended tosimulate the actual process gas as closely as possible.

Full load, full speed testing of a turboexpander compressorwith hydrocarbon gas proved to be somewhat difficult and tooexpensive. Furthermore, obtaining permits to conduct such tests is achallenge for most test facilities. To circumvent some of these issues,a mixture of nitrogen and helium was used in some turboexpandercompressor FLFSTs in early 2000. In this approach, a mixture ofnitrogen and helium formulated to simulate the molecular weight ofthe actual process gas is used in a closed test loop.

BACKGROUND

A major Middle East gas production company was contemplating thedevelopment of six large liquefied natural gas (LNG) plants. Thedesigns were to be similar, all based on the use of turboexpandercompressor units for maximum production of LNG. The turboexpandercompressor power was estimated to be in the range of 9.0 to 10.0 MW,assuming that some of the plants would have units in parallel operationto limit the size of the expanders to modules already demonstrated.

The company had experience with similar plants and had identifieda large turboexpander compressor as a major contributor to plantoutages or production limitations. A thorough review of the problemsexperienced in the existing plants was conducted with the responsibleoperating and rotating equipment engineers. The objective of thereview was to identify specific design features of the turboexpandercompressor that required upgrading from previous designs.

Several specific features of the existing turboexpander compressorwere identified as responsible for the great majority of equipmentoutages. In order of importance, these were:

• Dilution of lubricating oil by process gas and subsequent loss ofoil viscosity resulting in excessive vibration and bearing wear.

• Inadequate design of the thrust balancing provisions and pooroperation of the thrust balance mechanism resulting in thrustbearing overload.

• Clamping of inlet guide vanes due to lack of controls resultingin erratic unit performance.

The review group concluded that a new, large turboexpander

compressor application should incorporate features to addressthose deficiencies and that the turboexpander compressor shouldbe tested as thoroughly as possible prior to shipment.

Turboexpander compressors now employ magnetic radial and thrustbearings as a solution to the major problem of oil contamination.Thrust calculations and thrust balancing designs have improved andthe load can be measured with the active magnetic bearing system.Inlet guide vane controls were updated by the addition of an electricactuator but it was not certain that the control would be as accurateas desired. The active magnetic bearing system, however, lacksredundancy and presents the new concern of a component failureresulting in a loss of levitation under loaded conditions. In order todemonstrate the ruggedness of the design, an FLFST was planned inseries with the residue gas compressor for the same plant.

The turboexpander compressor package for this project used240 mm (9.45 inch) active magnetic bearings. Table 1 shows theturboexpander compressor wheel characteristics. This package wasto be installed in a closed loop with other LNG train equipment suchas a gas turbine (GT) driven residue gas compressor. The project GTprovides the power required to drive the residue gas compressor,which in turn delivers the required flow and pressure to theturboexpander compressor to bring the latter up to the full speed,full load condition. Table 2 shows the design guaranteed conditionsand Table 3 depicts the simulated conditions for the FLFST.

Table 1. Turboexpander Compressor Wheel Characteristics.

Table 2. Turboexpander Compressor Design/GuaranteedOperating Conditions.


Table 3. Turboexpander Compressor FLFS Operating Conditions.

FULL LOAD FULL SPEED TEST OBJECTIVES

The main objectives of the FLFS test that were agreed to by thecustomer and expander manufacturer are as follows:

• Verification of the mechanical integrity of the turboexpandercompressor with active magnetic bearings

• Verification of the control functions related to the inlet guidevanes of the turboexpander compressor unit, automatic thrustbalancing, and the antisurge valve while operating at FLFS

• Identification and correction of any faults or defects of theturboexpander compressor system and repeat of the FLFS test toverify that the issues were indeed rectified

• The FLFST sequence is shown in Figure 2

Figure 2. Full Load/Full Speed Test Sequence.

The following conditions were to be monitored and recorded:

• Startup

• Near trip speed operation

Full load operation, turboexpander compressor operates atmaximum continuous speed (MCS)

Partial load operation

• Design/normal speed operation

• Normal shutdown

• Emergency shutdown (ESD) due to process upset

• Two coastdown auxiliary bearing landing tests

• Verification of critical functions such as ESD and processshutdown system trips

THE CLOSED LOOP TEST SETUP

Figure 3 shows a schematic of the test setup and Figure 4 showsthe FLFST piping around the turboexpander compressor. Theturboexpander compressor unit was integrated into the residue gascompressor (RGC) test loop. The project residue gas compressorwas driven by its dedicated GT and delivered test fluid to theexpander at 64 barg (928.2 psig) and 50ºC (122ºF). The expanderextracted energy from the gas stream by expanding to 20 barg(290.1 psig) and cooling down to 221ºC (25.8ºF). The cold gasfrom the expander discharge was then fed to the recompressor,which used the expander power and boosted the test fluid to adischarge pressure of 31 barg (449.6 psig) before returning it to theRGC to repeat the cycle.

Figure 3. Closed Loop Process and Instruments Diagram for FullLoad/Full Speed Test.

Figure 4. Test Loop Around Turboexpander Compressor.


83

•

EQUIPMENT

The major components and supporting equipment in this testloop are listed below:

• Residue gas compressor, centrifugal compressor driven by a GT

• Residue gas compressor aftercooler

• Residue gas compressor auxiliary support system

• Test loop piping

• Flow measuring devices

• Pressure measuring devices

• Temperature measuring devices

• Pressure transducers, cabling, data acquisition/analysis equipment

• Startup seal gas supply, instrumentation air, etc.

• Gas analyzer and reporting system

• Vibration recording and analyzing equipment compatible withactive magnetic bearing system

• Residue gas compressor inline inlet strainer (60 mesh)

• Noise meter

The auxiliary equipment and components were as follows:

• Turboexpander compressor package with control system,including active magnetic bearing signal interlock with test facilityand startup seal gas supply

• Expander bypass, Joule Thompson (JT) valve

• Expander inlet quick shutoff valve

• Compressor surge control system and recycle valve

• Check valve downstream of the recompressor

• Expander inline inlet strainer (60 mesh)

• Compressor inline inlet strainer (20 mesh)

The majority of the operating parameters such as flow rates,pressures, temperatures, vibration, etc., were monitored and loggedby the automatic data acquisition system.

Considering the practical aspects of the FLFST, some designparameters could not be simulated. Figure 5 shows processparameters that were different during the FLFST compared to thenormal site conditions.

Figure 5. Comparison Between the Design and Test Parameters.

FLFST Operation

Before startup, the test loop was pressurized to 24 barg (348.1psig). The test header pressure reached 52 barg (754.2 psig) after theresidue gas compressor developed a stable pressure. Flow wasadmitted to the turboexpander compressor by opening the inlet guidevanes and closing the JT valve simultaneously. The recompressorantisurge valve was at the full open position at the startup. Theantisurge valve began closing to load up the recompressor when theturboexpander compressor speed reached approximately 5000 rpm.

The TEC speed was ramped up at 10 percent increments until thespeed closely approached the trip speed of 12,400 rpm and remainedat that speed for 15 minutes. Then the turboexpander compressorspeed was reduced to a maximum continuous speed (MCS) of 11,813rpm for two hours. The test loop equilibrium state was achieved byslowly adjusting the recycle valves of the recompressor and residuegas compressor. The turboexpander compressor speed was furtherreduced to the normal speed of 11,250 rpm and operated at this speedfor another two hours. At the end of this test run, the turboexpandercompressor speed was gradually increased to the shutdown speedand the turboexpander compressor tripped on high speed. Theturboexpander compressor rotor coasted down under normalconditions, i.e., the active magnetic bearing system was in levitatingmode during coastdown. Table 4 shows a sample of the FLFSTparameters that were monitored.

Table 4. A Sample of Test Parameters Monitored/Recorded.

The turboexpander compressor was restarted and the speed wasincreased to 10,550 rpm for about 15 minutes in order for thesystem to reach thermal equilibrium. A special method was appliedto bypass the active magnetic bearing controller and completelydisable the AMB amplifiers in order to activate delevitation.

INLET GUIDE VANE SENSITIVITYAND FLOW CONTROLLABILITY

The turboexpander compressor flow is linearly proportional tothe opening of its inlet guide vanes except in small opening andfull open positions. By controlling the sensitivity of the inlet guidevanes to within 1 percent, i.e., deviation between process signal toinlet guide vanes and feedback signal to actuator system, it couldbe demonstrated that the expander flow controllability is within 1percent of the total flow. As the trended data in Figure 6 show, thedifferences between the inlet guide vane input signals and thecorresponding feedback signals were mostly less than1.0 percentfrom ramp up to the FLFST condition. It is interesting to note thatthe injection of additional fluid to increase the test loop pressuredid not influence the inlet guide vane sensitivity or flowcontrollability. The inlet guide vane sensitivity remained within1.0 percent even when the expander inlet pressure was increasedto 64 barg (928.2 psig).

Figure 6. IGV Sensitivity Analysis.


AMB AUXILIARY BEARING LANDING TEST

Auxiliary bearings support the turboexpander compressor rotorwhen it is not levitated. Another function of the auxiliary bearingsis to catch the rotor upon loss of the magnetic field resulting indelevitation as the machine coasts down from full load and fullspeed (FLFS) to a full stop. To demonstrate the functionality of theauxiliary bearings and show their ability to support the rotor upondelevitation at FLFS, landing tests were performed during tur-boexpander compressor shop tests. To implement this test, theturboexpander compressor speed was increased to 10,550 rpm, andthen the expander was delevitated by intervening with the activemagnetic bearing control system. The delevitation signal shut offthe expander inlet quick shutoff valve and opened the JT valve. Theactive magnetic bearing controllers were bypassed by introducingjumpers in the control cabinet. As a result, all radial and axialamplifiers were disabled, and the rotor landed on the auxiliarybearings and coasted down to a complete stop. For both landings,it took approximately 2.4 seconds from the rotor landing until thequick shutoff valve shut off; it took 4.6 seconds for the turboex-pander compressor to coast down to a complete stop; the rotorlanded in the auxiliary bearings for a total of 7.0 seconds; and rotorwhirling stopped within 4.0 to 4.2 seconds. Figures 7 and 8 providedetailed records of these tests.

Figure 7. First Landing Test.

Figure 8. Second Landing Test.

Before and after the landing test, both the radial and axial airgaps between the rotor and auxiliary bearings were measured,compared, and contrasted (Table 5). The data showed that therewere no changes in gap dimensions after two landing tests.

Table 5. Comparison of Radial and Axial Air Gap Before and AfterLanding Tests.

The tear down and inspection showed that there were lighttouches on the compressor impeller blade tips. The rest of therotor and its corresponding stator parts such as the expanderwheel, shaft seals, sensor rings, thrust disk, and magneticbearings were found to have no touch marks and were inexcellent condition.

There were light marks on the ball bearing inner rings andlanding sleeves in both radial and axial surfaces but the ballbearings could roll freely.

ACTIVE MAGNETIC BEARING ROTOR VIBRATION

Before spinning the residue gas compressor for the FLFST,fine tuning of the active magnetic bearings, clearance, andtuning checks were carried out to ensure that the air gaps wereconsistent with the design values, transfer functions were up todate, and all the required securities were set correctly. Thebearing system was equipped with antivibration rejection andautomatic balancing system logic. The antivibration rejectionactivation deactivation limits were set at 3400 and 4600 rpm,respectively. The rotor first critical speed was estimated to beapproximately 37 Hz. At higher speeds the automatic balancingsystem took over the control function. These two systemsensured that the active magnetic bearing rotor always rotatesaround its inertia center. Figure 9 shows the turboexpandercompressor rotor vibration throughout the course of the FLFSTincluding both landing tests. The higher rotor vibration wasobserved during ramp up, at a speed range between 6,000 to9,000 rpm. The highest vibration reached 50 mm and wasmainly in the subsynchronous spectrum, from approximately37 percent to 44 percent of the synchronous frequency, andoccurred only during the startup period. The vibration at thislevel was considered normal compared to the alarm setting of90 mm. The tangential velocity component, i.e., exit fromthe inlet guide vanes, could cause swirling around theturboexpander wheel and resonate at subsynchronousfrequencies. At FLFST conditions, subsynchronous displacementswere almost nonexistent and overall vibration levels were about15 µm. The axial vibration was about 13 mm.

Figure 9. Rotor Total Displacement/Vibration.


85

The active magnetic bearing current chart, Figure 10, shows thecurrent for each bearing during the FLFST. The relatively flat curvesindicate that the rotor was quite stable at the FLFST conditions.

Figure 10. AMB Current.

The bearing temperature chart, Figure 11, shows that the bearingcoil temperatures were normal. The highest temperature of 70ºC(158ºF) was observed when the turboexpander compressor wasoperated near the trip speed. The rotor length expansion was measuredat 2150 mm. This measurement shows that the relative distance of therotor and housing was increasing during the test (Figure 11).

Figure 11. Shaft Extension.

ROTOR VIBRATION DURINGTURBOEXPANDER TRIP

One of the FLFST criteria was to observe the rotor behaviorduring shutdown and coastdown to stop. Upon a request for aturboexpander compressor trip, the normal sequence of the eventsis to close the quick shutoff valve and open the antisurge valve.Following this procedure, during turboexpander shutdown andcoastdown while the rotor was levitated, there were nosignificant changes in vibration levels, unbalance, temperatures, orbearing currents.

ACTIVE MAGNETIC THRUST BEARING ANDAUTOMATIC THRUST BALANCING SYSTEM

The turboexpander automatic thrust balancing system operatedflawlessly in conjunction with the active magnetic bearing thrustbearing control system. There were no indications of axial thrustbiases throughout the FLFST. The interlock logic between theactive magnetic bearing and turboexpander automatic thrustbalancing system was set such that when the axial current reached14 A, the turboexpander system would take action to open or closethe automatic thrust balancing valve to relieve the thrust load. Theaction automatically stopped when the bearing thrust current wasreduced to 12 A. During the FLFST, the valve opening varied from50 percent to 60 percent. Figure 12 shows the axial bearing currentduring the FLFST.

Figure 12. Axial Bearing Current.

EXPANDER INLET QUICKSHUTOFF VALVE TIMING

The project quick shutoff valve was used during the FLFST.Three trips were initiated. One trip was triggered by an overspeedtrip at the end of the four-hour mechanical running test, and twotrips were performed during landing tests. It took about 2.4seconds to complete the valve closing process. This durationincluded the signal transmission time, the valve actuator responsetime, and the stem traveling time. Based on the factory bench testreport, the valve closing time was 0.602 seconds. Therefore, thesignal transmission time in the test loop control system tookapproximately 1.8 seconds.

BEARING HOUSING, SEAL GAS, ANDVENT GAS TEMPERATURE MONITORING

The turboexpander compressor seal gas control system was not inoperation during the FLFST. The seal gas pressure was controlledmanually with a bypass valve. The bearing housing vent gastemperature alarm was initially set at 55ºC (131ºF). This setting wastoo low for the FLFST conditions and had to be revised. The activemagnetic bearing high temperature alarm was set at 110ºC (230ºF)and shut down was set at 130ºC (266ºF). Therefore, the vent gastemperature setting was revised to 95ºC (203ºF).

TEAR DOWN INSPECTION

After completion of the four-hour mechanical running and twolanding tests, the turboexpander compressor was disassembled forinspection. All parts were in good condition. The auxiliary ballbearings were replaced with a new set despite being in goodcondition. The turboexpander variable inlet guide vane assemblywas also removed and inspected. There were scratch marks on theinterfacing surfaces between the guide vanes and nozzle clampingrings as well as on the nozzle cover. These parts were sent to themetallurgical laboratory for determination of the root cause. Theconclusion was that debris carried by the gas stream of the test loopwas trapped in the nozzle grooves and was dragged into theinterfacing surfaces causing the scratch marks when the vanesmoved. All scratches were removed and the surfaces were restoredbefore the inlet guide vanes were reassembled.

FLFST RESULTS

The turboexpander compressor operation during the FLFST waswithout any major problems or unexpected events. UninterruptedFLFST continued for four hours at 12.5 MW and 11,250 rpm. Therotordynamics performance was stable and all dynamic parameterswere within the design limits and as predicted. The normal trips atFLFS and ESD trips with landing at FLFS into auxiliary bearingswere carried out successfully. Rotor coastdown and vibrationlevels for all trips were smooth and at safe levels. The inlet guidevane test for ease of operation under full pressure, FLFS, ramping


up, and closing down conditions did not show any faults, orindications of blow by or clamping problems. The response of theinlet guide vanes to process signals and controllability weredemonstrated to be consistent with other controls in the plant andhence could control plant flow with the desired precision. Theautomatic thrust balancing systems of the bearing andturboexpander maintained the axial position of the rotor in thedesired center position during the various tests and operatingconditions. Gas dynamics performance test results were asexpected and consistent with the predicted values. Expanderisentropic and compressor polytropic efficiencies were better thanthe guaranteed values.

LESSONS LEARNEDThe FLFST requires detailed planning and coordination with the

various facilities that are involved. The expander vendor had anearlier FLFST at a facility outside their group and this one waswithin their group. It turned out that more planning and attentionto details were necessary in the latter FLFST because each teamtended to leave something out anticipating that the other teamwould pick it up.

Turboexpander compressor units are normally skid-mountedpackages ready to be installed on a foundation. This packageshould be complete with all auxiliary and support systems before itis installed for the FLFST. There were delays, confusion andmanual (in lieu of automatic) operation because some componentswere not shipped to the test site.

The turboexpander compressor shutdown loops should bededicated and have absolute minimum response time to guaranteethe safety and security of the turboexpander and its processes.

The noise level for this turboexpander compressor wasestimated at 85 dBA with a noise jacket installed on the casings.The turboexpander, bearing housing, and compressor did not havea noise jacket during the FLFST and no background noisecorrection was applied. Noise levels were measured at 115 dBA,some 20 percent higher than calculated/expected. This test helpedto highlight the need to review and revise formulas and algorithmsused for noise level estimation.

The integrity and ruggedness of the auxiliary bearings weretested and demonstrated by multiple landing tests. Inspection ofrotor parts after landing tests showed that there was no damage.This could be considered as justification to delete landing tests thatnormally are requested during open loop air tests.

Gas dynamics performance tests of the expander and compressor

produced the same performance that was predicted at the designstage and that would have resulted from extrapolation of the openloop air test. Therefore one may conclude that the FLFST does notprovide any additional information through the gas dynamicsperformance tests.

PRESENT CONDITIONThe turboexpander compressor package was installed,

commissioned, and put into normal operation in early 2006. The unithas been in normal operation since then.

CONCLUSIONSThe end user and EPC for a large LNG facility in the Middle

East had specific requirements for the design, manufacturing, andFLFST of a turboexpander compressor package with an activemagnetic bearing system. The turboexpander vendor workedclosely and diligently with them and incorporated all the specialrequirements that were consistent with the turboexpander com-pressor design. The FLFST was carried out in a loop where theresidue gas compressor for the same project supplied the requiredboost for this package. The FLFST was conducted successfully andwith relatively few problems. All the specified FLFST criteria werefulfilled satisfactorily. The turboexpander compressor package isin normal operation at the present time.

REFERENCES

Aghai, R. , 2003, “Turboexpander Technology Evaluation andApplication in Natural Gas Processing,” Proceedings ofEighty-Second Gas Processing Association AnnualConvention, San Antonio, Texas.

Agahi, R., Ershaghi, B., Leonard, M., Bosen, W. and Brunet, M., 1994,“The First High-Power Natural Gas Turboexpander/Compressorwith Magnetic Bearings and Dry Face Seals,” Proceedings of the23rd Turbomachinery Symposium, Revolve Technologies Inc.,Calgary, Alberta, Canada.

Bergman, D. and Nijhuis, H., 1996, “Full Power, Full Pressure,Closed Loop Test of a Natural Gas Turboexpander,”Proceedings of the Twenty-Fifth Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 89-94.

“Dr. Judson S. Swearingen,” 1999, ORBIT, 20, (4), pp. 68-69.


87

Takao Ohama is currently the President of KOBELCO EDTICompressors Inc., in Corona California, a subsidiary company ofKobe Steel, Ltd. He is in charge of both oil-free and oil-floodedscrew compressors for process gas and industrial refrigeration andmanages the company. Mr. Ohama’s previous career for 25 yearswas as an engineer for oil-flooded gas screw compressors andmanaging the screw compressor engineering group when he waswith Kobe Steel Ltd., Japan. He and his staff developed thehigh-pressure screw compressor H series in 1997, which is the firstto be applied to 60 barG in the world as a series and expanded thatrange to 100 barG. Mr. Ohama also participated on the Task Forcefor the preparation of API 619 Fourth Edition.

Mr. Ohama graduated with a B.S. degree (MechanicalEngineering, 1979) and an M.S. degree (Mechanical Engineering,1981) from the Saga University, Japan.

Yoshinori Kurioka is currently the Application EngineeringManager of KOBELCO EDTI Compressors Inc., in CoronaCalifornia, a subsidiary company of Kobe Steel, Ltd. He is involvedin proposals for both oil-free and oil-flooded screw compressors.Mr. Kurioka’s previous assignment in his 15 year career with KobeSteel Ltd., was as an engineer for API 619 type bare shaft oil-freescrew compressors, R&D engineer for air packaged type oil-freescrew compressors, and application engineer for both oil-floodedand oil-free screw compressors. He recently served on the TaskForce for the revision of API 619 Fourth Edition.

Mr. Kurioka graduated with a B.S. degree (MechanicalEngineering, 1989) and an M.S. degree (Mechanical Engineering,1991) from the Tohoku University, Japan.

Hironao Tanaka is currently a Project Manager of KOBELCOEDTI Compressors Inc., in Corona California, a subsidiarycompany of Kobe Steel, LTD. He is involved in engineering theexecution of oil-flooded and oil-free screw compressors for variousprocess gas applications. Mr. Tanaka’s previous assignment in his12 year career with Kobe Steel, Ltd., has been a system engineerand a project engineer for oil-flooded compressors.

Mr. Tanaka graduated with a B.S. degree (MechanicalEngineering, 1992) and an M.S. degree (Mechanical Engineering,1994) from Kobe University, Japan.

ABSTRACT

Oil-free screw compressors have been used for process gasapplication since the 1970s. Oil-flooded screw compressors havebeen used in many process related applications since the 1980s.Oil-flooded screw compressors are covered in the latest edition ofAPI Standard 619 issued in 2004. Both oil-free and oil-floodedscrew compressors have been expanding into process gas com-pression applications. It is therefore of interest to present theauthors’ recent experiences and share the acquired knowledge bycomparing features with reciprocating compressors and/orcentrifugal compressors.

High reliability, low maintenance costs, simple foundations, lowoperational costs, low initial costs, low consumed power atunloaded condition, and suitability for process fluctuation such asgas composition and pressure are some of the basic attributes of therotary screw compressors. These attributes have resulted in asignificant demand for such machines, primarily as an alternate toreciprocating compressors.

INTRODUCTION

The purpose of this paper is to present the experience acquiredin the use of oil-flooded screw compressors in certain process gascompression applications and highlight the key points as comparedto other types of compressors. In recent years rotary screwcompressors have been applied at higher pressure and largercapacity than before. This paper presents the special features ofscrew compressors and provides data from actual applicationshighlighting those features.

HISTORY

In the late 1950s, a Swedish company developed the oil-floodedtechnique in a screw compressor and perfected the rotor profile toachieve higher volumetric and compression efficiencies. They thenlicensed compressor manufacturers in the USA, Europe, and Japanto manufacture these compressors and collected royalties.

Since the screw compressors have characteristics of both rotary(centrifugal) compressors and positive displacement compressors(reciprocating), such machines found rapid acceptance inpetrochemical and gas processing industries. In 1975, API 619(2004) was introduced to specify a screw compressor. This firstedition of API 619 (2004) looked only at oil-free screw compressors.During this period, the oil-free screw compressor was applied in

89

PROCESS GAS APPLICATIONS WHERE API 619 SCREW COMPRESSORSREPLACED RECIPROCATING AND CENTRIFUGAL COMPRESSORS

byTakao Ohama

President

Yoshinori KuriokaApplication Engineering Manager

Hironao TanakaProject Manager

KOBELCO EDTI Compressors, Inc.

Corona, California

andTakao Koga

General Manager of Sales

KOBELCO EDTI Compressors, Inc.

Houston, Texas

many unique applications such as butadiene, styrene monomerrecycle gas, linear alkyl benzene, soda ash, etc. Most of theseapplications are subject to dust and liquids that are likely to bepresent in the gas stream. In many cases, water injection was usedto control the compression process.

In the 1980s, oil-flooded screw compressors started appearing inprocess gas applications. Around the same time, cogenerationstarted to take off with gas turbines becoming necessary in moreand more applications. Also, oil-flooded screw compressors werefinding their way into light gases such as helium and hydrogen.Less sensitivity to changes in molecular weight made suchcompressors particularly suitable for hydrogen pressure swingadsorption (PSA) compressors. On helium and hydrogen feedcompressors, stringent oil carryover requirements made itnecessary to introduce activated charcoal absorbers in the oilmanagement system. Carbon dioxide compressors for the beverageindustry switched to oil-flooded screw compressors with an oilremoval system down to 10 parts per billion (ppb) by weight.

In the 1990s, the demand for higher volume oil-flooded andoil-free screw compressors resulted in compressor manufacturersdesigning and building machines in large frame sizes.

By the mid 1990s, high pressure oil-flooded screw compressorsstarted to find their way into fuel gas boosters and manypetrochemical and refinery applications. At the same time oil-freescrew compressors were finding strong acceptance as vapor recoverycompressors in both offshore as well as onshore applications.

GENERAL DESCRIPTION OF THETHREE TYPES OF COMPRESSORS

Before introducing actual applications, one needs to understandthe compression mechanism and typical mechanical limitation forcentrifugal, reciprocating, and screw type compressors.

Centrifugal compressors are continuous flow machines in whichone or more rotating impellers accelerate the gas as it passesthrough the impellers, which are shrouded on the sides. Theresultant velocity head is then converted into pressure. This occurspartially in the rotating element and partially in the stationarydiffuser.

Reciprocating compressors are positive displacement machineswith a piston compressing the gas in a cylinder. As the pistonmoves forward it compresses the gas into a smaller space, thusraising its pressure. There are two types of reciprocating compressors,called “lube” type with oil injection and “nonlube” as oil-free.

Screw compressors are also positive displacement machines butrotating twin rotors act as pistons that compress the gas in a rotorchamber (casing). Compression is done continuously by the rotation ofthe twin rotors. There are also two types of screw compressors: the “oil-flooded” type with oil injection, and “oil-free” with no oil injection.

Pressure, flowrate, and gas composition are the major factors to beconsidered in selecting the type of compressor. Table 1 showscomparison of three types of compressors with respect to pressures,flows, and gas compositions, etc.

Table 1. Comparison Table of the Three Types of Compressors.

Generally, reciprocating compressors are suitable for highpressure ratios, low flow, and low megawatt (MW) applications.

Centrifugal compressors are suitable for large flowrates. Screwcompressors are suitable for the following conditions.

• Pressure ratio limitations—Since it is a positive displacement typecompression, and has no valve movement, a high pressure ratio can beachieved. On oil-flooded screw compressors, there is no mechanicallimitation for pressure ratio. The only concern is efficiency.

• Capacity control—Oil-flooded screw compressors have an unloadercalled a slide valve and can provide stepless turndown (typically 100percent to 15 percent) with corresponding reduction in power.

• Impact of molecular weight of gases—There is almost no impactof molecular weight of the gases upon the performance of an oil-flooded screw compressor. Injected oil is a sealant and leakage iscontrolled. Therefore these compressors are highly efficient foreven the lowest molecular weight gases.

• Gases containing dust and polymers—In oil-free screwcompressors, any type of gas can be compressed. This is practicalbecause compression is done by displacement with continuosrotation, the rotor shaft is rigid so that effect of unbalance islimited, and there are no internal valves to hinder the operationfrom dust and polymers.

• Availability—High reliability resulting in compressor availabilityis the same as centrifugal machines and allows single machineoperation without a spare in critical services.

GENERAL DESCRIPTION OFOIL-FREE SCREW COMPRESSORS

A cutaway drawing of a typical oil-free screw compressor isshown in Figure 1. There are two rotors inside the casing of thescrew compressor. One rotor is referred to as male, and theother rotor is the female. The male rotor and the female rotormaintain a small clearance and do not contact each other. Tokeep phase with each other, a timing gear is furnished to drivethe other rotor.

Figure 1. Typical Cutaway Drawing of Oil-Free ScrewCompressor.

To isolate the rotor chamber from the bearing with an oilatmosphere, seals are furnished next to the rotor lobe on eachend of the machine. There are journal bearings outside the sealarea, which are typically sleeve type hydrodynamic bearings.Thrust bearings are located on the outer side of the journalbearings, and tilting-pad type is typically used.

The following are the major characteristics of the oil-freescrew compressors:

• Process gas is completely free of oil, there is no contamination, andtherefore any gas can be handled. In oil-free screw compressors, due to


the positive displacement compression, even polymer gas or dirty gascan be compressed.

• The rotor speed is higher than with oil-flooded screw because ofno oil turbulence in the rotor chamber, but does not exceed anycritical speed since the rotor shaft is to remain rigid. Rotor speed istypically higher than an oil-flooded screw machine so compressorframe size can be smaller than the oil-flooded type.

• Discharge temperature is typically high because of compressionheat. To avoid excessive heat deformation, cooling is required.Some applications utilize a process compatible fluid such aswater or solvent to cool the gas directly by injection into therotor chamber inlet.

• Due to its longer rotor span for seal area, rotor clearance, andlimits on discharge temperature, pressure ratios are limited for theoil-free screw compressors.

• Because of its high rotational speed, noise is rather high so thatsilencers on suction and discharge nozzles are typically required.Expansion and/or absorption type silencers are typically used incombination or separately. Frequency of the noise is high becauseits main frequencies are pocket passing frequency (rotationalspeed*lobe number) or its harmonics. The major noise is measuredat discharge piping. In the authors’ experiences it is apparent thatthe expansion is good for several discrete frequencies. The size ofthe expansion type silencer can be optimized by using a simulationto target the specific frequencies, i.e., pocket passing frequencyand its harmonics. To absorb this high frequency noise, internalabsorption type silencers are considered to be more effective thanexternal absorption type. Absorptive method is effective inabosorbing pulsation energy of frequencies ranging from 500 Hz toseveral thousand Hz. Further experience confirms the use of acombination of absorption and expansion type silencers to be moreeffective in noise reduction. By expansion type, 15 to 20 dB ofsound pressure level inside the piping can be reduced whereas, byinternal absorption, 25 dB of sound pressure level inside the pipingcan be reduced. In addition to the silencers, a noise enclosureenclosing just the compressor and gearbox is typically required ifthere is a sound requirement of 85 dBA at 1 m (3 ft) from thecompressor skid edge.

GENERAL DESCRIPTION OFOIL-FLOODED SCREW COMPRESSORS

A cutaway drawing of a typical oil-flooded screw compressor isshown in Figure 2. There are two rotors inside the casing as withthe oil-free screw compressors. However, here they contact eachother at lobe surface via an oil film.

Figure 2. Typical Cutaway Drawing of Oil-Flooded Screw Compressor.

Oil is supplied not only to the bearing and seal, but also tothe rotor chamber directly and oil will act as lubricant, coolant,

and sealant in the rotor chamber. Typically, the male rotor isdriven by a directly coupled two-pole or four-pole electricmotor and drives the female rotor. An external gear unit istypically not used since the tip speed of the oil-flooded screwcompressor is in the proper design range when driven at motorspeed. Since oil is injected into the rotor chamber, the seal areabetween the lobe and bearing is no longer necessary. There isone mechanical seal located at the drive shaft end. There aretypically sleeve type journal bearings on either end of the rotorlobes. Thrust bearings are typically tilting-pad type and arelocated on the outer side of the journal bearings.

The oil and gas mixture is discharged through the compressordischarge nozzle into an oil separation system locateddownstream of the compressor. Oil separated in the oilseparation system is circulated in the compressor lube system.

An unloaded slide valve is located in the compressor justbeneath the twin rotors and is used to adjust the inlet volume.The inlet volume of the compressed gas can be adjusted bymoving the slide valve, which is actuated by a hydrauliccylinder. A typical schematic diagram for an oil-flooded screwcompressor is shown in Figure 3.

Figure 3. Typical Schematic Diagram for an Oil-Flooded ScrewCompressor.

Compressor lubricant oil is present in the process side, so thelube oil selection is very different from other types of machines.The bulk of the oil is separated in the primary oil separator, but asecondary coalescing oil separator may be used as an additionalseparator. Separation of oil is one of the important factors for oil-flooded screw compressors. Typically, a combination of demistermesh pad and coalescing elements are used. For example, 0.1parts per million by weight (ppm wt) level can be achieved bycombination of a demister mesh pad and two stages ofcoalescing elements. Charcoal absorbers are occasionally usedfor more severe applications. Borocilicate microfiber is a typicalmaterial used in coalescing elements and submicronic particlesof oil can be separated from the compressed gas. Unlikereciprocating compressors, oil from the compressor has nodeterioration by piston rubbing so oil can be recirculated in the


91

system as lubricant for longer life. The lube oil circulation systemconsists of compressor lube lines, oil cooler, oil filters, and oilpump. The oil pump may be double or single configuration. Thedesign of a single pump system may be used when the pump isrequired only during startup. In such case, after the compressorstarts and discharge pressure is established, oil can circulate inthe system by utilizing gas differential pressure between suctionand discharge.

A slide valve is used to load and unload the compressor tomaintain suction pressure or discharge pressure. There is a spoolvalve actuated by air with solenoid valves to switch over the oillines to pressurize the slide valve cylinders to the load side or theunload side. A typical control range by slide valve is from 15percent to 100 percent stepless by inlet volume.

Below is a list of some of the major characteristics of theoil-flooded screw compressor:

• Power consumption savings by a built-in slide valve—The slidevalve as unloader adjusts the inlet volume of the compressor,and this equates as power savings. Figure 4 shows the basicprinciple of the slide valve mechanism. The slide valve islocated just beneath the rotors and moved in the axialdirection. The slide valve is moved typically by hydrauliccylinder with oil utilized from the compressor lube oil line.Moving the slide valve to the suction side attains full load, andunloading is achieved by moving the slide valve toward thedischarge port. At full load position, the entire length of therotor is utilized to draw the gas so that inlet volume of thecompressor can be maximized. By moving the slide valve tothe unloaded position (i.e., discharge side), the length of thecompression chamber is shortened. As a result, inlet volume ofthe compressor is reduced. Compression is done with less inletvolume of the compressor so that theoretical brake horsepoweris reduced.

Figure 4. Basic Principle of the Slide Valve Mechanism.

• High compression ratio limitation—Since the oil acts as acoolant and sealant the limit on compression ratio is veryhigh. Discharge temperature can be adjusted by oil flowrate,i.e., oil can be injected into the rotor chamber to absorb thecompression heat in the oil-flooded screw compressors.

When a very high pressure ratio is required, a tandemarrangement of two stage compressors combined in onecasing is available. Typically, this tandem arrangement isused when the pressure ratio is larger than 7:1, and can beapplied to ratios of more than 50:1. A typical cutaway drawingof a tandem arrangement oil-flooded screw compressor isshown in Figure 5. Since oil will act as a coolant at theintermediate stage, an external intercooler with piping forintermediate stage is unnecessary.

Figure 5. Typical Cutaway Drawing of a Tandem ArrangementOil-Flooded Screw Compressor.

• Low maintenance cost—Due to the lube oil system the rotors andmany other parts of the compressor have an oil film on their surface.The life of the rotors is long enough so that a spare set is not required.The mechanical seal is typically one per casing, so maintenance andreplacement cost for the seal are typically reduced.

• Single skid arrangement—The compressor and lube oil systemare integrated and packaged on a single skid. Thus, transportationand installation are completed in a short period.

• No cooling water jacket/no gas bypass cooler—Since oil acts ascoolant in the compression process, discharge temperature can becontrolled by the oil injection flowrate so that the casing structureis made simpler by elimination of a cooling water jacket. The gasbypass cooler can also be eliminated by oil cooling.

• Selection of oil is driven by the need to be compatible withprocess gas. Not only mineral-based oil, but synthetic oil hasrecently been used to expand the application range of oil-floodedscrew compressors. Hydrotreated mineral-based oil has typicallybeen used, but recently many are changing to synthetic oil. Thereare two kinds of synthetic oil: one is polyalphaolefin (PAO), andthe other is polyalkylene glycol (PAG). With PAG there are severalkinds of oil that differ in ratio of propylene oxide (PO) andethylene oxide (EO). For a process with a heavy hydrocarbon, bothmineral-based oil and PAO are subject to dilution; however, lessdilution can be expected with PAG. There is no difference fordilution ratio by process with heavy hydrocarbon between mineral-based oil and PAO; however, less dilution can be expected for PAG.

• Dilution rateMineral oil 5 PAO . PAG(PO).PAG(EO1PO).PAG(EO)PAG with EO 5 100 percent is hygroscopic; however,

it has no dilution for heavy hydrocarbon. By using PAG oil,oil-flooded screw compressors are now able to be used for heavyhydrocarbon applications as in refinery services.

GENERAL COMPARISON BETWEENDIFFERENT TYPES OF COMPRESSORS INSOME APPLICATIONS AND RECENT SITUATIONS

Hydrogen Service

Hydrogen is widely used in oil refining processes and manyprocesses in petrochemical fields. Hydrogen is typically generated inpressure swing absorption, membrane, or electrolyzing systems.Hydrogen generated by the above methods is usually produced atatmospheric pressure and then compressed typically up to 30 barG(435 psiG) by compressors.

Due to the very low molecular weight of hydrogen and high pressureratio needed, centrifugal compressors or oil-free screw compressors arerarely used for such applications. Reciprocating compressors have beentypically used in this service. Due to the advantage of low maintenancecost, oil-flooded screw compressors are increasingly being applied for


this application. Table 2 shows comparison for oil-flooded andreciprocating compressors in typical hydrogen service.

Table 2. Typical Comparison Table for Hydrogen Service.

As shown in Table 2, the reciprocating compressor has anadvantage of total brake horsepower (BHP) due to multistagecompression with an intercooling system. However, the oil-floodedscrew compressor has the slide valve to save power at the unloadcondition. Figure 6 shows the typical package for hydrogen servicewith an oil-flooded screw compressor.

Figure 6. Typical Package for Hydrogen Service Using Oil-FloodedScrew Compressor.

Oil-flooded screw compressors have an advantage due to thesmaller amount of installation area needed and less weight tosupport. In the case of a tandem arrangement, which is two stagecompressors arranged in one casing, compact skid arrangementscan be adopted on oil-flooded compressors. Another advantage ofthe oil-flooded screw compressor is longer times betweenmaintenance. The typical maintenance period for a reciprocatingcompressor is one to two years, while an oil-flooded screwcompressor is two to four years. Reciprocating compressorapplications typically require a spare compressor, so investmentand installation costs are doubled.

The gas industry field requires a longer maintenance periodsuch as two to four years. Equipment demands requirecontinuous operation, and the oil-flooded screw can meet thisdemand. Oil carryover from the oil-flooded screw compressoris managed by oil coalescing systems, which can reduce

carryover to 1.0 ppm—in some applications by adding charcoalabsorbers. Less than 50 ppb carryover by weight is achieved.

Vapor Recovery Unit (VRU)

In most offshore platform applications, crude oil or natural gasdrilling produces vapor gas as a by-product. This vapor by-productneeds to be recovered for environmental reasons. As a result, vaporrecovery units together with compression systems are used. Thetypical gas composition and operating condition is shown in Table 3.

Table 3. Typical Gas Composition and Operating Condition for VRU.

Gas composition of the recovered vapor can change due to welllocation and the age of the well. Even from the same well, the vaporgas composition and flowrate can fluctuate. Centrifugal compressorshave difficulty in this application because of unsteady gas compositionand flowrate. In recent years, lower costs have increased the use of oil-flooded and reciprocating compressors in this application.

In comparison with reciprocating compressors, oil-flooded screwcompressors are more widely used due to their gas flow adjustmentcapabilities, which can be adjusted by the internal slide valve. However,the “unpredictable” gas composition sometimes contains seriousamounts of sulfur, tar, or other unknown corrosive components as well asheavy hydrocarbons that are always present. Also, there are somedifficulties in using oil-flooded compressors due to serious dilution of oil.

Oil-free screw compressors are being increasingly used where thespecific heat coefficient (k value) is rather small and the dischargetemperature is lower for higher pressure ratios. A typical packagewith oil-free screw compressor for VRU is shown in Figure 7. Asshown in the picture, the skid needs to be very compact due torestriction of space, which is also a very important factor on VRU.

Figure 7. Typical Package for VRU Using Oil-Free Screw Compressor.


93

Lube oil does not come in contact with process gas withinoil-free screw compressors. Therefore, there are no dilutionproblems. In addition, heat insulation and electronic heat tracingare required to avoid condensation of gas in oil-flooded screwcompressors when the compressor is not running. Duringoperation, process gas temperatures need to be kept higher thandew point to avoid dilution of oil in oil-flooded screw compressors.In the case of an oil-free screw compressor, there is no concern dueto condensation of gas. Therefore the overall system is simple.

Regarding the gas flow change, oil-flooded screw compressorshave an advantage with adjustment by the internal slide valve andpower savings. However, the rate of change of the gas flow is veryslow, typically 20 to 30 years of operation, and generally changes overthe life of the field. Oil-free screw compressors can accommodate thischange by adjusting the operational speed. Replacement of gear andpinion combinations in a speed increasing gearbox makes thisprocedure possible. These parts are interchangeable and can bereplaced and maintained. A comparison table between oil-free screwcompressors and oil-flooded screw compressors is shown in Table 4.

Table 4. Comparison of Screw Compressor Features for VRUBetween Oil-Free and Oil-Flooded Types.

Fuel Gas Booster for Gas Turbines

Recently, the efficiency of generating electrical power by gasturbines has been significantly improved. High efficiency type gasturbines are used in many power plants utilizing natural gas as fuel.Many gas turbines require higher supply pressure of the fuel attypically 30 barG to 50 barG (450 psiG to 725 psiG) and natural gaspressure coming out from the pipeline is low. To boost the fuel gas tothe required pressure of the gas turbine, a fuel gas booster is required.For the fuel gas booster application, reciprocating compressors andcentrifugal compressors have been used primarily.

In the 1990s, high pressure oil-flooded screw compressors weredeveloped and started to be used for fuel gas booster applications.The oil-flooded screw compressors are very suitable for theseapplications, since the requirement of the fuel gas booster fits verywell with characteristics of oil-flooded screw compressor, i.e.:

• Suction pressure fluctuations

• Gas turbine load fluctuation, i.e., flowrate fluctuation

• Unstable gas composition (typically pipeline quality natural gas)

Also, this fuel gas booster application requires economicaloperation and the oil-flooded screw compressors with a slide valveas an unloader can provide significant power savings.

Because of suction pressure fluctuations, the compressor needsto be sized according to the design point, which is the lowestsuction pressure in specification. However, the actual suctionpressure is typically higher than the design point so that thecompressor is always operated in a partially unloaded condition. Atypical unload performance curve is shown in Figures 8 and 9.

Figure 8. Typical Unload Curve When Mass Flowrate is Constantwith Suction Pressure Change.

Figure 9. Typical Unload Curve When Suction Pressure isConstant with Mass Flowrate Change.

Oil-flooded screw compressors can be operated at higher suctionpressures by utilizing a slide valve with less brake horsepowerrequired in the unloaded condition. Reciprocating compressors andcentrifugal compressors cannot accommodate the big fluctuationof suction pressure so a suction control valve is typically requiredto control suction pressure close to design point at the compressorinlet. Thus, large power energy savings cannot be acquired withthese machines. The flowrate is typically rated with some range forgas turbines, since consumed fuel gas flowrate is varied by


atmospheric temperature. As a result, the compressor needs to becapable of operation at lower flowrates than the rated point. Theoil-flooded screw compressor can meet this demand by using theslide valve with power savings as well (refer to Figure 9).

This application typically requires a suction scrubber sinceunexpected water or liquid may be present in the gases. Gasturbines require very precise delivery pressure of fuel gas.Discharge pressure needs to be maintained regardless of the suctionpressure swing so that a spillback (bypass line) is always requiredfor quick load and suction pressure changes. The screw compressorslide valve control system accommodates these large changes.

In addition to the above, automatic operation is required withgas turbine operation so that a control panel with programming istypically required. Figure 10 shows a typical package for a fuel gasbooster unit using an oil-flooded screw compressor.

Figure 10. Typical Package for Fuel Gas Booster Service UsingOil-Flooded-Screw Compressor.

Except for large size machines, all equipment can be mounted ona single skid, including the oil separation system, suction scrubber,spillback line, and control panel. Sometimes, the compressorforward bypass line is provided when maximum suction pressure isabove the discharge pressure. A typical comparison of the screwcompressor features for a fuel gas booster application betweenoil-flooded screw, centrifugal, and reciprocating compressors isshown in Table 5.

Table 5. Typical Comparison Table for Fuel Gas Booster.

As shown in Table 5, brake horsepower at the design rated pointhas almost no difference among three types of compressors.However, there is a large difference at normal operation point andwhen the suction pressure is higher than design and less flowrate.From a cost and installation point of view, the oil-flooded screwcompressor has significant advantage for such applications.

Desulfurization Compressor

Recently, demand for desulfurization of vehicle gasoline anddiesel fuel is increasing all over the world. New regulations toprotect the environment have forced the oil refinery industry todevelop a desulfurization process. For this process, gas compressionis necessary mainly utilizing a hydrogen mixture.

The oil-flooded screw compressor has been proven in thisprocess, and demand for the screw compressor is increasing in thisapplication. Table 6 shows a typical comparison of compressors fordesulfurization process between oil-flooded screw, centrifugal, andreciprocating compressors.

Table 6. Typical Comparison Table for Desulfurization Compressors.

Hydrogen is the main gas component and H2S is typicallyincluded in the gas stream in ppm level. Gas composition is notstable due to change of desulfurization process and nitrogenoperation is required at startup. Therefore the compressor needs tohave the capability of operating at various conditions of gascomposition.

Pressure condition is typically very low pressure. Howeversuction pressure is higher when discharge pressure is high, whichcan change case by case with the process. The end users are oilrefineries, so longer times between maintenance periods and highreliability are required. In the past reciprocating compressors andcentrifugal compressors were typically used for this application.However, demand for the oil-flooded screw compressors hasbeen increasing.

In the 1990s oil-flooded screw compressors suitable for highsuction pressure and low pressure ratio were developed. The oil-flooded screw compressor can be suitable for gas compositionchanges due to positive displacement type of compression. Theslide valve allows the compressor to handle pressure and flowchanges with power savings.

Other than the desulfurization compressor, there is anotherapplication in the desulfurization process called “net gas booster,”which requires higher pressure ratio and larger size. Since this netgas booster contains hydrogen the oil-flooded screw compressorhas started to be used for this application instead of reciprocatingcompressors, for longer maintenance periods. A typical packageusing an oil-flooded screw compressor is shown in Figure 11. Anoise enclosure is not typically required for oil-flooded screw


95

compressors due to low noise, so the accessibility to the compressoris secured, which is also a very important factor from amaintenance standpoint.

Figure 11. Typical Package for Desulfurization Service UsingOil-Flooded Screw Compressor.

Application Chart

To get a better understanding, Figure 12 shows an applicationchart where the applications in this paper fall with each type ofcompressor. Although screw compressor applicable range isconfined to reciprocating compressor, and centrifugal compressorrange, there are applications with ranges where screw compressorsare used, as referred to in this paper because of the manyadvantages in using screw compressors.

Figure 12. Application Chart with Each Type of CompressorTypical Applicable Range.

CONCLUSION

Oil-free and oil-flooded screw compressors can be applied inmany applications. Some reasons for considering the screwcompressor are changes in process conditions, recent progress incompressor technologies, and application range of screwcompressors. There are many benefits for the customer such ashigh reliability, low initial cost, less maintenance cost, andpower savings.

REFERENCES

API Standard 619, 2004, “Rotary Type Positive DisplacementCompressors for Petroleum, Petrochemical, and Natural GasIndustries,” Fourth Edition, American Petroleum Institute,Washington, D.C.

BIBLIOGRAPHY

Ohama, T., Amano, Y., and Kawaguchi, N., 2000, “High PressureOil-Flooded EH Series Screw Compressors,” KobelcoTechnology Review, (23).

Ohama, T., Koga, T., and Kurioka, Y., 2004, “High PressureOil-Injected Screw Gas Compressors (API 619 Design) for HeavyDuty Process Gas Applications,” Proceedings of the Thirty-ThirdTurbomachinery Symposium, Turbomachinery Laboratory, TexasA&M University, College Station, Texas, pp. 49-56.


Paul B. Gallick is a Senior Applications Engineer for ElliottCompany Ebara Group in Jeannette, Pennsylvania. He has workedfor Elliott as a Compressor Product Engineer, a Lubrication and SealOil Systems Engineer, and currently as a Compressor ApplicationEngineer, for more than 21 years. He is responsible for compressorselection, specification review, and other aspects related to thepreparation of quotations for new and rerated compressors.

Mr. Gallick has a B.S. degree (Mechanical Engineering, 1978)from the University of Pittsburgh and is a registered ProfessionalEngineer in the Commonwealth of Pennsylvania.

Greg Phillippi is the Director of Process Marketing and Salesfor Ariel Corporation, in Mount Vernon, Ohio. He began his careeras a design engineer with Cooper Energy Services in 1978. In 1985he accepted a position as a Design Engineer with ArielCorporation. From 1985 to the end of 1999 he worked as a DesignEngineer and Design Engineering Manager at Ariel. In 2000, Mr.Phillippi accepted a position with ACI Services, Inc., inCambridge, Ohio, where he was deeply involved with marketing,sales and engineering. In January 2004, he moved backed to Arielin his present role. He has significant experience in the design andapplication of reciprocating compressors.

Mr. Phillippi has a BSME degree (1978) from Ohio NorthernUniversity and an MBA degree (2000) from Ashland University.

Benjamin F. Williams is an Applications Engineer for ArielCorporation, in Mount Vernon, Ohio. He primarily focuses onprocess and international applications for Ariel Corporation,where he has worked since 1997. Prior to working at Ariel, heworked for Lone Star Compressor Corporation, in South Houston,Texas, from 1985 to 1997. During his 20+ years in the compressorindustry (the majority in process compression), Mr. Williams hasworked in design, application, reapplication, and service of manydifferent compressor makes and models. He is specially trained in

Nuclear Propulsion Engineering and served on two nuclearsubmarines: the USS John Adams and the USS Nathan Hale.

ABSTRACTThis tutorial addresses the question of which compressor

type is better suited to a given application—a centrifugal orreciprocating design. The general application map will bepresented and discussed, as will the advantages and disadvantagesof each type of compressor. The application guidelines will beaddressed from the standpoint of reliability, cost, efficiency, size,and other more general application parameters such as molecularweight, compression ratio, and flow range, etc.

The intent of the tutorial will be to provide guidelines andcomparative information to be used by contractors and users todetermine which type of compressor will be the best fit for theirparticular application.

INTRODUCTIONThe tutorial is organized into four sections. The first, “HOW A

RECIPROCATING COMPRESSOR WORKS,” will be a short,very basic explanation of how a recip compressor works. The nextsection, “HOW A CENTRIFUGAL COMPRESSOR WORKS,”will do the same for a centrifugal compressor. The goal with thesetwo sections is to serve as a primer for the rest of the tutorial. Thethird section, and the primary content of the tutorial, will be wherethe two different machines are compared and contrasted. Finally,the fourth section is entitled “CASE STUDIES.” Here 12 differentsets of application conditions (four sets of inlet and outlet pressureseach at three different gas mole weights) are used to compare theperformance of the two machines.

HOW A RECIPROCATING COMPRESSOR WORKSHow a reciprocating compressor works will be explained by

discussing pressure versus time (Figure 1) and pressure versusvolume (Figure 2) diagrams.

113

WHAT’S CORRECT FOR MY APPLICATION—A CENTRIFUGAL OR RECIPROCATING COMPRESSOR?

byPaul Gallick

Senior Applications Engineer

Elliott Company

Jeannette, Pennsylvania

Greg PhillippiDirector Process Marketing and Sales

andBenjamin F. Williams

Application Engineer

Ariel Corporation

Mount Vernon, Ohio

Figure 1. Pressure Versus Time Diagram.

Figure 2. Pressure Versus Volume Diagram.

The P-T diagram (Figure 1) is a plot of the pressure of the gas inthe compression chamber versus the angle of crankshaft rotation,which is essentially time because crank angle and time are directlyrelated. Understanding the P-T diagram will help in understandingthe P-V diagram, which underlies all the theory of reciprocatinggas compressor operation.

The P-V diagram (Figure 2) is a plot of the pressure of the gastrapped in the compression chamber versus the volume of gastrapped in the compression chamber. The volume of gas trappedin the compression chamber is not linearly related to crankangle or time, so the two diagrams have different shapes andpurposes.

A reciprocating compressor is a positive displacement machine—meaning a certain volume of gas is drawn in to the compressionchamber where it is trapped, compressed, and released.

HOW A CENTRIFUGAL COMPRESSOR WORKS

A centrifugal compressor is a dynamic type of compressorwhere the pressure rise is accomplished by transfer of dynamic(motion-related) energy from the rotor to the gas.

• The basic aerodynamic components of a centrifugal compressorare the impellers, diffusers, and return channels. (Figure 3)

• Velocity (kinetic energy) is imparted from moving blade to gasin the impeller (Figure 4).

• Velocity (kinetic energy) is converted to pressure (potentialenergy) in the diffuser.

• For a given pressure ratio, more head is required to compress alow molecular weight gas than for a higher molecular weight gas(Figure 5).

• Increasing speed imparts higher kinetic energy, which convertsto higher pressure (potential energy) (Figure 6).

• The head/pressure ratio curve is essentially constant withchanging volume flow at fixed speed (Figure 7).

• Centrifugal compressor performance can be estimated using thefan laws. According to the fan laws, capacity is proportional tospeed, head is proportional to the square of the speed, and power isproportional to the cube of the speed. Since these laws assumeideal gases with constant k and Z values, they apply withreasonable accuracy to single-stage compressors or multistagecompressors with low pressure ratios. For a multi-stage centrifugalcompressor with a high pressure ratio, the laws still applydirectionally, but the accuracy of performance estimates using thefan laws is insufficient for most calculations.

Figure 3. Aerodynamic Parts of a Centrifugal Compressor.

Figure 4. Impeller Discharge Gas Velocity.

Figure 5. Gas Mole Weight Effect on Head.


Figure 6. Effect of Speed on Head.

Figure 7. Effect of Speed Range on Overall Flow Range.

CENTRIFUGAL VERSUSRECIPROCATING COMPARISON

The following text is organized by topic with comments for thattopic relating to reciprocating and centrifugal compressors.

Maximum Discharge (Outlet) Pressure

• Reciprocating—The “typical” reciprocating compressor is used fordischarge pressures up to 12,000 psi (828 bar). Special compressors(called hypercompressors) are used in low density polyethyleneproduction and discharge at pressures up to 50,000 psi (3500 bar).

• Centrifugal—Discharge pressures to 1450 psi (100 bar) forhorizontally split compressors. Discharge pressures up to 15,000psi (1034 bar) for radially split (barrel) compressors.

Minimum Suction (Inlet) Pressure

• Reciprocating—Can be applied with suction pressures atatmospheric or even a slight vacuum. In vacuum applications,precautions must be taken to prevent atmospheric air from leakinginto the cylinder through the piston rod packing.

• Centrifugal—Inlet pressures to atmospheric or below. Forsubatmospheric inlet conditions, special seal and buffering designsare employed to keep atmospheric air from being drawn into thecompressor.

Maximum Discharge (Outlet) Temperature

• Reciprocating—Discharge temperature limits will depend on theapplication and the seal element materials selected. In hydrogen richapplications, API 618 (1995) limits discharge temperatures to 2758F(1358C). For natural gas service the maximum discharge temperaturelimit is 3508F (1758C); although in most cases a more practical limit

is 3008F (1498C). Air compressors may run at discharge temperaturesin excess of 4008F (2048C).

• Centrifugal—Maximum temperature based on the compressordesign itself is typically 400 to 4508F (204 to 2328C). Highertemperatures are possible but require special designs such as centersupported diaphragms, less efficient seal materials, and hightemperature O-rings and sealants.

The process may also have discharge temperature limitationsdue to fouling, temperature limits of downstream components, andprocess efficiency.

Minimum Suction (Inlet) Temperature

• Reciprocating—The common compressor cylinder materials,cast gray iron and cast ductile iron, are acceptable for use attemperatures as low as 2408F (2408C) which typically occur inrefrigeration applications.

The lowest suction temperatures requested typically are in liquefiednatural gas boil-off gas applications. These inlet temperatures can beas low as 22608F (21628C). Compressor cylinders used for thisapplication require very special materials and are not offered by allmanufacturers.

• Centrifugal—Standard centrifugal compressor materials aretypically suitable for 220 to 2508F (219 to 2468C).

Refrigeration compressors in ethylene service typically haveinlet temperatures as low as 21558F (21048C), which requirespecial low temperature compressor materials.

Similar to reciprocating compressors, the lowest temperaturerequirement for centrifugal compressors is typically found inliquefied natural gas (LNG) boil-off gas applications. Centrifugalcompressor designs and materials for this service can accommodateminimum temperatures down to 22758F (21718C). Special lowtemperature stainless steels are typically used for this service.Special low temperature seals and O-rings are also required.

Maximum Flow

• Reciprocating—Reciprocating compressors are positivedisplacement type compressors. Capacity is limited by cylindersize, the number of throws available, and the availabledriver speeds. A “throw” is a location on the crankcase whereacompressor cylinder can be attached.

• Centrifugal—Centrifugal compressors can be sized for an inletflow of 400,000 acfm (680,000 m3/hr) in a single body.

The maximum flow through a centrifugal compressor is limitedby the choke point, which is the point at which the flow throughsome part of the compressor nears a velocity of Mach 1.

Minimum Flow

• Reciprocating—Similar to the maximum flow, the minimum flowin a reciprocating compressor is limited by cylinder size, stroke, andspeed. Very small reciprocating compressors are available.

• Centrifugal—Centrifugal compressors can be sized for flow aslow as a few hundred acfm. Unlike a reciprocating compressorwhere minimum flow is solely a function of compressor geometryand speed, the minimum flow for a centrifugal compressor islimited by an aerodynamic condition known as surge, which isa function of compressor geometry, speed, aerodynamic gasconditions, and system resistance.

Flow Range

• Reciprocating—Reciprocating compressors have the ability tochange flow (throughput) through speed control, the addition of fixedclearance to a cylinder (fixed or variable volume clearance pockets),cylinder end deactivation, and system recycle. Typical flow rangemight be 100 percent down to as low as 20 percent, and even lower.


115

The application will determine what type of capacity controlmethod is required and used. On low compression ratio applications(compression ratio less than 1.6, such as natural gas pipeline gas trans-mission) adding fixed clearance will have little if any effect on flow.These applications may use speed control or cylinder end deactivation.

In other applications, with higher compression ratios, it is quitecommon to use clearance pockets and cylinder end deactivation toregulate flow.

• Centrifugal—As discussed above, the flow range of a centrifugalcompressor is set by the surge and choke points.

Typical turndown for a fixed speed, multistage centrifugalcompressor is approximately 20 to 30 percent. With speedvariation or adjustable inlet guide vanes, the turndown can beincreased to 40 to 50 percent or more (Figure 9).

Weight

• Reciprocating—The weight of a reciprocating compressorobviously varies with size, which varies with speed, stroke, and rodload rating.

It is safe to say that on a mass per power basis the centrifugalcompressor will be lighter.

• Centrifugal—The weight of a centrifugal compressor variesdepending on compressor size (refer to “Size” below) and, to asmaller degree, on materials of construction.

The driver, baseplate, and auxiliary systems contributesignificantly to the weight of a centrifugal compressor package.

Size

• Reciprocating—Reciprocating compressors come in a widevariety of sizes. The size and weight of a reciprocating compressoris directly related to stroke, speed, and rod load rating. Stroke, inturn, can be related to the driver speed. In general, the higher thedriver rotating speed, the shorter the stroke and therefore thesmaller the compressor. Conversely, a slower speed compressorwill typically have a longer stroke and be physically larger.

• Centrifugal—The size of a centrifugal compressor is mainly afunction of flow capacity (sets the diameter) and number of stages(sets the length).

Casing outer diameters range from as small as 20 inches (500mm) to as large as 150 inches (3800 mm).

Reliability

• Reciprocating—Reciprocating compressors will probably neverbe as reliable as centrifugal compressors. The recip has many moreparts and more rubbing seals (pressure packing, piston rings, andrider rings) that wear and require more frequent replacement thanany seal or other part in a centrifugal.

In addition a recip has compressor valves, which though verysimple mechanical devices (simple spring-loaded check valves),require considerable maintenance.

Another significant factor affecting the reliability of the recip isthe cleanliness of the process gas. Wear life of the seals and thevalves will be considerably longer with a process gas that is free ofliquid and solid debris.

• Centrifugal—Reliability/availability of centrifugal compressorsis typically 98 to 99 percent.

Typical Maintenance Intervals

• Reciprocating—Maintenance intervals for a reciprocatingcompressor vary significantly with the application and followalong with the comments made in the reliability section.

Compressor valve and seal element intervals might be as short asa few months and as long as three to five years (some applications

may be even longer). This depends so much on the specifics of theapplication and the cleanliness of the gas.

A major overhaul (typically defined as completely goingthrough the machine to the point of replacing the bearings) may berequired only every 10 years or longer.

• Centrifugal—Per API 617 (2002), Seventh Edition, a centrifugalcompressor has to be designed for at least five years ofuninterrupted service. In clean service, a centrifugal compressorcan operate continuously for 10 years or longer.

Maintenance requirements are typically limited to replacingbearing pads and seal wearing parts.

Compressed Gas Molecular Weight

• Reciprocating—A reciprocating compressor has no limit withregard to molecular weight. Very light and very heavy gases arecompressed equally well. Over the range of molecular weightdifferent application configurations may be required. For example,very low molecular weight gases may present some seal challengesand very high molecular weight gases pose issues with efficiency.But nonetheless, the recip handles the whole range quite well.

• Centrifugal—Compression ratio is highly dependent onmolecular weight. Head is developed by increasing gas velocity tocreate kinetic energy and then converting the kinetic energy topressure in the diffuser. The amount of kinetic energy is a functionof gas velocity and mass or molecular weight.

Centrifugal compressors are used for a broad range of molecularweight including low molecular weight applications such ashydrogen recycle and high molecular weight applications usingrefrigerant gases with molecular weights over 100.

Compression Ratio

• Reciprocating—The maximum compression ratio that areciprocating can handle in one stage is limited mostly bycompressed gas discharge temperature. The piston rod loadgenerated by the compression ratio may also be a limit.

Typical compression ratios are 1.2 to 4.0.

• Centrifugal—Compression ratio is a function of gas molecularweight, compressibility, stage geometry, compressor speed, and thenumber of compressor stages.

For a specific gas, the limits to compression ratio are themechanical and rotordynamic limitations on speed and the numberof stages that can be accommodated in a single body. Dischargetemperatures resulting from high compression ratios can usually becontrolled by intercooling.

Materials

• Reciprocating—Reciprocating compressors are made of verycommon materials such as gray iron, ductile iron, carbon steel,alloy steel, and stainless steel, in cast, forging, or bar stock form.

Some compressor pistons and covers may be made of aluminum.For corrosive applications it is common to see stainless steel,typically 17-4PH or a 400 series, used for piston rods andcompressor valve seats and guards.

• Centrifugal—Materials for major components such as casings,nozzles, shafts, impellers, etc., are primarily carbon, alloy, and/orstainless steels. Components may be cast, forged, or fabricated.Some cast or nodular iron may be used for stationary components.

Material selection is primarily dependent on temperature, stress(pressure, torque), and gas composition (corrosive, erosive, etc).

Multiservice Capability

• Reciprocating—It is very easy to have a multiservice reciprocatingcompressor. The number of different services on a given


compressor crankcase (frame) is only limited by the number ofthrows available and the number of stages required for eachservice. Eight, 10, and even 12 throw frames are not uncommon.

• Centrifugal—While it is possible to have a multisectioncentrifugal compressor with different services/gases in eachsection, this is not typical.

Efficiency

• Reciprocating—Reciprocating compressors have a verycharacteristic adiabatic efficiency curve (Figure 8). As compressionratio drops, adiabatic efficiency drops. Efficiency changes withmolecular weight. Efficiency will also vary with several otherfactors, most significantly the compressor cylinder’s ratio of valveflow area to main bore diameter and piston speed.

Figure 8. Reciprocating Compressor Efficiency.

• Centrifugal—Polytropic efficiency is typically used for centrifugalcompressors rather than adiabatic. Adiabatic is commonly used forair compressors.

Typical polytropic efficiencies range from 70 percent to 85percent. Efficiencies approaching 90 percent are possible. In acentrifugal compressor, efficiency is primarily affected by theinternal leakage and mechanical losses.

Cost: Capital and Operating

• Reciprocating—Generally a reciprocating compressor will have alower capital cost but a higher operating cost compared to a centrifugal.The lower capital cost is a direct result of the machine consisting ofsmaller parts that cost less and are easier to manufacture. Higheroperating cost results from the recip containing more wearingparts requiring more maintenance and downtime—most specificallycompressor valves, which do not exist in a centrifugal.

• Centrifugal—The capital cost of a centrifugal compressor istypically higher than a reciprocating compressor operating at thesame conditions. This is primarily due to the fact that centrifugalcompressors require parts with more complex geometry such asimpellers and diaphragms. However, a centrifugal compressor hasfewer wearing parts, resulting in lower operating costs in terms ofreplacement parts, repairs, and downtime.

Minimum/Maximum Power

• Reciprocating—Reciprocating compressors vary in size from thevery small, under 10 hp (7.5 kW), to very large at 12,000 hp (9.0 MW).Even higher horsepower compressors are available from somemanufacturers for some very specialized applications, like very highpressure ethylene compression (“hyper” compressors).

• Centrifugal—Power developed is dependent on the mass flow ofthe gas compressed, the head required, and the efficiency. Thepower required to drive a centrifugal compressor can be as low as100 hp (75 kW) and as high as 130,000 hp (97 MW) or more.

Lead Time

• Reciprocating—Today, reciprocating compressor lead time isquite long due to the increased demand from the natural gasindustry. Lead time for a bare compressor will vary from 14 to 40weeks depending on size and manufacturer. Quite often, electricmotor driven compressor lead time is driven by the lead time of themotor—again depending on size (horsepower).

• Centrifugal—Typical lead times for a centrifugal compressortrain are in the range of 35 to 75 weeks. Lead time is mostsignificantly affected by the original equipment manufacturer(OEM)/subvendor shop loading, availability of any special materialsrequired (low temperature, corrosion resistant, etc.), special/uniquedesign requirements, and testing/inspection requirements.

Installation Time and Complexity

• Reciprocating—Installation time for a reciprocating compressorvaries significantly with size and location, and whether or not thecompressor is packaged. Packaged compressors are common todayup to 5000 hp (3.4 MW) of a high speed short stroke design.Installation time for these might vary from a few days to a coupleof weeks. Larger slow speed long stroke compressors assembled atsite might require three to four weeks to install.

• Centrifugal—Similar to a reciprocating compressor, the installationtime varies widely depending on the size of the compressor. Thenumber of main casing nozzle connections and the type of driveralso affect installation time.

The location can be a factor as well. Remote or offshorelocations can add to the installation time.

The compressor and driver are typically packaged on a base platecomplete with oil piping and wiring to junction boxes. Processequipment such as coolers and scrubbers and process control valvesare typically installed at site. Auxiliary systems such as lube oilconsoles, control panels, and seal buffer systems may also be installedseparately. Piping and wiring from these auxiliary systems and processequipment to the compressor train are typically done at site.

Installation time for a typical motor/gear driven compressor packageis two to three weeks. For a very large compressor or a gas turbinedriven compressor the installation time could be three to six weeks.

CASE STUDIES

APPENDIX A contains tables where the performance data forreciprocating and centrifugal compressors are compared for fourdifferent sets of pressure conditions, each for three different moleweight gases, for a total of 12 case study points. The purpose ofthese case studies is to compare performance and to help explainwhy a certain design compressor fits a certain application betterthan the other—from a performance perspective. Figure 9 is a chartshowing discharge pressure versus inlet flow intending to compareand contrast where each type of compressor fits best in this map.The case study points have been selected from this map.

Figure 9. NGPSA Compressor Coverage Chart. (Courtesy, NGPSAEngineering Data Book, 1994)


117

APPENDIX A

Table A-1. Case Study Performance Data, A.


Case Study Number 1A – Low Flow, Low Pressure, Medium Molecular Weight

Capacity = 600 acfm (1019.4 m3/hr) Suction Pressure = 5 psig (0.34 BarG) Suction Temperature = 80o F (27o C) Discharge Pressure = 45 psig (3.1 BarG) Molecular Weight = 18.83 (S.G. = 0.65) Reciprocating Centrifugal

Power, hp (kW) 79 (59) 150 (112)

Flow, scfm (Nm3/hr) 774 (1,244) 774 (1,244)

Power per flow, hp/scfm (kW/Nm3/hr) 0.102 (0.047) 0.193 (0.090)

Number of Stages 1 7

Equivalent Head, ft (m) 54,869 (16724) 59,230 (18,053)

Case Study Number 1B – Low Flow, Low Pressure, Low Molecular Weight

Capacity = 600 acfm (1019.4 m3/hr) Suction Pressure = 5 psig (0.34 BarG) Suction Temperature = 80o F (27o C) Discharge Pressure = 45 psig (3.1 BarG) Molecular Weight = 2.02 (S.G. = 0.07) Reciprocating Centrifugal

Power, hp (kW) 78 (58)

Flow, scfm (Nm3/hr) 773 (1243)

Power per flow, hp/scfm (kW/Nm3/hr) 0.101 (0.047)

Number of Stages 2

Equivalent Head, ft (m) 540,746 (164,819)

This case is too small for a centrifugal compressor.

Case Study Number 1C – Low Flow, Low Pressure, High Molecular Weight

Capacity = 600 acfm (1019.4 m3/hr) Suction Pressure = 5 psig (0.34 BarG) Suction Temperature = 80o F (27o C) Discharge Pressure = 45 psig (3.1 BarG) Molecular Weight = 28.05 Reciprocating Centrifugal

Power, hp (kW) 85 (63) 151 (113)

Flow, scfm (Nm3/hr) 774 (1,245) 774 (1,245)



Equivalent Head, ft (m) 36,702 (11,187) 38,578 (11,759)

Case Study Number 2A, High Flow, Medium Pressure, Medium Molecular Weight

Capacity = 10,000 acfm (16,990.1 m3/hr) Suction Pressure = 200 psig (13.8 BarG) Suction Temperature = 80o F (27o C) Discharge Pressure = 1000 psig (68.95 BarG) Molecular Weight = 18.83 Reciprocating Centrifugal

Table A-2. Case Study Performance Data, B.


119

Power, hp (kW) 18,909 (14,100) 21850 (16.306)

Flow, scfm (Nm3/hr) 145,830 (234,522) 145,830 (234,522)




Case Study Number 2B, High Flow, Medium Pressure, Low Molecular Weight


Power, hp (kW) 14,295 (10,660)

Flow, scfm (Nm3/hr) 145,830 (225,430)


Number of Stages 2

Equivalent Head, ft (m) 813,384

This case can not be accomplished with a reasonable

number of centrifugal compressor bodies/stages.

Case Study Number 2C, High Flow, Medium Pressure, High Molecular Weight


Power, hp (kW) 21,102 (10,660) 20,960 (15,642)

Flow, scfm (Nm3/hr) 154,014 (225,430) 154,014 (225,430)




Case Study Number 3A, Medium Flow, Medium Pressure, Medium Molecular Weight

Capacity = 3,000 acfm (5,097 Nm3/hr) Suction Pressure = 200 psig (13.79 BarG) Suction Temperature = 80o F (27o C) Discharge Pressure = 600 psig (41.37 BarG) Molecular Weight = 18.83 Reciprocating Centrifugal

Power, hp (kW) 3,945 (2,948) 4,252 (3,173)

Flow, scfm (Nm3/hr) 43,867 (70,537) 43,867 (70,537)

HP per flow, hp/scfm (kW/Nm3/hr) 0.090 (0.042) 0.097 (0.045)



Table A-3 .Case Study Performance Data, C.


Case Study Number 3B, Medium Flow, Medium Pressure, Low Molecular Weight

Capacity = 3000 acfm (5,097 Nm3/hr) Suction Pressure = 200 psig (13.8 BarG) Suction Temperature = 80o F (27o C) Discharge Pressure = 1000 psig (69 BarG) Molecular Weight = 2.02 Reciprocating Centrifugal

Power, hp (kW) 3,775 (2,815)

Flow, scfm (Nm3/hr) 41,861 (67,610)


Number of Stages 2

Equivalent Head, ft (m) 511,799 (155,996)



Case Study Number 3C, Medium Flow, Medium Pressure, High Molecular Weight

Capacity = 3000 acfm (5,097 Nm3/hr) Suction Pressure – 200 psig (13.8 BarG) Suction Temperature = 80o F (27o C) Discharge Pressure = 1000 psig (69 BarG) Molecular Weight = 28.05 Reciprocating Centrifugal

Power, hp (kW) 3,927 (2,928) 4,062 (3,031)

Flow, scfm (Nm3/hr) 45,477 (73,125) 45,477 (73,125)

HP per flow, hp/scfm (kW/Nm3/hr) 0.086 (0.040) 0.089 (0.041)



Case Study Number 4A, Medium Flow, High Pressure, Medium Molecular Weight

Capacity = 2000 acfm (3,398 Nm3/hr) Suction Pressure = 200 psig (13.8 BarG) Suction Temperature = 80o F (27o C) Discharge Pressure = 3500 psig (241.4 BarG) Molecular Weight = 18.83 Reciprocating Centrifugal

Power, hp (kW) 7,012 (5,229) 9,732 (7,263)

Flow, scfm (Nm3/hr) 29,027 (46,675) 29,027 (46,675)


Number of Stages 3 16 (2 bodies)


Case Study Number 4B, Medium Flow, High Pressure, Low Molecular Weight


Table A-4. Case Study Performance Data, D.


121

Power, hp (kW) 6,844 (5,134)

Flow, scfm (Nm3/hr) 27,908 (44,880)


Number of Stages 4

Equivalent Head, ft (m) 1,788,446



Case Study Number 4C, Medium Flow, High Pressure, High Molecular Weight


Power, hp (kW) 5,890 (4,392) 7,811 (5,829)

Flow, scfm (Nm3/hr) 30,318 (48,750) 30,318 (48,750)




REFERENCES

API Standard 617, 2002, “Axial and Centrifugal Compressors andExpander-Compressors for Petroleum, Chemical and GasIndustry Services,” Seventh Edition, American PetroleumInstitute, Washington, D.C.

API Standard 618, 1995, “Reciprocating Compressors forPetroleum, Chemical, and Gas Industry Services,” FourthEdition, American Petroleum Institute, Washington, D.C.

NGPSA Engineering Data Book, 1994, 1, Revised 10thEdition, Compiled and Edited in Cooperation with the GasProcessors Association, Copyright 1987 Gas ProcessorsAssociation.

BIBLIOGRAPHY

Basic Thermodynamics of Reciprocating Compression, ShortCourse Presented at the 2005 Gas Machinery Conference,October 2005.

Brown, R. N. ,1986, Compressors—Selection and Sizing, Houston,Texas: Gulf Publishing Company.

Paluselli, D. A, Basic Aerodynamics of Centrifugal Compressor,Elliott Company, Jeannette, Pennsylvania.

ACKNOWLEDGEMENT

The authors wish to acknowledge Mr. John F. Vanderhoff,formerly senior application engineer with Elliott Company andpresently with Westinghouse in Monroeville, Pennsylvania, whocontributed a significant amount of effort to this tutorial whileemployed at Elliott Company.


Robert C. White is a Principal Engineerfor Solar Turbines Inc., in San Diego,California. He is responsible for compressorand gas turbine performance predictionsand application studies. In his formerposition he lead the development ofadvanced surge avoidance and compressorcontrols at Solar Turbines. Mr. White holds12 U.S. Patents for turbomachinery relateddevelopments. He has contributed to

several papers, tutorials, and publications in the field ofTurbomachinery.

Rainer Kurz is Manager of SystemsAnalysis and Field Testing for SolarTurbines Inc., in San Diego, California. Hisorganization is responsible for conductingapplication studies, gas compressor andgas turbine performance predictions, andsite performance testing. He joined SolarTurbines in 1993, and has authored morethan 30 publications in the field ofturbomachinery and fluid dynamics.

Dr. Kurz attended the University of the German Armed Forces,in Hamburg, Germany, where he received the degree of aDipl.-Ing., and, in 1991, the degree of a Dr.-Ing. He was elected asan ASME Fellow in 2003.

ABSTRACT

This tutorial explains the surge phenomenon as well as the pre-cursors of surge and the damage that surge can cause to theequipment. Then, methods to keep the compressor system fromsurge, i.e., modern surge control systems, are discussed. Thisincludes the necessary instruments, algorithms used, as well as thepiping layout required. Sizing considerations for valves and othersystem components, in particular methods to correctly estimateallowable upstream pipe volumes, are described. Surge controlsystems for single body and multibody or multisection compres-sors will be explained, and attention will be given to the integrationof the surge control system with other aspects of the station andunit control. In a look forward, new methods for surge detectionand detection of system changes prior to surge are covered.

INTRODUCTION

Operation of centrifugal gas compressors can be defined by threeoperating parameters: speed, head, and flow. Centrifugal compressorshave a maximum head that can be achieved at a given speed. At thatpeak head there is a corresponding flow. This is a stability limit.Operation of the compressor is stable provided the head is lower (lessresistance in series with the compressor) and the flow is greater thanthese values. That is, the system is stable, as long as reductions in headresult in increases in flow. Surge occurs when the peak head capabilityof a compressor is reached and flow is further reduced. Depending onthe dynamic behavior of the compression system, system surge canoccur at somewhat higher or, seldom, lower flows than the peak headcapability. This is a particular issue in systems with low frequencypulsations (Kurz, et al., 2006). When the compressor can no longermeet the head imposed by the suction and discharge condition (whichare imposed by the compression system), flow reverses.

When a compressor approaches its surge limit, some of itscomponents (diffusers, impeller) may start to operate in stall. Stalloccurs when the gas flow starts to separate from a flow surface(Figure 1). Changing the operating point of a compressor alwaysinvolves a change in incidence angles for the aerodynamiccomponents. Just as with an airfoil (Figure 1), increasing the incidenceangle will eventually lead to stall. Stall in turbomachines often appearsas rotating stall, when localized regions of separated flow move alongthe diffuser at speeds below the rotational speed of the impeller (Day,1991). Surge is the ultimate result of system instability.

Figure 1. Progression of Stall. (NACA4412 airfoil at increasingangle of attack, based on data by Nakayama, 1988)

123

SURGE AVOIDANCE FOR COMPRESSOR SYSTEMS

byRobert C. WhitePrincipal Engineer

andRainer Kurz

Manager, Systems Analysis

Solar Turbines Inc.

San Diego, California

Attempting to increase the resistance in series with the compressorbeyond that which drives the compressor to peak head will resultin unstable operation. Flow will decrease and subsequently thehead capability of the compressor will also decrease. As the headcapability decreases, flow further decreases. Once the compressor canno longer meet the external head, flow reverses.

Surge is what happens after the stability limit of the compressionsystem is passed.

Not only is this detrimental to meeting the process objectives,the resulting axial and radial movement of the rotor can causedamage, sometimes severe, to the compressor. Surge can beavoided by ensuring the flow through the compressor is notreduced below the flow at peak head.

The surge avoidance system prevents surge by modulating a surgecontrol (bypass) valve around the compressor. A typical systemconsists of pressure and temperature transmitters on the compressorsuction and discharge lines, a flow differential pressure transmitteracross the compressor flowmeter, an algorithm in the control system,and a surge control valve with corresponding accessories.

A surge avoidance system determines the compressor operatingpoint using the pressure, temperature, and flow data provided bythe instrumentation. The system compares the compressoroperating point to the compressor’s surge limit. The differencebetween the operating point and the surge limit is the control error.A control algorithm (P1I1D) acts upon this difference, or “error,”to develop a control signal to the recycle valve. When opened, aportion of the gas from the discharge side of the compressor isrouted back to the suction side, and head across the compressor isprevented from increasing further. When the operating pointreflects more flow than the required protection margin flow, thesurge control valve moves toward the closed position and thecompressor resumes normal operation.

There are five essentials for successful surge avoidance:

1. A precise surge limit model—It must predict the surge limitover the applicable range of gas conditions and characteristics.

2. An appropriate control algorithm—It must ensure surgeavoidance without unnecessarily upsetting the process.

3. The right instrumentation—Instruments must be selected tomeet the requirements for speed, range, and accuracy.

4. Recycle valve correctly selected for the compressor—Valvesmust fit the compressor. They must be capable of large and rapid,as well small and slow, changes in capacity.

5. Recycle valve correctly selected for the system volumes—Thevalve must be fast enough and large enough to ensure the surgelimit is not reached during a shutdown. The piping system is thedominant factor in the overall system response. It must be analyzedand understood. Large volumes will preclude the implementationof a single valve surge avoidance system.

BACKGROUND

The first turbocompressors were manufactured at the turn ofthe 1900s. They were originally developed by steam turbinemanufacturers and were widely used for ventilation purposes indeep shaft mining, in particular the coal industry. At that time,the method of producing an impeller relied upon fabrication. Itwould be decades before technology would allow highlyefficient turbocompressors to be made. It was not until late inWorld War II that sufficient money was pumped into technologyto allow the production of efficient high-speed compressors. In1947 to 1948, Ingersoll-Rand and Clark designed the firstcentrifugal compressors for natural gas transmission. InSeptember 1952, El Paso Natural Gas became the first companyto use large gas-turbine-driven centrifugal compressors fornatural gas transmission.

The pressure ratio across an early riveted impeller would haveonly been on the order of 1.2:1. This would have meant that to

reach a final working pressure of 7 barg, a turbocompressor neededas many as 10 or 11 stages. A single modern impeller can producea pressure ratio as high as 8:1.

As the performance of compressors increased, so did thepotential for damage from surge. Not only the damage tocompressors needed to be avoided but avoiding upsets to theprocess became essential. To address these ever growing needs,surge avoidance systems evolved.

Early surge controls were pneumatic. They monitored thepressure differential (DP) across the compressor versus the DPacross a flow measuring element through a balance beam thatcontrolled the pneumatic signal to a recycle valve. With higherperformance, higher stressed compressors, and more challengingapplications, better surge controls were required.

Initially electronic surge controls were models of thepneumatic ones. They were faster, less complicated, morereliable, and required less maintenance than their pneumaticpredecessors. With the advent of the microprocessor, surgeavoidance systems became more algorithmically intense, surgelimits were modeled via polynomials, and asymmetrical controlschemes came into use. Modern surge controls ensure surgeavoidance while maximizing the operating range ofthe compressor.

THE SURGE LIMIT MODEL

In order to avoid surge it must be known where thecompressor will surge. The more accurately this is predicted, themore of the compressor’s operating range that is available to theuser (Figure 2). A compressor’s operation is defined by threeparameters: head, flow, and speed. The relationship between thecompressor’s operating point and surge can be defined by anytwo of the three.

Figure 2. Surge Avoidance System Schematic.

The first two models on the left of Figure 3 involve speed. Thespeed of the compressor at an operating condition is stronglyinfluenced by changes in gas composition, because the machineMach number will change. The head versus flow relationship onthe right provides a means for modeling the surge limit withoutbeing affected by gas conditions or characteristics. Theparameters of the surge limit model on the right can bemeasured in terms of head across the compressor and head acrossthe flowmeter.


H K T SG Zp

PP

D

S= ⋅

−

⋅ ⋅ ⋅

σ

σ

1

Figure 3. Surge Limit in Different Systems of Reference.

This is the basic equation for head. In the head across thecompressor and the head across the flowmeter monitoring the flowthrough the compressor, there are common terms. These commonterms (units, gas temperature, specific gravity, and compressibility) areequal in both equations and can be cancelled. This results in asimplified model that is referred to as reduced head versusreduced flow.

THE CONTROL ALGORITHM

A surge avoidance control needs to be able to react appropriatelyto changes in power or the process. There are two very differentsituations that the system must respond to.

If the operating point slowly crosses the protection line, that is,at the same rate it has been moving left for the past several hours,opening of the recycle valve should be small and slow. Theinterdiction of the surge avoidance control should be unnoticeable.It should be as though the compressor had infinite turndown.

Conversely, if the operating point races across the compressormap, the recycle valve should begin opening before the operatingpoint crosses the protection line. Reaction of the control should beaggressive to protect the compressor. In this case we are lessconcerned about the process, as it has already been impacted.

A sudden change in the system produces a control response.This is a standard control test. Ping it and see how it rings. Figure4 reflects reactions of variously tuned controls. Low gains producea slow response. A critically damped control produces anaggressive response but settles down quickly. If the gains are toohigh the system will oscillate.

Figure 4. Reaction of a Control System to an Error Signal.

What does a surge avoidance system do most of the time?Hopefully nothing! Then, with very little margin it must actaggressively, probably requiring gains higher than could bemaintained stable, to protect the compressor. To avoid instabilitythe gains are reduced to close the valve. Once surge has beenavoided, the control system should bring the process back onlineslowly and smoothly to avoid further upsets.

The need for extremely high gains is driven by the following:surge avoidance systems are normally built up out of commonlyavailable process control plant components. As such thesecomponents are designed for ruggedness, reliability, and lowmaintenance. In general they are not focused on speed of dataacquisition. Information about changing process conditions isoften 1/10 of a second old. As will be seen in later sectionssignificant advances in surge control valve action have been maderecently. However, the response of the valve is typically thedominant lag in the system.

INSTRUMENTATION

To avoid surge, the control needs to know where the compressoris operating in relation to surge in real time. Again, how close theprotection margin can be placed to surge depends on howaccurately and how quickly the change in flow is reported to thecontrol. Correctly selected instrumentation is essential. The systemmust have accurate measurements of the suction and dischargepressures and temperatures, and the rate of flow. Flow is the mostimportant parameter as it will move the fastest and farthest as thesurge limit is approached. Ideally, the flow transmitter should be anorder of magnitude faster than the process. Unfortunately, comparedto pressure and temperature transmitters, flow transmitters tend to beslow. Even the best surge avoidance control will allow a compressorto surge, if it is connected to a slow transmitter.

Flow-Measuring Devices

Most commercially available flow-measuring devices areaccurate enough for surge avoidance; however, it is the transmitterthat slows things down. A differential pressure transmitter’sresponse time is inversely proportional to its range; thus, thestronger the signal, the faster the response.

Devices that develop high DP signals are desirable. Those withlow signal levels tend to have low signal-to-noise ratios. Transmittersfor low DP signal ranges typically have slow response times.Devices that create an abrupt restriction or expansion to the gas, suchas orifices, cause turbulence and, subsequently, create noise.

It is preferable to place the flow-measuring device on thesuction side of the compressor. Typically, variations in pressures,temperatures, and turbulence of the gas are less upstream of thecompressor. Also, the device must be inside the innermost recycleloop (refer to Figure 2).

At a minimum, failure of the device will cause the compressorset to be shut down until the device can be replaced. If the failureresults in pieces being ingested by the compressor, it can causean expensive overhaul. For this reason, devices that arecantilevered into the gas stream are not recommended. Low costflow-measuring devices do not always result in cost savings in thelong run.

Low permanent pressure loss (PPL) devices are oftenrecommended; however, their benefits may be marginal. The lostpower cost impact of operating a device can be calculated. Forexample, a flowmeter developing a 100 inch H2O signal and a 50percent PPL flowing 100 mmscfd (50 lb/sec) is equivalent toabout 20 hp.

As noted, strong signal devices are highly preferred. Pitot types(annubars and verabars) have a relatively low signal level, around25 inches H2O. In the middle are orifices and venturis with amoderate signal of around 100 inches H2O. Compressor suction-to-eye provides a strong signal (around 700 inches H2O) with theadded benefit of not causing any additional pressure loss.

SURGE AVOIDANCE FOR COMPRESSOR SYSTEMS 125

N H

Q

H

N2 Q2

H = Head Q = Flow N = Speed

Surge

Margin

Head Rise

to SurgeTurndown

F

Error Signal

Gain Too

Low

Critical

Damping

Gain Too

High

High and

Low Gains

Too Slow

Optimum

Too Fast - Unstable

Smooth Response

F

HREDUCED

PP

D

S=

−

σ

σ

1

Suction-to-Eye Flow Measuring

Suction-to-eye uses the inlet shroud or inlet volute of thecompressor as a flow-measuring device. This feature is now availableon many compressors. The design requirements for the inlet voluteand the flow measuring device have several things in common.Performance of the first stage impeller and the device is dependenton the uniform direction and velocity of the flow presented to it.

Critical to the operation of suction-to-eye flow measurement is theplacement of the eye port. As the impeller approaches surge an area ofrecirculation begins to develop at the outer perimeter of the inlet to theimpeller. If the eye port is placed too close to the impeller’s outerperimeter the relationship of the DP to flow will be affected.Fortunately the meter factor (C’) typically remains nearly the same forthe same surge margin. Hence, selecting the meter factor at the desiredsurge protection margin will contribute to effective surge avoidance.

In a typical pipeline application (600 psi suction pressure) suction-to-eye will develop 25 psid (692 inches H2O). This is nearly seventimes the differential of an orifice plate. Typically the signal to noiseratio is low and there is no additional permanent pressure loss. Forsurge avoidance the suction-to-eye method is strongly recommended.

Compressor Instrumentation

Optimal performance of any control system is dependent on thespeed, accuracy, and resolution of the instrumented processconditions. To achieve optimal performance the instruments shouldhave performance specifications an order of magnitude better thanthe requirements for the system. Typical gas compressor systemshave a first-time constant of about one second; hence, noinstrument should have a first-time constant of greater than 100ms. The surge control system is expected to discriminate betweensingle digit percentages of surge margin; hence, measurement ofthe process parameters should be accurate to 0.1 percent. The finalcontrol elements (recycle valves) probably can resolve 1 percentchanges in their command signals; hence, the process variablesshould be resolved to at least 0.1 percent (10 bits) of their normaloperating range. Over-ranging transmitters degrade resolution.

THE SURGE CONTROL VALVE

Characteristics

Earlier it was discussed how the control should react differentlyto gradual and rapid approaches to surge. Likewise, the valve mustaddress these two very different requirements. For the gradualapproach, it should behave like a small valve and produce smooththrottling. For the rapid approach case, it should act like a large fastvalve to handle sudden major changes.

There are three general valve characteristics (Figure 5): quickopening, where most of the valve’s capacity is reached early in itstravel; linear, where capacity is equal to travel; and equal percentage,where most of the capacity is made available toward the end of thevalve’s travel. All three types of valve have been used in variousconfigurations as recycle valves.

Figure 5. Valve Types.

Equal percentage valves, and in particular noise-attenuating ballvalves, are recommended for surge avoidance systems with asingle surge control valve. They perform like smaller valves whennearly closed and bigger valves when close to fully open. Figure 6is a comparison of two types of equal percentage valve. For a givenvalve size, the noise-attenuating ball valve is often twice the costof the globe valve, but it provides approximately three times theflow coefficient (Cv) or capacity. Also, it is more reliable as it isless susceptible to fouling and improper maintenance.

Figure 6. Ball and Globe Valves Compared.

Employing a valve with an equal percentage characteristic mayprovide the capacity needed to avoid surge during a shutdownwhile maintaining enough resolution at less than 50 percentcapacity to provide good control at partial recycle. With an equalpercentage characteristic the valve typically has greater resolutionthan a single linear valve selected to fit the compressor.

Multiple Valves

If the volumes on either side of the compressor are large, amultiple valve approach may be needed. If an integrated approachis used, the total valve capacity will be reduced.

Probably the most common is the hot and cold recycleconfiguration (Figure 7). Usually the cooled (outer) valve ismodulating and the hot (inner) valve is a quick opening on-offtype. Generally the two valves are sized independently. If thecooled valve has a solenoid, its capacity can be considered withthat of the shutdown valve; subsequently the shutdown valve canbe smaller.

Figure 7. Hot and Cold Recycle Valve Arrangement.

An alternate to this configuration is having a second cooledvalve in parallel with the first (Figure 8). This arrangementprovides some measure of redundancy. In the control the twovalves are operated in cascade. That is, they have different setpoints, say 9 percent and 10 percent surge margin. Under normalmovements of the operating conditions only the 10 percent surgemargin valve (primary valve) will open. If movement is fastenough to push the operating point down to 9 percent, the secondvalve (secondary valve) will open. If the primary valve becomesfouled and no longer positions properly, the control can place it in


the secondary position and the secondary becomes the primaryvalve. This change can be made without taking the compressoroffline.

Figure 8. Parallel Recycle Valves.

The advantages of the two parallel valves do not come without aprice. In normal operation 2 percent to 5 percent of the pressurerise across the compressor will be lost across the cooler. In theshutdown scenario the required flow through the cooler to avoidsurge may be two or three times the normal flow. This will resultin four to nine times the pressure drop across the cooler. Thisadditional pressure drop may increase the needed recycle valvecapacity significantly.

Recycle valves need to be fast and accurately positionable. Theyalso need to be properly sized for both the compressor and thepiping system. A valve well suited for modulating recycle aroundthe compressor may not be suitable for a shutdown. (Refer to the“REVIEW OF SYSTEM VOLUMES” section below.)

For some two-valve applications, single purpose valves may besuitable, one for controlled recycling and one for shutdown. Alinear characteristic valve is appropriate for the controlledrecycling and a quick opening characteristic globe or ball valvefor shutdown.

For the applications where the compressor speed lines arefairly flat (little increase in head for a decrease in flow) from thedesign conditions to surge, extra fast depressurization may berequired. To achieve this, two quick opening valves may beemployed. In this case a single 6 inch linear characteristic valveis replaced by two 4 inch quick opening valves. The two 4 inchvalves should have slightly less flow capacity (Cv) but they willopen nearly 45 milliseconds faster. For linear valves 50 percenttravel equals 50 percent capacity. For quick opening valves,capacity approximately equals the square root of travel. As suchthe two 4 inch valves will have 70.7 percent of their fully opencapacity at 50 percent open. Comparing the two arrangements250 ms after the shutdown is initiated, the two 4 inch quickopening valves will have 56 percent more flow capacity than thesingle 6 inch linear valve.

For throttling, the valves are operated in cascade or split range.For most controlled recycling only one valve is opened. Althoughthe valves have a quick opening characteristic the valves aresmaller thus the capacity per percent travel is less. The two quickopening valves operated in cascade or split range will have thesame Cv as the 6 inch linear at 25 percent travel.

Valve Actuation

As previously discussed, there are two operational scenariosfor the surge avoidance system: modulating (minimum flowcontrol) and rapid depressurization for shutdown. By inserting athree-way solenoid valve into the positioner’s output, the valvecan be made to open with either a proportional (4-20 mA) signalfor modulating control, or a discrete (24 VDC) signal for totalfast opening.

The primary difference between a surge control valve and astandard control valve is in its actuation system. The preferredactuator for surge avoidance is spring return, fail open. Thisdesign is simple, reliable, and ensures the compressor isprotected in the event of a power failure. Both spring and

diaphragm and spring and piston actuators are used. Thespring and diaphragm actuator is most commonly used onglobe valves. The spring and piston actuator is more oftenused on ball valves. The more powerful spring and pistonactuators are required on rotary valves due to the greaterforces required to accelerate the mass of the ball. Some ballvalves are not suitable for surge control applications becausetheir shafts and attachments to the ball are not strong enoughto transmit the torque required to open these valves at therequired speeds.

Surge control valves need to be able to open very quickly. Assuch their actuators will have strong springs, very large airpassages, and shock absorbers at their end of travel. This mustbe considered when sourcing recycle valves for surgeavoidance.

The accessory unique to a sound surge control valve assembly isthe single sided booster or exhaust booster. This is essentially adifferential pressure relief device. Opening the booster vents theactuator pressure to atmosphere. The threshold for opening is about0.5 psid. There is a small restriction (needle valve) between thecontrol pressure from the positioner via the three-way solenoidvalve and the top of the booster. Small slow reductions in pressure(opening the valve) do not cause the booster to open. Large fastreductions in pressure developing more than 0.5 psid across therestriction, cause the booster to open. If the solenoid valve isde-energized, the top of the booster is vented to atmosphere and thebooster fully opens.

Standard industry quick-exhausts are not recommended for thisapplication. They have a high threshold for opening (typically 2 to4 psid) and an equally high threshold for reclosing. Although theymay work well for fully opening the valve, they will not work wellwith the positioner.

Positioners should be selected for high capacity and quickresponse to changes in their control signals. Most of the majorvalve manufacturers have released second and third generationsmart positioners that are suitably fast for this application.

Figures 9 and 10 show globe and ball valves with their preferredinstrumentation configurations.

Figure 9. Globe Valve Assembly.


4 - 20mA

Limit Switch Closed

Limit Switch Open

24VDC

Exhaust

Booster

Needle Valve &

Check Valve

Three-way 24VDC

Solenoid Valve

Position

Transmitter

4 - 20 mA

Pressure Regulator

(Airset 35 - 50 psig)

Instrument

Air Supply

Electro- pneumatic

Positioner

Yoke Mounted

4 - 20 mA

6 - 30 psig4 - 20mA

F

Figure 10. Ball Valve Assembly.

Recycle Valve Sizing Tool

The recycle valve needs to be sized based on the expected operatingconditions of the compressor. A valve-sizing program can facilitatematching a recycle valve to a compressor. The compressor data areentered into the tool in its normal form (pressures, temperatures, heads,speeds, and flows). Various operating conditions for a specificapplication are then entered, such as the minimum and maximumoperating speeds, pipe operating pressures, temperatures, relief valvesettings, and cooler data, if applicable. The tool calculates theequivalent valve capacities or Cvs from that data.

Typically the surge limit of a compressor equates to a singlevalve capacity or Cv (Figure 11). The valve can be selected basedon valve Cg, Cv, and Xt tables from surge control valve suppliers.As previously described, a single surge control valve applicationwill have an equal percentage characteristic. Once a valve isselected several performance lines of a specific opening can bedeveloped and overlaid on the compressor map. The equalpercentage characteristic valve should be at about two-thirds travelat the surge conditions. The valve evaluation in Figure 12 shows sucha valve with its flow characteristic when 60 percent, 70 percent, and100 percent open, superimposed on the compressor map.

Figure 11. Almost Constant Cv at the Surge Limit.

Figure 12. Valve Matched to Compressor.

REVIEW OF SYSTEM VOLUMES

Design of the piping and valves, together with the selection andthe placement of instruments, will significantly affect theperformance of an antisurge control system. This should beaddressed during the planning stage of a project because thecorrection of design flaws can be very costly once the equipmentis installed and in operation.

As described above, the control system monitors the compressoroperating parameters, compares them to the surge limit, and opensthe recycle valve as necessary to maintain the flow through thecompressor at a desired margin from surge. In the event of anemergency shutdown or ESD, where the fuel to the gas turbine isshut off instantly, the surge valve opens immediately, essentially atthe same time the fuel valve is closing.

The worst case scenario for a surge control system is anemergency shutdown (ESD), particularly if the compressor isalready operating close to surge when the engine shutdown occurs.If an ESD is initiated, the fuel supply is shut off immediately andthe compressor will decelerate rapidly under the influence of thefluid forces counteracted by the inertia of the rotor system. Figure13, which displays data based on test data and theoreticalconsiderations, indicates a 30 percent drop on compressor speedwithin the first second after shutdown. A 30 percent loss in speedequates to approximately a loss in head of about 50 percent. Thevalve must, therefore, reduce the head across the compressor byabout half in the same time as the compressor loses 30 percent ofits speed.

Figure 13. Simplified System and Transient Characteristic.

The larger the volumes are in the system, the longer it will taketo equalize the pressures. Obviously, the larger the valve, the betterits potential to avoid surge. However, the larger the valve, thepoorer its controllability at partial recycle. The faster the valve canbe opened, the more flow can pass through it. There are, however,limits to the valve opening speed, dictated by the need to controlintermediate positions of the valve, as well as by practical limits tothe power of the actuator. The situation may be improved by usinga valve that is only boosted to open, thus combining high openingspeed for surge avoidance with the capability to avoid oscillationsby slow closing.


4 - 20mA

Closed Limit Switch

Open Limit Switch

4 - 20 mA

80 - 100 psig

24 VDC

Electro-pneumatic

Positioner

Yoke Mounted

4 - 20 mA

6 - 30 psig

Position

Transmitter

4 - 20 mA

Three-way

24VDC

Solenoid Valve

Exhaust

Booster

Needle Valve &

Check Valve

If the discharge volume is too large and the recycle valve cannotbe designed to avoid surge, a short recycle loop (hot recycle valve)may be considered (Figure 7), where the recycle loop does notinclude the aftercooler.

While the behavior of the piping system can be predictedquite accurately, the question about the rate of deceleration forthe compressor remains. It is possible to calculate the powerconsumption for a number of potential steady-state operatingpoints. The operating points are imposed by the pressure in thedischarge volume, which dictates the head of the compressor. Fora given speed, this determines the flow that the compressor feedsinto the discharge.

In a simple system, the boundaries for the gas volume on thedischarge side are established by the discharge check valve,compressor, and recycle valve (Figure 2). The volume on thesuction side is usually orders of magnitude larger than thedischarge volume and, therefore, can be considered infinite. Thus,to simplify the analysis of a system, the suction pressure can beconsidered constant. This is not a general rule, but is used tosimplify the following considerations. This yields the simplifiedsystem, consisting of a volume filled by a compressor and emptiedthrough a valve (Figure 14).

Figure 14. Train Deceleration and Valve Opening.

The basic dynamic behavior of the system is that of a fixedvolume where the flow through the valve is a function of thepressure differential over the valve. In a surge avoidance system, acertain amount of the valve’s flow capacity will be consumed torecycle the flow through the compressor. Only the remainingcapacity is available for depressurizing the discharge volume. Insuch a system, mass and momentum balance have to be maintained(Sentz, 1980; Kurz and White, 2004). From this complete model,some simplifications can be derived, based on the type of questionsthat need to be answered.

For relatively short pipes, with limited volume (such as thesystems desired for recycle lines), the pressure at the valve and thepressure at compressor discharge will not be considerablydifferent. Also, due to the short duration of any event, the heattransfer can also be neglected. Therefore, mass and momentumconservation are reduced to:

The rate of flow through the valve is calculated with the standardISA method (ANSI/ISA, 1995) (Qstd is the standard flow. Fp is thepiping geometry factor. It is usually not known and can be assumedto be 1. The pressure is assumed to be constant in the entire pipevolume. It is thus the same just upstream of the valve and at thedischarge pressure of the compressor):

and

The compressibility Z2 can be calculated with the Redlich-Kwongequation of state. Equations (3), (4), and (5) mean that thedischarge pressure change depends on the capability of the valve torelease flow at a higher rate than the flow coming from thecompressor. It also shows that the pressure reduction for a givenvalve will be slower for larger pipe volumes (V). Kurz and White(2004) have shown the validity of the simplified model.

The discharge pressure p2 in Equation (3) is a function of thecompressor operating point, expressed by:

Alternatively, a lookup table, showing the head-flow relationshipfor the compressor, can be used.

The behavior of the compressor during ESD is governed by twoeffects. The inertia of the system consisting of the compressor,coupling, and power turbine (and gearbox where applicable) iscounteracted by the torque (T) transferred into the fluid by thecompressor (mechanical losses are neglected). The balance offorces thus yields:

Knowing the inertia (J) of the system and measuring the speedvariation with time during rundown yields the torque and, thus, thepower transferred to the gas:

If the rundown would follow through similar operating points,then P~N3, which would lead to a rundown behavior of:

Regarding the proportionality factor (k) for power and speed, thisfactor is fairly constant, no matter where on the operating map therundown event starts. Thus, the rate of deceleration, which isapproximately determined by the inertia and the proportionalityfactor, is fairly independent of the operating point of the compressorwhen the shutdown occurred, i.e., the time constant (dN/dt [t = 0])for the rundown event is proportional to k/J. However, the higher thesurge margin is at the moment of the trip, the more head increase canbe achieved by the compressor at constant speed.

While the behavior of the piping system can thus be predictedquite accurately, the question about the rate of deceleration forthe compressor remains. It is possible to calculate the powerconsumption for a number of potential steady-state operatingpoints. The operating points are imposed by the pressure in thedischarge volume, which dictates the head of the compressor. Fora given speed, this determines the flow that the compressor feedsinto the discharge.


[ ]dp

dt

k p

VQ Qv

2 2=⋅

−

Q F c Yp p

p SG T Zstd p v= ⋅ ⋅ ⋅ ⋅−

⋅⋅ ⋅

1360

12 1

2 2 2

0 5.

( )Q Qp T Zv std

std=ρ

ρ 2 2 2, ,

( )p

p

k

k

h Q N SG

ZT

h

N

Q

N

Q

N

kk

2

1 12

2

11

287

1

= +−

⋅⋅

⋅

=

+ +−,

α β γ

T JdN

dt= − ⋅ ⋅2π

( )P T N J NdN

dt= ⋅ ⋅ = − ⋅ ⋅ ⋅2 2 2π π

( ) ( )( )

( )

dN

dt

k

J

NN dN

k

Jdt c N t

tk

J Nt

= → = + → =− −

−

=

∫∫2

22

2

2

12 2

1

20

π ππ

The model described above, which accounts for the primaryphysical features of the discharge system, can be used to determinewhether the combination of discharge volume and valve size canprevent the compressor from surge during an ESD. It allows thetwo important design parameters to be easily varied to avoid surgeduring ESD. The surge valve size and opening speed can beincreased for a given discharge volume or the maximum allowabledischarge volume for a given configuration of valves andcompressor characteristic can be limited. The second method,which has the advantage of being more transparent for the stationdesign, is used here.

The simplified model calculates the maximum discharge volumewhere the head across the compressor can be reduced by half inone second, based on the assumption that this reflects the speeddecay during an ESD as outlined above. Therefore, the calculationof the instant compressor speed is replaced by a fixed, presumed tobe known, deceleration rate. The assumption is made that thepower turbine and compressor will lose about 20 to 30 percentspeed in the first second of deceleration. This is, for example,confirmed by data from Kurz and White (2004) showing a 30percent speed reduction of a gas turbine driven compressor set,and, where the gas turbine driven configurations lost about 20 to 25percent speed in the first second, while the electric motor drivenconfiguration lost 30 percent speed in the first second. As a result ofthe loss of 25 percent speed, the head the compressor can produceat the surge line is about 56 percent lower than at the initial speed,if the fan law is applied. A further assumption is made about theoperating point to be the design point at the instant of the ESD.

Any ESD is initiated by the control system. Various delays in thesystem are caused by the time for the fuel valve to shut completely,the time until the hot pressurized gas supply to the power turbineseizes, and the opening time of the recycle valve. ESD data show itis a valid assumption that the surge control valve reaches full opensimultaneously with the beginning of deceleration of the powerturbine/compressor. This is the starting time (T0) for the model.

Usually, the suction volume (no check valve) is more than threeorders of magnitude greater than the discharge volume and istherefore considered at a constant pressure. The general idea isnow to consider only the mass flow into the piping volume (fromthe compressor) and the mass flow leaving this volume through therecycle valve. Since the gas mass in the piping volume determinesthe density and, thus, the pressure in the gas, we can for any instantsee whether the head required to deliver gas at the pressure in thepipe volume exceeds the maximum head that the compressor canproduce at this instant. Only if the compressor is always capable ofmaking more head than required can surge be avoided.

A further conceptual simplification can be made by splitting theflow coefficient of the recycle valve (cv) into a part that isnecessary to release the flow at the steady-state operating point ofthe compressor (cv,ss) and the part that is actually available toreduce the pressure in the piping volume (cv,avail).

The first stream and, thus, cv,ss of the valve necessary to coverit are known. Also known is the cv rating of the valve. Thus, theflow portion that can effectively reduce the backpressure isdetermined by the difference:

The model is run at constant temperature. Most of thecompressor systems modeled contain aftercoolers. The thermalcapacity of the cooler and the piping are much larger than thethermal capacity of the gas; thus, the gas temperature changes arenegligible within the first second. The flow calculated above ineach step of the iteration is then subtracted from the gascontained in the discharge and a new pressure in the pipe volumeis calculated. The calculation yields the maximum allowablepiping volume for the set parameters that will not cause surge atESD. Kurz and White (2004) validated this approach, using actual

test data. The method described can easily be expanded tosituations where a relatively small suction volume leads to a fastincrease in suction pressure.

Application

The model described above can be used in a software simulationprogram to rapidly evaluate whether a selected valve is sizedcorrectly for the piping volume.

The model iteratively determines the maximum allowabledischarge volume for a given valve configuration. This isimportant, because the valve size can be determined early in theproject phase. With a known valve configuration, the stationdesigner can be provided with the maximum volume of piping andcoolers between the compressor and the check valve, that allowsthe system to avoid surge during an ESD.

The calculation requires specification of the head-flow-speedrelationship of the compressor, and the definition of the surge lineas a function of either compressor speed, compressor head, orcompressor flow. Further, the valve needs to be described by itsmaximum capacity (Cv), as well as by its capacity as a function ofvalve travel and the opening behavior, including the delay. Thedischarge check valve is assumed to be closed as soon as therecycle flow exceeds the compressor flow, i.e., once thedepressurization begins. The calculation procedure is started byinitiating the deceleration of the train and the valve opening. Foreach time step, the compressor head and flow, based on speed andsystem pressures, and the flow through the valve, based on systempressures and valve opening, are calculated. The mass of the gastrapped between the recycle valve and compressor discharge issubsequently determined, yielding a new discharge pressure. Ifsurge occurs, i.e., if the flow drops below the flow at the surge line,the backwards flow through the compressor is assumed to increasewith time in surge, with a recovery once the required head drops 1percent below the head at surge. The modeling of the backwardsflow is not critical, and is only made to avoid numericalinstabilities, because the only information that is expected fromthe model is whether or not the compressor will surge for thegiven configuration.

Figure 13 shows the deceleration of the engine’s power turbine,and thus the driven compressor, following an ESD. Also shown isthe response of the recycle valve.

Figures 15, 16, and 17 show typical results of these simulations.In Figure 15, the discharge volume is small enough, and while theactual flow of the compressor approaches the minimum allowableflow (surge flow) at about 500 ms after the initiation of the ESD,so surge can be avoided. In Figure 16 the compressor surges about700 ms after the initiation of the ESD. For this configuration, eitherthe valve size has to be increased, or the discharge volume has tobe reduced, to avoid compressor surge during an ESD. In Figure17, the system is severely underdesigned and will requiresignificant changes including the possible addition of another valvein a hot bypass mode.

Figure 15. Actual Flow and Flow at the Surge Line During ESD—Surge Avoided.


c c cv avail v v ss, ,= −

Figure 16. Actual Flow and Flow at the Surge Line During ESD—Compressor - Surge at 0.7 Sec.

Figure 17. Actual Flow and Flow at the Surge Line During ESD—Multiple Surges.

RECYCLE ARRANGEMENTSFOR SPECIFIC APPLICATIONS

The arrangement of recycle loops impacts the operationalflexibility, as well as the startup and transient behavior ofa compressor station. In the following section, variousarrangement concepts are described, together with their basicadvantages and disadvantages.

Recycle Configurations for Single Compressors

• Basic recycle system (Figure 18)1Small discharge volume, fast recycle response2Although some partial recycle can be maintained, 100 percentrecycle cannot.

Figure 18. The Basic Recycle System (Hot Recycle).

• Aftercooler inside recycle loop (Figure 19)1100 percent recycle possible2Additional discharge volume impacts recycle response.

Figure 19. Aftercooler Inside Recycle Loop.

• Precooler and aftercooling (Figure 20)1Small discharge volume, fast recycle response1If there is significant heat in the suction header may improvecompressor performance2Requires an additional cooler

Figure 20. Precooler and Post Cooling.

• Cooled recycle loop (Figure 21)1Small discharge volume, fast recycle response1No inline pressure loss2Requires an additional cooler, although smaller than a precooler

Figure 21. Recycle Cooling.

• Hot recycle valve and cooled recycle valve (Figure 22)1Provides modulating surge control valve and shutdown valveideally suited for their purpose2More components, more cost

Figure 22. Hot and Cooled Recycle Valves.


Fig. 18: The Basic Recycle System (Hot recycle)

F

• Parallel recycle valves (Figure 23)1Provide good modulating surge control and fast shutdown valves1Provide some level of redundancy2More components, more cost

Figure 23. Parallel Recycle Valves.

Recycle Configurations for Multiple Compressors

• Hot unit valves and cooled overall recycle valve (Figure 24)1Provides good modulating surge control and fast shutdownvalves1Provides some level of redundancy2More components, more cost

Figure 24. Hot Unit Valves Cooled Overall Recycle Valve.

• Compressor 2 significantly larger than compressor 1: recyclevalve for second compressor and overall recycle valve (Figure 25)1Provides good modulating surge control and fast shutdownvalves1Provided some level of redundancy2More components, more cost

Figure 25. Recycle Valve for Second Larger Compressor andOverall Recycle Valve.

• Low ratio (low discharge temperature) individual compressorsconnected in series (Figures 26 and 27)

In either configuration any compressor can be started and putonline with the recycle valve closed. As the compressor begins tomake head its discharge check valve will open and its bypass checkvalve will close. In the configuration with a common recycleheader (Figure 27), multiple parallel recycle valves are necessary.This is due to the fact that the pressure ratio over the recycle valveis reduced if one or more of the compressors upstream are shutdown. This in turn requires added flow capacity in the recyclevalve. This configuration also requires a number of additionalblock valves.

Figure 26. Low Ratio Compressors in Series: Recycle forIndividual Compressor.

Figure 27. Low Ratio Compressors in Series: Common Recycle Header.

RECYCLE CONNECTION CONSIDERATIONS

Connecting Upstream or Downstream of Scrubbers

Connecting upstream of the suction scrubber will prevent debrisleft in piping from entering the compressor and will add volume inthe recycle path to extend time before overheat. Connectingdownstream of the suction scrubber will avoid the pressure dropacross the scrubber thus pressure rise at the compressor suctionwill be faster and higher.

Startup Considerations

The design of the antisurge and recycle system also impacts thestartup of the station. Particular attention has to be given to thecapability to start up the station without having to abort the startdue to conditions where allowable operating conditions areexceeded. Problems may arise from the fact that the compressormay spend a certain amount of time recycling gas, until sufficientdischarge pressure is produced to open the discharge check valve(Figure 3) and gas is flowing into the pipeline.

Virtually all of the mechanical energy absorbed by the compressoris converted into heat in the discharged gas. In an uncooled recyclesystem, this heat is recycled into the compressor suction and thenmore energy added to it. A cubic foot of natural gas at 600 psi weighsabout 2 lb (depending on composition). The specific heat of naturalgas is about 0.5 Btu/lb (again depending on composition). OneBtu/sec equals 1.416 hp. If the recycle system contains 1000 cubicfeet, there is a ton of gas in it. One thousand four hundred sixteen hpwill raise the temperature of the gas about 1 degree per second. Thisapproximates what happens with 100 percent recycle. At 100 percentrecycle, eventually this will lead to overheating at the compressordischarge. The problem usually occurs when there is a long periodbetween the initial rotation of the compressor and overcoming thepressure downstream of the check valve.

Low pressure-ratio compressors often do not require aftercoolers.There are several strategies that can be employed to avoidoverheating the uncooled compressor during startup:

• Accelerate quickly

• Delay hot gas re-entering the compressor

• Dilute hot gas re-entering the compressor

• Throttled recycle

Compressors without cooling must be accelerated and placedonline quickly to avoid overheating. Uncooled compressor sets


C1 C2

F

cannot be started and accelerated to idle. They must be acceleratedquickly through the point where the discharge check valve opens andthe recycle valve closes. If acceleration slows when the dischargepressure is met, and recycle valve closes slowly, a shutdown may stilloccur. Often standard start sequences are very conservative and canbe shortened to reduce the time it takes to get a compressor online.

Extending the length of the recycle line downstream of therecycle valve increases the total volume of gas in the recyclesystem. This reduces the heat buildup rate by delaying when thehot gas from the compressor discharge reaches the suction. Someheat will be radiated through the pipe walls. If the outlet is farupstream into a flowing suction header, dilution will occur.

Figures 28 and 29 outline a solution to a rather difficult startingproblem for a compressor station without aftercooling capacity. InFigure 28, to start the first unit is relatively easy, because there isvirtually no pressure differential across the main line check valve,and therefore the unit check valve will open almost immediately,allowing the flow of compressed gas into the pipeline. However, ifone additional unit is to be started, the station already operates at aconsiderable pressure ratio, and therefore the unit check-valve willnot open until the pressure ratio of the starting unit exceeds thestation pressure ratio. Ordinarily the unit would invariably shutdown on high temperature before this can be achieved. By routingthe recycle line into the common station header (Figure 29), theheat from the unit coming online is mixed with the station suctionflow. This equalizes the inlet temperature of all compressors; higherfor the compressors already online, lower for the compressorcoming online. With this arrangement overheating of a compressorcoming online is nearly always avoided.

Figure 28. Original Station Layout.

Figure 29. Improved Station Layout.

Figure 30 shows the problem of a conventional system that includes3000 ft of 24 inch pipe without aftercooling. The temperature in therecycle line starts to rise and, assuming a shutdown set point of 350ºF,the compressor would shut down after about 20 minutes.

Figure 30. Temperature Build Up.

Figures 31 and 32 outline the startup event with the revisedsystem. The power turbine and the compressor start to rotate oncethe gas producer provides sufficient power. Subsequently, the gastemperature rises, but, because the discharge pressure required toopen the check valve is reached fast enough, overheating can beavoided. The temperature rise in the recycle loop during startup isshown as a function of power turbine and compressor speed(Figure 31), gas producer speed (Figure 32), and time (in minutes)(Figure 33). The power turbine starts to turn at about 75 percent gasproducer speed, at which point the temperature starts to rise. Afterthe discharge check valve opens (at 0.2 minutes after the compressorstarts to rotate, 95 percent gas producer speed and 70 percentpower turbine speed), the temperature drops rapidly.

Figure 31. Temperature (ºF) Versus NPT (Percent).

Figure 32. Temperature (ºF) Versus NGP (Percent).

Figure 33. Temperature (ºF) Versus Time (Min.).


HP

HP

HP

HP

HP

HP

Further analysis of the startup problem indicates the advantageof throttling the recycle valve, rather than starting the unit with therecycle valve fully open.

With the valve sizing tool described previously it can be determinedexactly what valve opening will be required to maintain a specificsurge margin at steady-state operation. As the compressor isaccelerating, flow is increasing. The pressure in the discharge is lowerand the pressure in the suction is higher than they would be if the com-pressor at this speed were steady-state. This is due to the effect of thesuction and discharge volumes. This also causes the flow to be higherand subsequently the surge margin will be higher. As such, if the valveis set at a fixed position to obtain a fixed small surge margin, the actualsurge margin will be higher during acceleration.

To use this strategy safely the control must be able to sense a lossof acceleration (flame out) and if detected open all recycle valvesimmediately. As the volumes up and downstream of the compressorcause the surge margin to be higher during acceleration they makesurge avoidance more challenging with loss of speed.

Figure 12 illustrates this: at 70 percent open setting, the startupof the compressor is significantly closer to the surge line thanat 100 percent open setting. For any given speed, the powerrequirement of the compressor is lower when it is closer to surgethan when it is farther in choke. Therefore, for a given amount ofavailable power, the start is quicker if the compressor operatescloser to surge. If the rate of acceleration is quicker, the heat inputinto the system is lower. Actively modulating the surge duringstartup is virtually impossible as the parameters defining the surgelimit of the compressor are too low to be practically measured.Returning to Figure 5 the surge limit of a compressor matches wellwith a fixed travel (constant Cv) line for a recycle valve. As such,a compressor can be started with a fixed recycle valve position.

OUTLOOK

There are ongoing efforts to improve surge avoidance systems.One line of efforts attempts to increase the stability margin of acompressor by active (Epstein, et al., 1994; Blanchini, et al., 2001)or passive means (Arnulfi, et al., 2000). Other efforts try toincrease the accuracy of determining the surge margin (McKee andDeffenbaugh, 2003) by detecting the certain precursors of surge.However, most of the ideas will remain valid even if some of thenew methods, currently in an experimental stage, are introduced.This is due to the fact that surge avoidance is a systems issue andmeaningful gains can be made by better understanding theinteraction between the compressor, the antisurge devices (controlsystem, valves), and the station piping layout (coolers, scrubbers,check valves).

CONCLUSION

This paper has addressed the key factors that must be consideredin the design of surge avoidance systems. The most important pointis the realization that surge avoidance must be viewed in terms ofthe total system and not as an isolated item looking only at thecompressor itself.

NOMENCLATURE

A = Flow areacv = Flow coefficient (cv = Q/√SG/Dp)C = Compressible valve coefficientFp = Piping geometry factorh = HeadHcooler = Gas cooler heat transferJ = Inertiak = Isentropic exponentK = ConstantKv = Valve coefficientL = Pipe lengthN = Speed (1/s)p = Pressure

Q = Volumetric flowSG = Specific gravitySM = Surge margin (percent)T = Temperaturet = TimeV = VolumeY = Expansion factorZ = Compressibility factorα,β,γ = Constantsρ = Density

Subscripts

avail = Availablecompr = Compressorop = Operating pointp = Polytropicsurge = At surgestd = At standard conditionsss = Steady-statev = Valve1 = Compressor inlet2 = Compressor discharge

REFERENCES

ANSI/ISA S75.01, 1995, “Flow Equations for Sizing Control Valves.”

Arnulfi, G. L., Giannatasio, P., Micheli, D., and Pinamonti, P.,2000, “An Innovative Control of Surge in IndustrialCompression Systems,” ASME 2000-GT-352.

Bakken, L. E., Bjorge, T., Bradley, T. M., and Smith, N., 2002,“Validation of Compressor Transient Behavior,” ASMEGT-2002-30279.

Blanchini, F., Giannatasio, P., Micheli, D., and Pinamonti, P., 2001,“Experimental Evaluation of a High-Gain Control forCompressor Surge Suppression,” ASME 2001-GT-0570.

Day, I. J., 1991, “Axial Compressor Performance During Surge,”Proceedings of the 10th International Symposium on AirBreathing Engines, Nottingham, United Kingdom, pp. 927-934.

Epstein, A. H., Ffowcs Williams, J. E., and Greitzer, E. M., 1994,“Active Suppression of Compressor Instabilities,”AIAA-86-1994.

Kurz, R. and White, R. C., 2004, “Surge Avoidance in GasCompression Systems,” TransASME Jturbo, 126, pp. 501-506.

Kurz, R., McKee, R., and Brun, K., 2006, “Pulsations inCentrifugal Compressor Installations,” ASME GT2006-90070.

McKee, R. J. and Deffenbaugh, D., 2003, “Factors that AffectSurge Precursors in Centrifugal Compressors,” Proceedings ofGMRC Gas Machinery Conference, Salt Lake City, Utah.

Nakayama, Y., 1988, Visualized Flow, Oxford, United Kingdom:Pergamon Press.

Sentz, R. H., 1980, “The Analysis of Surge,” Proceedings of theNinth Turbomachinery Symposium, Turbomachinery Laboratory,Texas A&M University, College Station, Texas, pp. 57-61.


Donald H. Mansfield is Manager ofBladed Rotor Application Engineering forElliott Company, Ebara Group, inJeannette, Pennsylvania. He has more than33 years experience in the Compressor andTurbine Application Engineering, ProjectManagement, and Marketing groups withinElliott Company. Currently, he is responsiblefor steam turbine, gas expander, and axialcompressor selection standards, specification

reviews, and all aspects related to proposal preparation for new andrerated equipment.

Mr. Mansfield received a B.S. degree (Mechanical Engineering,1970) from the University of Pittsburgh.

Douglas W. Boyce is a SeniorApplication Engineer for Elliott Company,Ebara Group, in Jeannette, Pennsylvania,in the Global Services operation. He hasbeen in the application engineeringfunction at Elliott Company for his entiretenure of 38 years. This experience isequally divided in the company’s steamturbine, compressor, and vacuum technologyproduct lines. Present responsibility is in

engineering of reapplied compressors and steam turbines.Mr. Boyce received a B.S. degree (Mechanical Engineering,

1969) from Lehigh University. He is a registered professionalengineer in the State of Texas.

William G. Pacelli is the Manager of YRSteam Turbine Product Engineering forElliott Company, Ebara Group in JeannettePennsylvania. He has 24 years’ experiencein turbine design and service engineering.Currently he is responsible for YR steamturbine product engineering, design, anddevelopment.

Mr. Pacelli received a B.S. degree(Mechanical Engineering, 1982) from the

University of Pittsburgh. He is a registered professional engineerin the Commonwealth of Pennsylvania.

ABSTRACTSteam turbines may be rerated, reapplied, or modified to meet

several specific goals. This tutorial will review possible reasonsfor these changes, including optimizing performance, improvingreliability, reducing maintenance requirements, solving operatingproblems, extending equipment life, and, finally, replacement ofthe turbine, either in whole or part, due to a catastrophic failure,normal wear, or problems found during an inspection. Whateverthe project, it normally requires clear goals specified by the userand buyer. The vendor can then review these goals and determinewhat can realistically be accomplished within the constraints of theexisting equipment, the project budget, and the time allowed. Mostof the equipment limitations will be discussed in this tutorial aswell as the presentation of a case study. For many of these projects,the vendor will not have access to the turbine since it will stillbe operating at the user’s facility. Therefore, closer than normalcoordination between the user and the vendor is required to ensuresmooth and timely completion of the project.

INTRODUCTIONThe first part of this tutorial will define many of the reasons for

rerating or reapplying a steam turbine, as well as what hardware orsoftware may be available to accomplish the task. This is followedby a listing of the information needed from the user and anexplanation of the limitations faced by the vendor—specificallythose that are not usually encountered with new equipment.Finally, a case study of an uprate of a mechanical drive turbineis presented.

PERFORMANCE CHANGESThe basic power and/or speed rating of a steam turbine may

change for many reasons. The most common one is an increase (ordecrease) in the power required by the driven machine due toa plant expansion or debottlenecking. Other reasons include areapplication of the turbine to drive a different machine, a searchfor increased efficiency, a change in the plant steam balance, or achange in steam pressure or temperature.

Power

An increase in power usually requires more steam flow areainside the turbine, which may or may not be possible within thephysical limits of the existing casing. A decrease in power isalmost always possible simply by blocking off some flow area, butmaintaining efficiency usually requires a more sophisticated

135

GUIDELINES FOR SPECIFYING AND EVALUATINGTHE RERATING AND REAPPLICATION OF STEAM TURBINES

byDonald H. Mansfield

Manager of Bladed Rotor Application Engineering

Douglas W. BoyceSenior Application Engineer, Global Services

William G. PacelliManager, Product Engineering, YR Steam Turbines

Elliott Company


solution. Reapplication of a turbine often is the most difficultproblem for the vendor since the power, speed, and steam conditionscan all be considerably different from those for which the turbinewas initially designed. This is discussed in more detail later inthis paper.

Efficiency

More efficient blading may be available for older turbines.Considering the cost of energy today, it may make sense to investin new blades (both rotating and stationary) if the gain in efficiencyis large enough. However, higher efficiency blading often requiresmore axial spacing along the rotor, and there may not be enoughroom in the existing casing.

There are simpler ways to improve or maintain efficiency.Leakage through labyrinth seals can be reduced by up to 80 percentby integrating brush type seals with the usual stationary labyrinthsas shown in Figure 1.

Figure 1. Labyrinth-Brush Seal.

The brush seal consists of bristles that are angled slightly withshaft rotation. They can tolerate some deflection and still springback to their original position. They can be incorporated betweenthe labyrinth teeth as well as at the ends. For higher pressures,brush seals may require pressure balancing to avoid excessivedownstream deflection. Note that due to the angled bristles, somebrush seals may not tolerate reverse rotation. Figure 2 shows brushseals in a heavy metal retainer.

Figure 2. Brush Seal in Metal Retainer.

Labyrinth teeth can be damaged by rubs, especially duringstartup or coastdown when the turbine rotor passes through alateral critical speed. Rubs may also occur at startup due todifferent rates of thermal expansion between the seal and the rotor.The rub opens the clearance and reduces efficiency. Retractablepacking is a possible solution if your turbine has this problem. Thelabyrinth ring is circumferentially divided into segments that arespring-loaded to hold them apart and therefore give a verygenerous clearance. Once the turbine starts and steam pressurebuilds up on the outside diameter of the seal, it overcomes thespring pressure and it closes the seal to normal clearance. When theturbine trips, steam pressure is reduced and the spring again opensthe clearance for coastdown.

Almost all turbine stages have seals between each diaphragmand the shaft, but many turbines do not have tip seals between therotor blade tips and the casing/diaphragm. If room permits, tipseals can be added to increase efficiency. Tip seals are moreeffective when used in higher pressure stages and stages withgreater reaction. Tip seals are shown in Figure 3.

Figure 3. Tip Seals.

Any of the seals mentioned above can combine the features ofthe brush seal and retractable seal.

RELIABILITY AND MAINTENANCE

The trend in most industries is for longer runs and shorterturnarounds. There is also a lot of pressure to eliminate unplannedshutdowns and reduce required maintenance.

Electronic Controls

Perhaps the biggest change in steam turbines in the past 20 yearshas been the conversion of the speed control and trip systems fromtotally mechanical to totally electronic. Many existing turbines canbenefit from a change to electronic controls.

Figure 4 shows a typical mechanical governor system for astraight-through turbine. The governor is powered off the turbineshaft by a worm and wheel drive. The governor itself has manyinternal moving parts including flyweights, springs, an oil pump,accumulator pistons, and several valves. These governors are quitereliable, but with all those moving, contacting parts, wear andmaintenance are inevitable.


Figure 4. Mechanical Governor Straight Through Turbine.

An electronic governor replaces the worm gear with a toothedwheel. Multiple noncontacting magnetic speed pickups read off thetoothed wheel and provide the signal to the electronic governor.The power cylinder in Figure 4 is replaced with an electronic orpneumatic actuator. The pilot valve and restoring linkage remainintact. Many electronic governors are also more versatile in thatthey can be programmed to use parameters other than speed aloneto control the turbine. Most electronic governors can also interfacewith the user’s distributed control system (DCS).

Things get much more complicated for mechanical controls whenthey are applied to an extraction turbine, as shown in Figure 5. Herethe power cylinder controls the added extraction pressure regulator,which includes even more linkages and springs as well as a pneumaticsignal to monitor extraction steam pressure. The extraction pressureregulator sends signals to either prepilot cylinder on either or bothservos. This system can maintain a set turbine speed and an extractionpressure level. However, the linkages require maintenance, thelinkages and springs wear, and if there is a change to extractionpressure, the linkages and springs need adjustment.

Figure 5. Mechanical Governor Controls for an Extraction Turbine.

An electronic governor designed for extraction turbines works asdescribed above for straight-through turbines, with the additionthat it receives an electronic signal for extraction steam pressureand outputs signals to the electronic (or pneumatic) actuators thatreplace the prepilot cylinders on each servo motor. Again othersignals (such as a compressor discharge pressure signal) can be fedto and processed by the electronic governor. Here a change to theextraction pressure is a simple set point change.

Electronic governors are not failproof, but they are available inredundant and triple modular redundant formats, so failure ofelectronic components will cause the governor to switch to backupcomponents while the failed components are replaced.

Most multivalve turbines still require hydraulic servo motors since

electronic ones are not powerful enough to move the valve racks. Wecan go a step further, however, and eliminate the prepilot and pilotvalves, in addition to the linkages, cams, and rollers shown as therestoring linkage in Figures 4 and 5. Figure 6 shows that the servomotor can be fed by a way valve. The way valve takes the oil directlyfrom the control oil header and directs it to the proper side of theservo motor. The electronic governor sends a control signal to theactuator coil that is an integral part of the way valve. The restoringlinkage is replaced by one or more linear variable differentialtransformers (LVDTs) that provide feedback to the control system.The way valve may have dual actuator coils for redundancy.

Figure 6. Way Valve Retrofit.

Turbines are still tripped by dumping the oil that holds the tripand throttle valve open. Figure 7 shows the arrangement formerlyused on most turbines. The overspeed trip was initiated by a spring-loaded trip pin or a weighted Bellville spring that struck amechanical lever that actuated a dump valve. The solenoid dumpvalve shown was for remote tripping purposes.

Figure 7. Mechanical Overspeed Trip.

Electronic trips, which are standard on most new turbines, usethree noncontacting magnetic speed pickups that read off a toothedwheel. This is the same method that sends the electronic governorits signal. The electronic overspeed trip control box uses two out ofthree voting logic to trip the turbine and eliminate spurious trips.Two solenoid dump valves are piped in parallel to provideredundancy. It is sometimes difficult to find enough room on an


137

existing turbine to mount the necessary speed pickups, but beyondthat it is fairly simple to add an electronic overspeed trip.

Monitoring Systems

There are many steam turbines that have been in operation for 30or more years that do not have the level of instrumentation that isconsidered standard today. Radial and axial vibration probes, as wellas bearing temperature instrumentation, may be fitted to existingturbines. Readouts from these instruments can be fed to sophisticatedmonitoring equipment that not only can display the current readings,but also can determine trends that may uncover a potential problembefore it causes an unexpected outage. The instruments couldindicate that a change to a different bearing (tilt shoe, spherical seat)or orientation, such as load between pad, would be beneficial.

The main stumbling block to retrofitting vibration probes isfinding the space in the bearing housings to mount them with acorresponding free area on the shaft to read from. The shaft areamay also have to be treated to reduce electrical runout to anacceptable level. Bearing temperature instrumentation willnormally require some machining of the bearing retainers andhousings to allow the wires to exit the turbine.

The user should be aware that it is fairly common for machinesthat have been operating satisfactorily for many years to show highvibration readings once probes are installed. The excessivevibration may have always been there, but it just has not been to thepoint where it has caused a problem. In such cases, the vibrationlevel is normally still not a problem. Although it exceeds currentstandards, if the vibration has not caused any operating ormaintenance problems in the past, there is no reason to believe thatit will be a problem in the future. With the probes installed,however, it is now possible to monitor and trend the vibrationlevels and to avoid potential future problems.

OPERATING PROBLEMSMany operating problems may be solved by some of the items

that have already been discussed. Monitoring systems can uncoverproblems such as a bowed rotor or fouling. There are also muchbetter tools to determine rotordynamic characteristics than therewere 30 years ago. The seals that were discussed can eliminatewater contamination of the oil and excessive gland leakage. Theelectronic controls can cure some process control problems or speedfluctuations. Speed fluctuation can also be caused by improperlysized governor valves, or valve and seat wear. Performanceproblems have to be analyzed on an individual basis. Again, thereare more sophisticated tools available now than there were just afew years ago, such as computational fluid dynamics (CFD).

LIFE EXTENSIONThere are several options available to extend the life of an

existing turbine.Most types of rotor damage can now be repaired by machining

off the damaged area, rebuilding it with weld, and remachining therotor. Some casing damage may be repaired in a like manner.

Better materials are often available to solve erosion problems.Blades also can be protected by a stellite overlay in critical areas.Some coatings are also available that may help with erosion,corrosion, or fouling problems.

A bearing upgrade or something as simple as an at-speedbalance could also extend the life of a turbine by eliminating avibration problem.

REPLACEMENTIf the user knows in advance that they will be replacing their

turbine, they have many options. A new turbine is certainly an option.Or there may be used turbines on the market that can be refurbishedto meet the requirements. The user may even have another turbinethat they would like to reapply to the existing service.

If the turbine is damaged in a wreck, it is an entirely different

matter. This is an emergency and a quick solution is essential. A newturbine is usually out of the question because the lead time is too long.

Depending upon the extent of the damage, most originalequipment manufacturers (OEMs) and shops that specialize inturbines can accomplish fairly major repairs in a short period oftime by utilizing welding, plating, and other proven repairtechniques. Blades and other specialized parts might be availablein a reasonable amount of time.

Catastrophic damage will require a search for a suitablereplacement. Again, most OEMs and turbine specialty shops keeptrack of equipment that is available on the used equipment market.The vendor may even have used turbines of his own to reapply. Asuitable turbine may be overhauled with minor changes and beback in service in a fraction of the time required for a new unit.

If no suitable turbine can be found on the used market, it mayeven be possible to reapply another of the user’s turbines (possiblyin a less critical service) to replace the damaged turbine. It may thenbe feasible to replace that turbine with either a new or used unit.

STARTING THE PROCESS

Define the Objective

Although not considered a vital part of the process, possibly themost critical part of the project is to define the objective of theproject. Many times projects are initiated without a clear objective.This lack of a defined goal often leads to failure of the entire project.

The process starts by an analysis of the driven machine to determinethe actual power requirement and speed and evaluation of steamconditions at all the expected operational points, including excesspower margin. Steam pressure drops in the supply, exhaust, andextraction piping must be accurately calculated. On extraction units,the extraction pressure and flow must also be defined.

In the initial purchase, many turbine standards were considered,including corporate, American Petroleum Institute (API), NationalElectrical Manufacturers Association (NEMA), industrial, andlocal governmental standards. It is up to the engineering staffdirecting the changes to determine which, if any, of these specifi-cations are required to be applied. Some of the originalspecifications or specifications that have been enacted since theunit’s original commissioning may need to be applied. One note ofcaution: older turbines must be carefully analyzed, as many olderunits cannot meet current standards or may only do so at consider-able added lead time and expense.

Construction Rating

Turbines undergoing a change in inlet pressure, temperature,and/or exhaust pressure require that maximum casing pressure andtemperature rating be checked. This includes a review of the casingand flange ratings to ensure the casing is suitable for the new con-ditions. Multistage turbines with changes in pressure and flowrequire a verification of the first stage maximum allowablepressure. In the case of a multistage extraction turbine, a change inflow or extraction pressure will also require that intermediatecasing and extraction flange be reviewed.

A change in inlet temperature will require a verification of thecasing material. Typical limits are given in Table 1.

Table 1. Casing Material Temperature Limits.


Casing Material Maximum Temperature

(Deg F) ASTM A278 Class 40 600

ASTM A216 Grade WCB 775 ASTM A217 Grade WC1 825 ASTM A217 Grade WC6 900 ASTM A217 Grade WC9 1000

There are many other materials that have been used over theyears, but these are the major casing materials currently in use.Before exceeding these limits, a thorough review of the metallurgyof the existing casing must be done.

Once the casing rating has been determined, the original hydrotestpressures must be determined. This can be done by contacting theoriginal OEM, reviewing the original construction documentation (i.e.,hydrotest certificates, data sheets, correspondence), or reviewing theturbine casing for markings. The hydrotest pressure must meet the newconditions based on the latest industry standards. The API guidelinesrefer to ASME Boiler & Pressure Vessel Code - Section VIII (2004).In some cases, the casing will need to be rehydrotested to the newconditions. This may need to be done to one or several sections or theentire turbine casing.

A decision must be made as to whether NEMA limits need to betaken into account. For new construction, most OEMs verify thecasing design will ensure the casing will meet the maximum designpressure and temperature plus 120 percent of the rated pressurecombined with a temperature increase of 50ºF. This verification isalso included in the evaluation of the flange rating.

Flange Sizing

If during the modification of the turbine, the flow increases orthe specific volume decreases, the size of the inlet flange must bereviewed. Typical maximum value for the inlet velocity is 175ft/sec. Equation (1) determines the inlet velocity.

where:V = Velocity in flange (ft/sec)M = Mass flow (lb/hr)d = Diameter of the inlet (inch)ν = Specific volume of steam (cu ft/lb)

In smaller turbines, the inlet flange with the steam chestand control valving can easily be changed. On larger turbines,increasing the inlet size may be very difficult. Some turbines havethe ability to add an additional inlet on the existing steam chest bywelding on a flange followed by a local stress relief. Figure 8shows a flange undergoing modification from an 8 inch inlet to a10 inch inlet.

Figure 8. Increasing a Flange in Size.

Some turbines may have a blank flange that can be easilyremoved and piped for additional inlet area. Refer to Figure 9 for atypical steam end with dual inlet capability.

Figure 9. Inlet Casing with Dual Inlet Flanges.

The trip and throttle (T&T) valve must also be reviewed, but thiswill be discussed in a later section. As a worst case option, thesteam velocity can be allowed to exceed the 175 ft/sec limit as longas the correct pressure drop is taken into account and additionalacoustic protection is provided as the noise level will increase.

The extraction and exhaust line sizes are also areas that need tobe considered. The maximum value for an extraction line is 250ft/sec. The maximum for a noncondensing exhaust is 250 ft/sec andfor a condensing exhaust it is 450 ft/sec. Typically there are notmany options to upgrade an extraction or exhaust connection. Themost practical solution is to increase the exhaust header size asclose to the turbine casing as soon as possible. Again theappropriate pressure drop calculations must be done to determinethe pressure at the flange. It will be an iterative procedure withgood communication between the engineering staff involved in thererate and the engineering staff with site responsibility.

Nozzle Ring Capacity

After the external issues have been decided, the first internalcomponent requiring review is the nozzle ring or nozzle block.This is the inlet to the control stage that ultimately controls theamount of steam a given turbine will be able to pass. If during theoriginal manufacture of the turbine, the nozzle ring had additionalarea available, the turbine may be uprated by adding additionalnozzles. Depending on the design, this can be as easy as installinga new nozzle ring. Some small turbines have nozzles machined intothe steam end and modifying these turbines is difficult, if notimpossible. It may be cost effective to purchase a replacementturbine if this is the case.

If the change in flow is significant, the nozzle may have to bereplaced with a nozzle ring that has an increased nozzle height. Thiswill normally require increased blade height on the turbine rotor. Themaximum nozzle height will be dictated by the height of the openingin the steam end of the turbine. The maximum number of nozzles ineach bank or segment of the steam end is dictated by the originaldesign parameters. Many times, the maximum nozzle area in a steamend is matched to the volumetric area of the inlet of the turbine.

The same issues affect the extraction nozzle ring on an extractionturbine although the extraction diaphragm in some instances can bechanged or modified to allow for additional nozzle area.

Steam Path Analysis

Once the inlet nozzling has been reviewed, the remainder of thesteam path must be reviewed. An increase in flow may require anincrease in diaphragm nozzle height and rotor blade height in someor all stages. As stage flow increases, the heat drop across the stagewill increase with the associated decrease in velocity ratio. Thevelocity ratio is defined as the ratio of the blade at the pitchdiameter to the steam jet velocity, as defined in Equation (2).


139

V M d=. /0509 2ν

VDN

ho =

π224 ∆

where:Vo = Velocity ratioD = Pitch diameter of rotating bladesN = Shaft rotational speedDh = Enthalpy drop across the stage

In order to increase the efficiency, the diaphragm and blade pathareas must be carefully matched. Each stage design has an optimumvelocity ratio with the equivalent maximum efficiency. During themodification, some of the stages will require changes for mechanicalor other reasons. Some of the stages will be acceptable with less thanoptimal efficiency. Decisions must compare the loss in efficiencyversus the cost of modifying each stage.

Rotor Blade Loading

With the change of the power, flow, or speed of a turbine, eachrotating row of blading requires review of its mechanical properties.This includes a review of the speed limits and analysis of bothGoodman and Campbell diagrams. The blade mechanical speedlimits are related to the disk and blade geometry, stress values formaterial, and shroud design.

The Campbell diagram graphically compares the blade naturalfrequency to the turbine speed range. The natural frequencies of theblading can be obtained from the OEM or determined experimentally.If the speed range is not changed in the modification, the Campbelldiagrams will not change except for any blading that has beenchanged. Figure 10 illustrates a typical Campbell diagram.

Figure 10. Campbell Diagram.

The Goodman diagram will evaluate the combined effects ofalternating stresses and steady-state stresses in the turbine rootand base of vane locations. This diagram contains a Goodmanline, which in theory represents the minimum combination ofsteady-state stress and alternating stress, above which a bladefatigue failure could occur. Figure 11 is a typical Goodmandiagram. The OEM will normally have minimum allowable values

for a Goodman number. There may be several values depending onthe location of the blade within the steam path; typically partial arcadmission and the moisture transition stages will have higherlimits. A significant change in the stress levels of the blading mayrequire replacement or redesigning of the blading. In some rarecases, no blading design can be found and low Goodman numbersmay need to be accepted.

Figure 11. Goodman Diagram.

Thrust Bearing Loading

In many cases, an increase in flow and speed may increase thethrust values. Depending on the age of the turbine in question, areplacement thrust bearing may be required. There are highercapacity thrust bearings available that will fit in the existing cavity.In the most extreme cases, a bearing housing with a larger thrustbearing may be required.

Governor Valve Capacity

During the design cycle, the capacity of the governor valve mustbe checked to ensure the correct flow area is present. The valvearea must be carefully matched to the area of the nozzle ring. Inmost cases the nozzle area should be the limiting point of the flowin a valve bank. Setting the valve as the limit will not provide themost efficient conversion of pressure into velocity.

If the valve(s) and seat(s) require an increase in size, most casingscan accept a larger size valve and seat. Some single valve turbinesare able to have the steam chest size increased to the next larger size.

Shaft End Torque Limit

With an increase in power or a decrease in speed, the shaft endstress will need to be reviewed. The basic equation required tocalculate shaft stress is given in Equation (3).


τ =321000

3

P

Nd

where:τ = Shaft shear stressP = Power (hp)N = Shaft speed (rpm)d = Shaft diameter (inch)

The limit of the shaft shear stress is a function of the materialand the heat treatment as well as the shaft end design. Typicalturbine shaft ends are single or double keyed, NEMA taper,hydraulically fit, or integral. Each of these has different allowablestresses depending on the stress risers within the design. Typicalturbine shaft materials are given in Table 2.

Table 2. Turbine Shaft Materials.

The loading of the driven machine may also have an effect onthese values. For example, a generator drive may reduce theallowable stress further to account for short circuit loading.

Speed Range Changes

Although this has been briefly discussed in other sections, caremust be taken when making significant changes to the speed range.The most important is the blading speed limits (refer to theprevious section, Rotor Blade Loading), but other areas requireanalysis such as lateral critical speeds, torsional critical speeds,coupling speed limits, etc.

Lateral critical speeds are a function of the rotor design, bearingdesign, and bearing support design. Typically the original criticalspeed is denoted on the nameplate for the turbine and also in thevendor literature. If no changes are being done to the rotor andoperational data agree with the noted critical speed, this valuenormally will not change. Should changes be made to the rotor orbearing system, a new lateral analysis will need to be done toconfirm the speed of the lateral critical speed. A typical report maycost between $3000 and $30,000 depending on the complexity ofthe rotor and what existing data can be reused.

Torsional critical speeds will be required. Each body in thestring must be modeled as well as each coupling and gear set.Direct drive units are not normally an issue, but strings with gearsrequire the torsional to be checked. These may not be as readilyavailable as the lateral critical speed. Normally this documentationis supplied by the vendor in the form of a report with engineeringdocumentation. The cost to complete a report will vary from $3000to $10,000 per body.

The turbine governor, which will need to be modified orreprogrammed to meet the new speeds as well as the tripmechanism, could either be mechanical or electronic. Other drivensystems will require review such as a gear (if used) or any shaftdriven oil pumps.

Auxiliary Equipment Review

Each turbine has auxiliary equipment that will require somereview during any modification. These may include the T&T valve,other steam block valves, leakoff and sealing steam system, surfacecondenser system, relief valve sizing, lubrication and control oilsystems, valve actuation system, and supervisory instrumentation.

The T&T valve may be acceptable as is for the change in theflow, but may be upgraded for reliability or the addition of a manual

exerciser. Some T&T valve OEMs are no longer actively pursuingthis market, so an upgrade may be more expensive than a completenew valve of a current design. For extraction turbines, the nonreturnvalve should also be looked at to determine acceptability oflong-term continued operation. Other steam valves in the systemshould be reviewed for proper sizing and good mechanical operation.

The leakoff and sealing steam system must be reviewed todetermine if additional leakoff flow will be experienced with thenew conditions. API 612 (1995) and 611 (1997) both require 300percent capacity in the leakoff system. If a change has been madeto the first stage pressure or the exhaust pressure, this additionalcapacity may be difficult to achieve without major rework. Thegland condenser will require a review along with the ejector orvacuum pump to ensure these are of adequate capacity. From apiping point of view, it may be a good decision to replace some orall of the leakoff and drain piping if significant corrosion or buildupis noted on the inner diameter (ID) of the piping from the turbine.

On a condensing turbine, the surface condenser system must bereviewed to ensure the proper capacity is available with the currentcooling water temperature and available cooling water flow. Thehotwell capacity and pump sizing must be verified to ensure the correcthotwell level can be maintained. Instrumentation connected with thecondenser system should be reviewed and upgraded at this time.

When modifying a backpressure turbine, the relief valve sizingmust be reviewed once the final flow capacity is determined toensure the exhaust casing is protected from overpressurization.With a condensing turbine, the relief valve or rupture disc sizingmust be reviewed to ensure the exhaust casing is prevented fromgoing over the maximum pressure rating, which is normally 5 psig.

The lubrication and control oil system must be carefully checkedto ensure the new required oil flow and the proper cooling capacityare available. If the control oil system is being modified, check toensure the proper oil pressure and flow are available.

CASE STUDY

This example illustrates the rerate procedure for a large mechan-ical-drive steam turbine in a process plant. The turbine of interestdrives a compressor train in ethylene charge gas service. The com-pressors were to be rerated as part of a plant expansion, and theas-built turbine power rating was insufficient for the planned com-pressor rerate.

Necessary turbine performance for the driven equipment modi-fications is shown in Table 3.

Table 3. Summary of Performance Changes.

There is no extraction steam requirement in this application.The scope of hardware change to affect this magnitude of power

increase was beyond that of an efficiency upgrade alone. Steamflow could roughly be expected to increase by the same 25 percentas rated power. The flow limit of each steam path section wasevaluated in order to use the existing casing.

Using Equation (4), inlet steam flow, G, is:


141

Commercial Designation

Material Typical Maximum shear stress (PSI)

AISI C-1040 Medium Carbon Steel 5000 AISI 4140 Chrome Moly Alloy

Steel 11500

AISI 4340 Nickel Chrome Moly Alloy Steel

12500

ASTM A470 Class 4, 7, 8

Nickel Chrome Moly Vanadium Alloy Steel

Forging

11500, 12500,12000

Parameter Change (Rerate vs. Original)

Rated Power +25% Rated Speed -1% Steam Inlet Pressure No Change Steam Inlet Temperature No Change Exhaust Pressure As Required

S R P G. . × =

where S.R. is the steam rate in lb/hp-hr and the rated power, P, isin horsepower. For this rerate, power is 25 percent higher than theoriginal design. Estimated inlet steam inlet flow is:

For this particular turbine, the original design specifications andturbine layout were available for reference. As a first step, theengineer prepared a turbine summary shown in Table 4.

Table 4. Turbine Summary.

The engineer next qualified the turbine casing for this rerate bycalculating velocities at the inlet and exhaust flanges. Exhaustpressure for this rerate is permitted to vary, so the engineerassumed a pressure of 4 in HgA. Table 5 shows a comparison offlange velocities with NEMA maximum velocity.

Table 5. Flange Velocity Estimate.

Inlet velocity exceeding NEMA maximum did not disqualify theturbine casing at this phase of the rerate. However, this area of theturbine would have to be carefully checked after actual performancewas calculated.

The engineer proceeded to the next step of selecting a preliminarysteam path for the rerate. However, axial space available on therotor and casing dimensions allowed larger geometry nozzle and

buckets only on the Curtis control stage. The same high strength900 lb nozzle and bucket profiles were retained. Existing stagingdownstream of the Curtis stage would be retained or replaced asindicated by shop inspection.

Calculations showed that the existing 12-stage steam path couldnot pass the required additional flow. As flow in the modelincreases, stage pressure at each successive nozzle row alsoincreases. This takes place in an actual turbine because nozzle areais fixed. In this case, pressure at the first stage nozzle exit rose tothe point at which even new larger nozzles became choked.

In general, flow can increase through a certain size nozzleuntil the resulting pressure drop declines to a critical ratio ofapproximately 0.6. For ratios higher than 0.6, nozzle flow cannotincrease and the nozzle is described as choked.

The only available means of reducing the first stage nozzle exitpressure was to remove nozzles and buckets, starting with the secondstage. The engineer determined that the required power could onlybe achieved by removing the existing second and third stages.

Pressures at each remaining downstream nozzle row were higherthan in the original power rating. The diaphragm of the new secondstage (original fourth stage) was replaced with a reinforcedweldment. Removal of the second and third stages relieved the firststage nozzle exit pressure to an acceptable ratio with inlet pressure.Diaphragms and rotor discs of these stages were replaced with aflow tunnel to smooth the flow in the empty area of the casing.Performance with the new 10-stage steam path is shown in Table 6.

Table 6. Calculated Rerate Performance.

The rerate stage selection is compared with the originalarrangement in Table 7. Note that stages downstream of theoriginal third stage are unchanged.

Table 7. Rerate Staging Comparison with Original Staging.


7 50 37 180 125 348 563. , . , /× × =hp lb hr

Frame size 2NV-11 Rated Power 37,180 hp Rated Speed 4200 rpm Maximum Continuous Speed

4200 rpm

Inlet Steam Pressure 600 psig Inlet Steam Temperature 750°FExhaust Pressure 2 in. Hg Abs. Steam Rate 7.50 lb/hp-hr Inlet Flange Size 10” CL900FF Exhaust Flange Size 87 inches × 110

inches rectangular (100 inches diameter equivalent)

Exhaust Quality 92% Staging Description Curtis control stage

11 Rateau stages, all impulse type, 32 inch to 54 inch pitch diameter

Rotating Blade Heights 2 inch average on the Curtis stage increasing to 16 inches

Shaft Diameter Between Rotor Disks

12.5 inches

Nominal Shaft End Diameter

7.0 inches

Journal Bearings 7 inch diameter, tilt pad, at inlet end 8 inch diameter, tilt pad, at exhaust end

Thrust Bearing 112.5 square inches Main Casing Seals Labyrinth Governor type Oil relay

Flange Calculated Velocity

NEMA Table 8-1

Maximum 10” Main Inlet 196 175 87” × 110” Exhaust

289 450

Parameter Calculated

Rated Power 46,350 hp Rated Speed 4150 rpm Rated Steam Flow 336,210 lb/hr Inlet Steam Pressure 600 psig Inlet Steam Temperature 750°FExhaust Pressure 4 in. Hg Abs. Exhaust Quality 91%

Casing Stage No.

Original Design

Nozzle Ht.

Rerate Design

nozzle ht. 1 1.125 in 1.250 in 2 0.813 in - - - 3 0.938 in - - - 4 1.125 in 1.125 in 5 1.500 in 1.500 in 6 1.875 in 1.875 in 7 2.375 in 2.375 in 8 2.450 in 2.450 in 9 3.600 in 3.600 in

10 4.170 in 4.170 in 11 6.500 in 6.500 in 12 14.76 in 14.76 in

To confirm suitability of this turbine steam path, the engineerconducted a series of limit checks, starting with flange velocities,shown in Table 8.

Table 8. Rerate Flange Velocity.

NEMA SM 23 (1991) maximum inlet velocity may be regardedas a suggested limit. Exceeding this velocity did not disqualify theturbine casing. To provide sufficient flow area in the steam chestand reduce losses, all governor valves were replaced.

Blade stress was acceptable for all rows of buckets. Governorspeed range did not change, so blade frequency was of concernonly on the new Curtis bucket rows. Evaluation showed that therewas no interference with natural modes for these buckets within theoperating speed range.

The engineer evaluated the turbine shaft end diameter at thererate power of 46,350 hp. From Equation (6), we have minimumshaft end diameter, d, equal to:

Maximum allowable torsional shear stress for the existing shaftmaterial was 11,500 psi. The equation for diameter then became:

The new conditions required a minimum shaft diameter of 6.8inches. Since the existing shaft end diameter was 7.0 inches, nonew shaft or subarc weld buildup of the existing shaft end wasrequired.

CONCLUSION

Rerating or reapplying steam turbines can save considerabletime and money compared to buying new units. It is usuallypossible to get more power and/or speed out of existing units bychanging items in the steam path. Efficiency can often be improvedat the same time. Older steam turbines can be brought up topresent day turbine standards by retrofitting items such asmore efficient seals, modern bearings, electronic controls and,monitoring/trending systems. Turbine life can be extended bychanging to better materials, adding coatings, and repairingexisting damage. These courses of action, however, have morephysical constraints and normally require closer coordinationbetween the buyer, user, and vendor than in the purchase of a newmachine. Attention to detail in these matters will result in asuccessful project.

REFERENCES

ASME Boiler & Pressure Vessel Code - Section VIII - PressureVessels, 2004, American Society of Mechanical Engineers,New York, New York.

API Standard 611, 1997, “General-Purpose Steam Turbines forPetroleum, Chemical, and Gas Industry Services,” FourthEdition, American Petroleum Institute, Washington, D.C.

API Standard 612, 1995, “Special-Purpose Steam Turbines forPetroleum, Chemical, and Gas Industry Services,” FourthEdition, American Petroleum Institute, Washington, D.C.

NEMA Standard Publication SM-23, 1991, “Steam Turbines forMechanical Drive Service,” National Electrical ManufacturersAssociation, Washington, D.C.


143

Flange Calculated Velocity

NEMA Table 8-1

Maximum 10” Main Inlet 189 175 87” × 110” Exhaust

279 450

( )( )( )d

P

S N=

321 0003

,

( )( )( )d

d inches

=

=

321 000 46 350

11500 4150

68

3, ,

,

.

Dr. Klaus Brun is currently ProgramManager at Southwest Research Institute,in San Antonio, Texas. His experienceincludes positions in business development,project management, engineering, andmarketing at Solar Turbines, GeneralElectric, and ALSTOM Power. Dr. Brun isthe inventor of the Single Wheel RadialFlow Gas Turbine, the Semi-ActiveCompressor Plate Valve, and co-inventor ofthe Planetary Gear Mounted Auxiliary

Power Turbine. He has authored over 40 papers on turbomachinery,given numerous technical lectures and tutorials, and published atextbook on Gas Turbine Theory. He is the Chair of the ASME-IGTIOil & Gas Applications Committee, a member of the Gas TurbineUsers Symposium Advisory Committee, and a past member of theElectric Power and Coal-Gen Steering Committees.

Dr. Brun received his Ph.D. and M.Sc. degrees (Mechanical andAerospace Engineering, 1995, 1992) from the University ofVirginia, and a B.Sc. degree (Aerospace Engineering, 1990) fromthe University of Florida.

J. Jeffrey Moore is a Principal Engineerat Southwest Research Institute, in SanAntonio, Texas. His professional experienceover the last 15 years includes engineeringand management responsibilities atSolar Turbines, Inc., Dresser-Rand, andSouthwest Research Institute. His interestsinclude rotordynamics, seals and bearings,finite element analysis, controls, andaerodynamics. He has authored more than

10 technical papers in the area of rotordynamics and aerodynamicsand has given numerous tutorials and lectures.

Dr. Moore received his B.S., M.S., and Ph.D. degrees (MechanicalEngineering, 1991, 1993, 1999) from Texas A&M University.

ABSTRACT

This tutorial provides an overview of the latest edition ofAmerican Petroleum Institute (API) Code API 616 (1998) and alsoprovides a brief summary of API 614 (1999), 617 (2002), 670

(2000), 671 (1998), and 677 (1997) as they apply to gas turbinedriven compressors. Critical sections of the codes are discussed indetail with a special focus on their technical interpretation andrelevance for the purchasing scope-of-supply comparison, machinetesting, and field operation. Technical compliance issues forpackage, core engine, instrumentation, and driven compressor areaddressed individually. As API 616 (1998) forms the backbone formost oil and gas combustion turbine acquisitions, it is coveredin more detail. Some recommendations for acceptance ofmanufacturer exceptions to API and the technical/commercialimplications will be provided. A brief discussion of the relevanceNFPA 70 (2002) to gas turbine driven compressor sets in oil andgas applications will also be included. This tutorial course isintended for purchasing, operating, and engineering staff ofturbomachinery user companies.

INTRODUCTION

American Petroleum Institute (API) specifications are generallyapplied to oil and gas turbomachinery applications rather than tolarge industrial power generation. Oil and gas applications of gasturbines have requirements that are inherently different from thoseof the electric power industry. Namely, oil and gas applications,and customers require:

• High availability/reliability• Ruggedness• High power/weight ratio• Efficiency often not critical

While industrial power generation customers have differentcritical requirements, namely:

• Cost of electricity• Efficiency• Cost of operation and maintenance (O&M)

Because of these inherent market differences, oil and gas customersoften insist on compliance with API codes and are willing to acceptthe resultant higher turbomachinery package costs.

OIL AND GAS TURBOMACHINERY APPLICATIONS

Many applications within the oil and gas industry require theusage of turbomachinery equipment for compression, pumping,and generation of electricity. These applications are generally

145

API SPECIFICATION REVIEW FOR GAS TURBINE DRIVEN TURBOCOMPRESSORS

byKlaus Brun

Program Manager

andJeff Moore

Principal Engineer

Southwest Research Institute

San Antonio, Texas

divided into three separate areas from the production of oil to thesales of the refined product. These are upstream, midstream, anddownstream. Upstream refers to the oil and gas production andgathering process, midstream is the oil and gas transportationprocess, and downstream is the refining and distribution of oil andgas products. For each area and defintiion, different processes thatrequire turbomachinery are listed below.

Upstream

• Self-generation—Power generation to meet needs of oil fieldor platform• Enhanced oil recovery (EOR)—Advanced technologies to improveoil recovery• Gas lift—Injecting gas into the production well to help lift the oil• Waterflood—Injection of water into the reservoir to increasereservoir pressure and improve production• Gas reinjection—Reinjection of natural gas into the reservoir toincrease the reservoir pressure• Export compression—Initial boosting of natural gas pressurefrom field into pipeline (a.k.a. header compression)• Gas gathering—Collecting natural gas from multiple wells• Gas plant and gas boost—Processing of gas to pipeline quality(i.e., removal of sulfur, water, and CO2 components)• Gas storage/withdrawal—Injection of gas into undergroundstructure for later use: summer storage, winter withdrawal

Midstream

• Pipeline compression—Compression station on pipeline to“pump natural gas” typically 800 to 1200 psi compression• Oil pipeline pumping—Pumping of crude or refined oil• LNG plant—Liquefied natural gas (LNG) plant; NG is cooledand compressed for transportation in liquid form• LNG terminal—Loading and unloading of liquid natural gasonto LNG tanker (vessel)

Downstream

• Refinery—Processing of crude oil into its various sellablecomponents• Integrated gasification combined cycle—Advanced process toconvert heavy oils and pet-coke into synthesis gas and hydrogen• Methanol plant—Plant that produces methyl alcohol frommethane (natural gas)• Fischer-Tropsch liquid—Artificial gasoline produced from coaland/or other lower cost hydrocarbons• Fractional distillation column—Distillation tower to separatecrude oil into its gasoline, diesel, heating, oil, naphtha, etc.• Cracking—Breaking large hydrocarbon chains to smaller chains(chocker, visbreaking, thermal, catalytic)• Blending—Mixing of hydrocarbons to obtain sellable refineryproducts• Distribution—Delivery of oil and gas products to end-users

TURBOCOMPRESSOR COMPONENTS AND LAYOUT

Most major elements of a typical turbocompressor can beclassified as either inside the package or outside the package.Inside the package generally refers to all equipment inside the gasturbine enclosure, while outside the package refers to equipmentconnecting to the package, such as ancillaries and auxiliaries. Themain turbocompressor’s equipment, classified by this definition, islisted below.

Inside the Package

• Fuel system and spark igniter• Natural gas (control valves)• Liquid (pumps, valves)

• Bearing lube oil system • Tank (overhead, integral)

• Filter (simple, duplex)• Pumps (main, pre/post, backup)

• Accessory gear• Fire/gas detection system• Starter/helper drive

• Pneumatic, hydraulic, variable speed alternating current(AC) motor

• Controls and instrumentation (on-skid, off-skid)• Seal gas/oil system (compressors)

Outside the Package

• Enclosure and fire protection system• Inlet system

• Air-filter (self-cleaning, barrier, inertial, demister, screen)• Silencer

• Exhaust system• Silencer• Stack• Lube oil cooler (water, air)• Fuel filter/control valve skid• Motor control center• Switchgear, neutral ground resistor • Inlet fogger/cooler

Within API there are standards that address each one of theabove package elements; however, the focus of most APIspecifications is on packaging and ancillaries, rather than on gasturbine core components.

GAS TURBINE PACKAGE SYSTEMS

As turbocompressors are complex mechanical devices, theyrequire electronic and mechanic controls as well as instrumentations.The control system of a gas turbine driven compressor must, as aminimum, provide the following functions through the use of ahuman machine interface (HMI):

• Machinery monitoring and protection• Equipment startup, shutdown, and protective sequencing• Stable equipment operation• Alarm, shutdown logic• Backup (relay) shutdown

• Driven load regulation• Fuel/speed control• Process control• Surge control

• Communication (supervisory control and data acquisition,SCADA) interface

API sets minimum functionality, redundancy, alarms, andshutdown requirements of the HMI. The control system alsointerfaces gas turbine instrumentation for sensing and emergencyalarms/shutdowns. Typical instrumentation that is provided with aturbocompressor package can be divided into sensors for controlinstruments and alarm/shutdown instruments. Control sensingprovides feedback for supervisory control of the machine whilealarm/shutdown sensors (or switches) are primarily intended toprotect the machine from damaging failures. However, there are anumber of instruments on the package that serve both functions.For example, compressor discharge temperature measurements canbe used both for control and high temperature shutdown. A list oftypical instrumentation, classified as sensing and control versusalarm and shutdowns devices, is shown below.

Sensing and Control Instrumentation

• Temperature sensors (resistance temperature detectors [RTDs]and thermocouples)

• Lube oil drains• Lube oil cooler in/out


• Lube oil header• Fuel gas• Bearing metal (radial and axial thrust bearings)• Gas path (T1, T2, T4, T5-spread)• Driven compressor (Ts, Td)• Driven generator stator coils

• Pressure sensors• Gas path (P1, P2, P4)• Fuel gas• Air inlet filter Delta P• Lube oil header• Lube oil filter Delta P• Driven compressor (Ps, Pd)

• Flow sensors• Surge control (driven compressor)

• Position sensors (proximity probes)• Measures relative movement between housing and shaft• Gas turbine (GT) bearings X, Y, Z • Gearbox bearings X, Y, Z• Driven equipment X, Y, Z

• Velocity probes or accelerometers• Measures absolute motion • Gas turbine case• Main gearbox case• Auxiliary gearbox case• Driven equipment case

Critical Alarm and Shutdown Switches

• Lube oil level, temperature, and pressure• Lube/seal oil filter Delta P• Lube oil drain temperature• Casing vibration• Proximity probe vibration• Axial thrust position• Fuel gas temperature and pressure• Gas and fire detectors• Inlet filter Delta P• Magnetic speed pickup (shaft speed)• Exhaust temperature T-5 spread• Flame-out (flame detector)• Bearing metal temperatures• Compressor discharge pressure• Seal gas or oil pressure and temperature (driven compressor)• Seal vent leakage (dry gas seal)• Buffer air Delta P• Synchronization/grid frequency (driven generator)• Manual emergency shutdown (ESD) button

COMPRESSOR STATION LAYOUT

Most turbocompressors are installed in gas compressionapplications, most often along major gas pipelines. Sinceturbocompressor codes and standards sometimes affect equipmentoutside the package, the turbocompressor is tightly integrated intothe station process. The major equipment at a pipeline compressionstation is listed below:

• Turbocompressors and ancillaries/auxiliaries• Surge recycle loop with surge control valves• Gas coolers (intermediate, discharge)• Suction scrubbers, discharge scrubbers• Station isolation and bypass valves• Suction and discharge valves• Compressor loading and bypass valves• Unit and station control room• Fuel gas heaters• Fuel gas filter and control skid • SCADA system• Fire fighting water system

Fire fighting CO2 system• Emergency power supply (diesel generators, tanks)• Instrument and seal gas air compressor• Slug catcher, scrubbers, filters• Flare and/or blowdown• Station and unit flowmeters• Pig launcher• Gas treatment plant (dehydration, CO2 removal, sour gas treatment)

CODES AND STANDARDS

Many codes and standards are used by turbocompressormanufacturers and users to specify, select, and characterize theirequipment. Codes are generally used for the following purposes:

• Codes provide design guidelines and definitions• Example: Vibrations, materials, service connections, package

design• Intent to facilitate, manufacture, and procurement• No code “designs” a gas turbine• Codes are by nature rather general. Any particular applicationmay require modifications.

For turbomachinery applications in the oil and gas sector, themost commonly used standards are:

• API 616 (1998)—Gas turbines • API 617 (2002)—Centrifugal compressors• API 614 (1999)—Lube oil system• API 670 (2000)—Machinery protection• API 613 (2003)—Load and accessory gear• API 677 (1997)—Accessory drive gear• API 671 (1998)—Flexible couplings• NFPA 70 (2002)—Electric code• ASME PTC-22 (1997)—Gas turbine testing• ASME PTC-10 (1997)—Compressor testing• ASME B133—Gas turbines• API RP 686 (1996)—Machinery installation

Some other international codes that are also applicable are:

• ISO codes (ISO 3977, 1995-2002—Gas turbines)• IEC/CENELEC (electrical and fire systems)• Local government codes• European Union environmental and health compliance• User specific specifications

• Upstream—Most companies follow API specifications.• Midstream—Some companies utilize API, but interpretation

is generally handled with more flexibility.• Downstream—API is a critical specification for most

downstream process applications.

Of the above codes, API 616 (1998) and 617 (2002) are the mostcritical codes when evaluating or purchasing a turbocompressorpackage. However, API 614 (1999) and 670 (2000) are stronglyreferenced in both API 616 (1998) and 617 (2002) and should, thus,also be critically reviewed when purchasing a turbocompressor.

WHAT IS API?

The American Petroleum Institute is the primary trade organizationfor the U.S. petroleum industry. API has over 400 membercompanies that cover all aspects of the oil and gas production.API is an accredited American National StandardsOrganization (ANSI) and started developing industry specificcodes in 1924. Currently, API publishes about 500 standards,which are widely referenced by the Environmental ProtectionAgency ( EPA), Occupational Safety & Health Administration(OSHA), Bureau of Land Management (BLM), AmericanSociety of Mechanical Engineers (ASME), and other codes and

API SPECIFICATION REVIEW FOR GAS TURBINE DRIVEN TURBOCOMPRESSORS 147

•

regulations. API’s philosophy for developing codes is based onthe following principals:

• Improve safety• Improve environmental performance• Reduce engineering costs• Improve equipment interchangeability• Improve product quality• Lower equipment cost• Allow for exceptions within reason

However, as with most other codes, API specifications often lagtechnology developments, especially in the rapidly changing gasturbine and compressor markets.

API STANDARDS AS PURCHASING SPECIFICATIONS

API standards are often used as a convenient purchasing document.They provide the means for a customer to normalize the quotations byforcing all machinery vendors to quote on similar scopes. API codesalso provide a common language and set of rules between vendor andcustomer to limit misunderstandings (e.g., definitions of efficiencies,test procedures, vendor data, data sheets).

All API codes clearly state in their foreword that exceptions areallowed, if they lead to an improved or safer technical offer. Forgas turbine applications in the oil and gas sector, API 616 (1998) isthe foundation for almost all purchase specifications. API 617(2002) is used for most centrifugal compressor applications. Also,API references National Fire Protection Agency (NFPA) NFPA 70(2002) electric code for hazardous locations. This specification isfundamental in most machinery standards and is, thus, a criticalrequirement for oil and gas applications. Namely, oil and gasturbocompressors must at least meet Class 1, Division 2 (Zone 2),Group D. Some Applications require Division 1 (Zone 1).

API STANDARD TOPICS

The API 616 (1998) and 617 (2002) codes can be divided by aset of standard topics. These are:

• Definitions• ISO rating, normal operating point, maximum continuous

speed, trip speed, etc.• Mechanical integrity

• Blade natural frequencies, critical speeds vibration levels,balancing requirements, alarms and shutdowns

• Design requirements and features• Materials, welding, accessories, controls, instrumentation,

inlet/exhaust systems, fuel systems• Inspection, testing, and preparation for shipment

• Minimum testing, inspection, and certification • Documentation and drawing requirements

Often overlooked, but critically important are the API data sheetsas they clearly form the technical basis of a typical proposal. Withinthese data sheets, the application specific issues are addressed, suchas customer site, operating conditions, basic equipment selection,and equipment minimum integrity requirements. Thus, for apurchaser of a turbocompressor it is important to always fill out (asa minimum) the data sheets for API 616 (1998, gas turbine), API617 (2002, compressor), API 614 Appendix D (1999, lube oilsystem), and API 670 Appendix A (2000, machinery protection).

When filling out these data sheets a couple of industry acceptednorms should be remembered:

• Cross-out requirements that are not required or cannot becomplied with.• Include notes for critical technical comments. By including themon the data sheets these comments are elevated to a contractualtechnical requirement.

Fill out as much info as is available—even a partially filled outsheet is better than no sheet.

API 616—GAS TURBINE

The following section describes a brief review of some criticalpoints of API 616 (1998) with engineering opinions, interpretations,and comments (in italics). Some common manufacturer exceptionsare also discussed. This discussion (in italics) presents only anengineering opinion of the authors, and the opinions are clearlydebatable and open to other interpretations. A careful review of APIspecification applicability should be performed for every gasturbine purchase based on the specifics of the application.

The reader should note that API 616 (1998) does not apply toaeroderivative gas turbines (i.e., it is intended for industrial typegas turbines only).

Section 1.0—Alternative Designs

Section 1.0 reemphasizes that alternative designs are allowed.This is also briefly discussed in the API 616 (1998) Foreword. It isimportant to note that when alternative designs are proposed, themanufacturer must explicitly state the deviation and explain whythe alternative design is superior to the API standard. Theseexplanations should focus on safety, reliability, and efficiency ofthe equipment, while cost is generally not an acceptablejustification for an API deviation.

Section 2.0—References

All reference standards included in this section are automaticallyincluded in the standard. As this is difficult to review and beaccountable for all referenced specifications, all manufacturers ofmachinery take general exception to this requirement.

Section 3.0—Definitions

Section 3.0 provides a valuable discussion of commonnomenclature and basic definitions. Some of the more criticalitems in this section are:

• 3.17—ISO conditions are defined as T = 15C, P = 1.0133 bar,RH = 60 percent.• 3.19—Maximum allowable speed: speed at which unit can safelyoperate continuously per manufacturer• 3.22—Maximum continuous speed: 105 percent highestdesign speed• 3.26—Normal operating point: Usually the performanceguarantee (heat rate, power) point for a gas turbine. This is definedby speed, fuel, and site conditions.• 3.38—Rated speed: power turbine speed at which site ratedpower is achieved• 3.42—Site rated conditions: The worst site condition at whichstill site rated power can be achieved. This must be provided by theuser as a requirement.• 3.45—Site rated power: The maximum power achievable at siterated conditions. This must be provided by the manufacturer. Note:Unless specifically stated by purchaser, this is not a guaranteepoint. Thus, it is recommended that the user explicitly states thatthis must be a guarantee point.• 3.50—Turbine trip speed: speed where controls shut unit downfrom the fuel gas supply to the gas turbine• 3.51—Unit responsibility: Prime package contractor that hascontractual responsibility to the purchaser. Either GT orcompressor manufacturer can be prime, but in more complexprojects the compressor vendor is usually selected prime (gasturbine is off-the-shelf—compressor is customized). As thecompressor is usually only about 30 percent of the cost of theturbocompressor package, this arrangement often leads tocommercial complexities and project financing requirements forthe compressor vendor.


•

Section 4.0—Basic Design

This section in API 616 (1998) covers the basic design of the gasturbine itself. Included in this section are pressure casingrequirements, combustor design, casing connections, shafts,bearings, seals, dynamic requirements, and material qualitystandards. This section is very comprehensive. However, as mostgas turbine manufacturers are generally unwilling to makecustomized changes to their core engine, this is also the section inwhich most critical exceptions can be expected. A number ofrelevant items of this section are discussed below:

• 4.1.1—Equipment must be designed for 20 years and three yearsof uninterrupted service. Hot section inspection must be performedevery 8000 hours. Many gas turbines on the market cannot reachthree years of uninterrupted service and/or require combustioninspections at 4000 hours. However, as this is a design requirement,vendor interpretations of this vary widely.• 4.1.2—The gas turbine vendor has unit responsibility unlessotherwise specified.• 4.1.5—Speed range operating requirements are: Two-shaft gasturbines 50 to 105 percent rated speed, single-shaft gas turbines 80to 105 percent rated speed. This refers to the output shaft only.• 4.1.14—On-skid electric must meet NEC NFPA 70 (2002). Thisis a critical requirement that generally applies throughout all oiland gas plants and has many significant packaging implications,which will be discussed later.• 4.1.16—Most gas turbine internals shall be exchangeable at site.This requirement disqualifies most aeroderivative gas turbines asthey are usually overhauled by core-engine exchange only.• 4.1.17—Unit shall meet performance acceptance criteria, both inthe factory and at site. Most manufacturers take exception to siteperformance guarantee unless a field test is specified andperformed.• 4.1.18—Vendor shall review customer’s installation drawings.However, this does not imply that vendor certifies or warrants thecustomer’s installation design.• 4.1.21—Gas turbine must meet site rated power with no negativetolerances. This does allow for additional test uncertainties.• 4.2.1—Casing hoop stresses must meet ASME Section VIII(2004). A standard requirement throughout API. This defines hydrotest procedures.• 4.2.7—Openings for borescope inspection must be provided forentire rotor without disassembly. Borescope inspection of the firstcompressor stages can be through the inlet and power turbinethrough the exhaust. Few manufacturers provide borescope pointsthat allow access to all compressor stages.• 4.2.9—Field balancing (if required) without removal of casing isacceptable.• 4.3—Combustor design: As this section aims to apply to all typesof combustors (can, can-annular, annular, dry low emission[DLE], diffusion) it is somewhat general.• 4.3.1—Combustors must have two igniters or igniters and cross-fire tubes (in each combustor for multiple cans). This requirementis not clear for annular combustors and most can design combustorshave a single ignitor in each can but do have cross-fire tubes.• 4.3.2—A temperature sensor is required in each combustor. Thisis not feasible in most gas turbine designs, as a thermocouple willnot survive extended periods of operation at gas turbine firingtemperatures. Most manufacturers provide T4 (gas generatorturbine outlet) for two-shaft engines and/or T5 (exhaust temperature)for single-shaft engines. Firing temperature (T3) is usuallycalculated and displayed on HMI.• 4.3.7—Fuel Wobbe Index range must be indicated in theproposal. Most small gas turbines can handle 610 percentvariation, even with DLE combustion.• 4.4—Casing connections: Section outlines good designpractices. Most manufacturers have few exceptions in this sectionor have reasonable justifications in their deviations.

4.5.1.2—Shafts shall be single piece heat-treated steel. Stackedrotors with a tie-bolt are not allowed. This is difficult to meet formost manufacturers and is not critical if a stacked rotor has beenoperationally proven and meets API dynamic requirements.Stacked rotors also have significant advantages for repair andmaintenance.• 4.5.2.1—Rotor shall be designed for overspeed up to 110 percentof trip speed. This does not imply that the rotor must be tested tothis speed.• 4.5.2.2—All rotor components must withstand instantaneousloss of 100 percent shaft load. Namely, coupling failure should notlead to catastrophic power turbine failure.• 4.6—Seals: Renewable seals are required at all close clearancepoints. Some manufacturers utilize internal lip-seals. Thisrequirement makes the most sense for interstage seals.• 4.7—Rotordynamics: This extensive section describes minimumrotordynamic requirements and basic operational definitions.• 4.7.1—Critical speeds: When the frequency of a periodic forcingphenomenon corresponds to a natural frequency of that system, thesystem may be in a state of resonance. When the amplificationfactor is greater than or equal to 2.5, that frequency is called acritical speed. Most gas turbines operate above their first criticalspeed. API provides required separation margin between theoperating speed range and the critical speeds. The most commonsource of excitation is unbalance but other sources include rubs,blade pass, gear mesh, and acoustic. The flexibility and potentialresonances of the supporting structure should be considered,especially for three-point mount baseplates.• 4.7.2—Lateral analysis: This section recognizes that gas turbinesare designed as standard products. Once the design is established,no modifications should be required on a particular application.An exception to this is if the drive coupling mass is significantlydifferent from the design coupling. A new lateral analysis is onlyrequired on new prototype engines.• 4.7.3—Torsional analysis: Torsional analysis is required, butmany manufacturers take exception to this for gas turbine drives.API requires that intersections between the torsional naturalfrequencies and one-times (1◊) the running speed must beseparated by at least 10 percent. API also states that intersectionsat two and higher orders of running speed shall preferably beavoided, but most manufacturers take exception to this.• 4.7.4—Vibration and balancing: A progressive balancingprocedure where no more than two components are added to therotor at a time between corrections is required by API. However,since most gas turbines’rotors are built-up designs with tie-bolts,this procedure is not possible. For these built-up rotors,component balancing should be performed and careful attentionto rotor-runout should be made after assembly. Blade masssorting routines help to minimize the amount of correctionrequired. Sufficient speed during low speed balance should bemade to ensure that the blades are seated in the roots orfur-trees. During assembly balance, one should distribute thecorrection planes to several locations along the rotor, since theexact location of unbalance is not known. A residual unbalancecheck is encouraged by API, but given the resolution of modernbalance machines, this check is not required. High speedbalancing of the entire rotor is acceptable but is discouraged.Clearly, rotors that require high speed balancing to achievevibration limits during test will likely require field balancing. Therequired API balance limits are reasonable and should beadhered to on both the component and assembly level. Careshould be made to minimize electromechanical runout by theproximity probes. API permits subtraction of the run-vector fromthe vibration level at running speed, but few customers acceptthis practice.• 4.8—Bearings: applies to all gas turbine bearings.

• 4.8.1.1—Hydrodynamic radial and thrust bearings are preferred.These should be thrust-tilt pad, radial-tilt pad, or sleeve bearings.


•

4.8.2.5—If rolling element bearings are used they must meet50,000 hours of continuous operation. Few industrial gas turbinesutilize rolling element bearings and aeroderivative engines are notapplicable to API 616 (1998).• 4.8.3.3—The bearing shells shall be horizontally split. Manyoriginal equipment manufacturers (OEMs) take exception to thisrequirement.• 4.8.4.2—Hydrodynamic thrust bearings shall be selected at nomore than 50 percent of the ultimate load rating at site power. Thisrequirement should not be taken exception to.• 4.8.5.2—Bearing housings: Replaceable labyrinth type bufferseal required. Lip-seals are not acceptable—this cannot be met bysome manufacturers.• 4.8.5.3—Two radial proximity probes in all radial bearings (X, Y),two axial proximity probes in the thrust bearing, and a key phasorshould be provided (when space allows). Many manufacturers willonly provide a single thrust bearing proximity probe.• 4.9 —Lubrication: specifies lubrication requirements and refersto API 614 (1999).• 4.9.2—Synthetic lube oil is acceptable in the lube oil system.However, most users prefer mineral oils for cost reasons.• 4.9.5—Integrated or separate lube GT/compressor oil systemsare both acceptable.• 4.9.7—Lube oil system must meet API 614 (1999). Most moderngas turbines have the lube oil tank integrated into the skid but anoverhead tank is also acceptable.• 4.10—Materials: This section describes basic material, casting,welding, and forging requirements and essentially states that gooddesign rules should be followed. Low grade carbon steel is notpermitted. All materials of construction must be stated in thevendor proposal (and most manufacturers do not comply with this).Purchaser may require 100 percent radiography, magnetic particleinspection, and/or liquid penetrant inspection on all welds.• 4.11.3—A stainless steel (SS) nameplate to include ratedperformance and conditions, critical speeds, maximum continuousspeed, overspeed trip, and fuel type. Few manufacturers providethis relatively simple and valuable requirement adequately, unlessthe purchaser insists.

Section 5.0—Accessories

Section 5 deals with all accessories that are required to operate a gasturbine (i.e., mostly package items, instrumentation, controls, andinlet/exhaust systems). Section 5 also includes critical sections on gasand liquid fuel systems. Some of the main points of this section are:

• 5.1—Starting and helper drivers: Starter motors decouple fromthe turbine shaft after startup while helper motors stay connected.• 5.1.1.1—Electric, pneumatic, electrohydraulic, gas engine, andsteam turbine are all acceptable starter/helper motors. Mostmodern gas turbines utilize AC starter motor with variablefrequency drives (VFD).• 5.1.1.7—Starter motors must be supplied with all necessarygears, couplings, and clutches.• 5.1.2.1—Starters shall be rated 110 percent of starting torquerequired. Vendor to supply torque curves. Most AC starter motorsare severely underrated for the application and overheat aftermultiple starts.• 5.1.3—Turning gear: Device to slowly rotate to avoid shaftdeformation. Must be separate device from starter motor. This isgenerally not required on smaller turbines, and, if required, mostmanufacturers utilize the starter motor.• 5.2.1—Vendor to supply accessory gear for starting, auxiliaryequipment (e.g., lubricating oil pump). Vendor to also supplyseparate load gear, if required. All gears shall meet API 613 (2003).Accessory gears shall be rated for 110 percent of required load.Most gas turbine lube oil systems utilize engine driven main pump,undersized AC pre-post pump, and undersized direct current (DC)backup pump.

5.2.2—Vendor to supply all couplings and guards for load andaccessory shafts. Couplings to be sized for maximum continuoustorque and meet API 671 (1998).• 5.3—Gas turbine to be supplied on single mounting plate(baseplate). This is typically interpreted to allow for skid mounting.Skid to be designed to limit worst-case shaft alignment change to50 micrometers. For some offshore applications this is difficult tomeet without gimbled three-point mount. Supporting the baseplateat three pounds can result in baseplate vibration modes in theoperating speed range of the GT or driven equipment. Thisfoundation flexibility may also adversely affect the critical speeds.Careful finite element analysis should be performed, although it isnot required by API 616 (1998).• 5.3.2—A single piece baseplate is preferred. The baseplate shallextend under driven equipment. As the compressor and gas turbineare often supplied by different vendors, this requirement is oftennot practical; tightly bolted gas turbine and compressor skids arecommon and generally function well when properly engineered.• 5.3.2.4—Skid to permit field leveling.• 5.4.1.2—All controls and instrumentation shall be suitable foroutdoor installation. This is usually not necessary or practical forinside enclosure devices and devices that are in a control room.• 5.4.1.3—Gas turbine instruments and controls shall meet API670 (2000).• 5.4.1.4—GT control system must provide for safe startup,operation, and shutdown of gas turbine. There is an implicitguarantee in this statement that the gas turbine mustfunction/operate properly.• 5.4.1.6—Unit shall continue to operate for specified time afterAC failure. This is only possible with redundant control system andnot practical for some applications as safe relay shutdown onbackup batteries is preferred (i.e., if the unit operation is notcritically important, it is often safer from a system perspective toshut down on an AC failure than to continue operating).• 5.4.2.1—Automatic start to require only single operator action.Although most modern control systems have a single buttonstart-stop operation, in reality the operator has to usuallyacknowledge and reset a host of alarms during the startup process.• 5.4.2.2—Control system to completely purge unit prior tostartup. This is a critical safety requirement to avoid explosions ina restart attempt. A purge crank of five to 10 minutes is typical.• 5.4.3.4—Load control to limit driven equipment speed to 105percent rated speed.• 5.4.4.1—Alarm/shutdown system to be provided to protect unitand operator. Both normal and emergency (fast) shutdown (i.e., cutoff fuel supply) must be available.• 5.4.4.3—Fuel control must include separate shutoff and ventvalves, separate from fuel control valve, with local and remote trip.A control valve that also acts as a shutdown valve is not permittedand should not be accepted by purchasers for safety reasons.• 5.4.4.8.1—Alarm and shutdown switches shall be separate. Thisis often not practical for gas turbine internal sensors (e.g., flamedetector, axial thrust position).• 5.4.4.8.3—Alarm and trip settings should not adjustablefrom outside housing. This is not relevant with modern controlsystems. Most alarm and shutdown levels are set from the unitcontrol system.• 5.4.4.9—All instruments (other than shutdown switches) shallbe replaceable without shutting unit down. This is often notpractical on alarm/shutdown switches from a single sensor (referto 5.4.4.8.1).• 5.4.5.1.1—Off-skid or on-skid control system to be suppliedwith unit. As most oil and gas facilities have a control roomavailable that does not have NFPA 70 (2002) hazardous arearequirements, it is more common to utilize off-skid control system.• 5.4.5.1.1—Any on-skid control system must meet hazardousarea requirements (NFPA 70, 2002).• 5.4.6.2—On-skid electric systems must meet NFPA 70 (2002).


• •

5.4.7—Instrumentation: The API 616 (1998) data sheets must befilled out by purchaser to indicate the required instrumentation.Manufacturers will often cross out lines to indicate noncompliance.The vendor must include in his proposal a full list of instrumentation.API 670 (2000) governs basic requirements for temperature andvibration sensors.• 5.5.3.1—Air inlet and exhaust system including inlet filter,inlet/exhaust silencers, ducting, and joints must be provided byvendor. As this is application specific, some vendors will takeexception to this.• 5.5.3.3—Inlet system shall be designed for maximum 4 inchH2O pressure drop at site rated power. In very humid or moistenvironments, it is difficult to meet this requirement with wetbarrier filters, especially during startup.• 5.5.3.4—Inlet/exhaust system design life is 20 years. For marineenvironments this requirement implies the use of SS, which is veryexpensive.• 5.4.2.2—Bolts, rivets, or fasteners are not permitted in inlet system.• 5.5.3.9—A gas turbine compressor cleaning system must beprovided. This is quoted as an option by most manufacturers.• 5.5.4.4—Inlet filter system requires walkways and handrails.This is not necessary for very small gas turbine units.• 5.5.5—Inlet/exhaust silencer: Internals to be SS. External can becarbon steel.• 5.7.3—Unit must be supplied with gas detection system, firedetection system, and fire suppression system to meet NFPA. Firesuppression shall be automatic based on thermal detection. Typicalsystems are: Thermal, thermal gradient, lower explosive limit(LEL), and flash detector. CO2 and water mist are typically utilizedfor suppression. Water is safer for personnel but also requires moremaintenance.• 5.7.5—When enclosure is supplied it must be weatherproof andinclude fire detection/suppression, ventilation/purging, lights, anddoors and windows.• 5.8.1.2—Gas fuel system: Must include strainer, instrumentation,manifolds, nozzles, control valve, shutoff valve, pressure regulator,and vent valve. A strainer is often not supplied. Some vendorsprovide a cartridge filter instead. If a strainer or filter is used, theyshould be supplied with a differential pressure sensor.• 5.8.1.3—Fuel gas piping must be stainless steel. This is criticallyimportant, especially in sour gas applications.• 5.8.1.4—Liquid fuel system: Must include pump, atomizing air,two shutoff valves, instrumentation, control valves, flowdividers, nozzles, and manifolds. Newer systems usually employ avariable speed pump instead of a positive displacement pump witha control valve.• 5.81.5.2—If dual fuel system is provided it must allow bumplessbidirectional transfer. All gas turbine vendors require a reductionin load during fuel transfer.• 5.8.2.2.2—Vendor must review customer’s fuel supply system.This does not imply a vendor certification of guarantee.• 5.8.2.4—Heating value of fuel cannot vary by more than610 percent. This often protects the manufacturer againstwarranty claims.• 5.8.4.2—Purchaser must specify required site emissions levelsfor NOx, CO, UHC, et al., to manufacturer (i.e., it is thepurchaser’s responsibility to assure that the gas turbines meetenvironmental codes. The vendor only confirms [or not] therequired emissions values).

Section 6.0—Inspection, Testing, and Preparation for Shipment

In this section basic inspection, testing and shippingrequirements are defined. Some important issues in thissection are:

• 6.3.2.3— Hydrostatic tests: Vessels and piping must be perASME Section VIII (2004).• 6.3.3—Mechanical running test: A required four-hour no load,

full speed (maximum continuous speed) test to verify that thecomplete gas turbine package (including all auxiliaries except forinlet/exhaust system) operates within vibration and operationalcontrol limits. Contract coupling should be used. Rotordynamicsignature and vibrations must be recorded. This is a basicmechanical integrity test, and compliance with these requirementsis critical.• 6.3.4—Optional tests: These tests are not optional if purchasermarks them in the API 616 (1998) data sheets.• 6.3.4.1—Optional performance test: unit full load tested toASME PTC 22 (1997) (discussed later).• 6.3.4.2—Optional complete unit test: Similar to mechanical runtest but to include driven equipment. This test is often combinedwith a performance test and is called a full load string test.• 6.3.4.3—Gear test: Gear must be tested with unit duringmechanical run test.• 6.3.4—Other optional test: sound level, auxiliary equipment, fireprotection, control response, spare parts.• 6.3.4.6—If unit fails mechanical run test, complete disassemblyand reassembly are required.• 6.4—Preparation for shipment: Short-term shipping is just aplastic wrapping or tarp. Long-term shipping requires crating,surface anticorrosion treatment, inert gas (nitrogen) fill of allvessels including GT casing. Generally, long-term shippingpreparation is recommended as there are often delays in thecommissioning of a new plant, which may lead to extended storagetimes of the equipment.

Section 7.0—Vendor’s Data

In this section the minimum documentation requirementsfrom the proposal to as-built drawings are defined. The detaileddocument requirements are listed in the API 616 (1998) datasheets. Most manufacturers take exception to specific documentrequirements by crossing out lines in the API data sheets. As aminimum, the purchaser should insist on performance maps,performance calculations, mechanical, hydraulic, layout, andelectrical drawings, process and instruments diagram(P&IDs), test and inspection results, as-built parts lists,operation and maintenance manuals, installation manual, andtechnical data.

API 614—HIGHLIGHTS

API 616 (1998) refers to API 614 (1999) for the lubricating oilsystem. API 614 (1999) is a generic specification for all lubricat-ing oil (LO) systems, but a couple of important requirements andissues should be emphasized:

• LO system must be design for 20 years service life. This impliesall SS for marine applications.• A single full capacity pump with a full size backup pump is required.Most GT manufacturers provide smaller sized backup pumps that onlyallow for a safe shutdown but not continued operation.• Duplex oil filters with smooth transfer are required (i.e., noshutdown should be required when replacing filter cartridges).• All wetted surfaces must be stainless steel. Most manufacturerswill provide a coated carbon steel LO tank integrated into the skid.• Lube oil tank must have minimum of eight minutes retentiontime. Manufacturers with integrated skid LO tanks generallycannot meet this requirement, especially if a combined compressorLO system with wet seals is utilized.• All pumps, valves, switches, and sensors to have block andbleed valves.• Sight indicators and levels sensors for are required for tanks.• A skid integrated lube oil tank is acceptable; an overhead gravity feedtank is preferred. This provides continues flow in case of an electricalfailure and sometimes eliminates the need for a DC backup pump.• Tank must have vent with flame trap, and an exhaust fan ispreferred to avoid the flammable vapor gas mixture accumulation.


•

API 670 HIGHLIGHTS

API 670 (2000) covers machinery protection and control systemsuch as alarm and shutdown switches. It is beyond the scope of thispaper to cover API 670 (2000) but a couple of important items arelisted below:

• API 670 (2000) provides minimum design, installation, andaccuracy standards for package alarms/shutdown switches andsensors. It covers all instruments specified in the API 616 (1998)data sheets.• All instrumentation and wiring must meet NFPA 70 (2002)hazardous area classifications.• Signal and power wiring must be separate.• Alarms and shutdown switches must be separate and separatestainless steel housings. This is often not practical with modernsensors and control systems.• Bearings must have metal temperature sensors. Mostmanufacturers only offer this as an optional feature.• Axial thrust bearings must have two proximity probes. Only oneproximity probe is usually provided.

FACTORY PERFORMANCE TESTS

There are a number of tests that are performed on a gas turbinedriven compressor in the factory. The gas turbine performance testspecification most commonly used is ASME Performance TestCode 22 (ASME PTC-22, 1997) or manufacturer specificderivatives of it. For the driven compressor, API requires only amechanical test run, but most operators insist on a closed loop testalso to characterize the performance. Closed loop tests are usuallyperformed per ASME PTC-10 (1997).

PTC-10 Compressor

• Closed loop test to determine performance of compressor• Can also identify aerodynamic stability issues if tested tofull load• Type I: actual gas and full pressure; Type II: simulated gas andreduced pressure• Type I typically used for high pressure/high energy applications• Reduces risk of startup delays due to vibrations or lack ofperformance

PTC-22 Gas Turbine

• Full speed, full load test for four hours• Typically against a water break or generator/load cells• Determines maximum output power, specific fuel consumption,and efficiency

Some operators opt to perform field testing instead ofextensive factory testing. The risk of this test is that if problemsare identified in the field it may be difficult to easily correctthem. However, if testing is intended to only verify a very tightperformance guarantee, field testing is typically adequate.

NATIONAL FIRE PROTECTION ASSOCIATION (NFPA) 70

API specifications refer to the NFPA 70 (2002) code forelectrical wiring and safety requirement. To identify the properclassification of a subject, NFPA 70 (2002) provides thefollowing guidelines.

Hazardous Location Summary Electrical Requirements

• Class 1—Flammable gases, vapors, or liquids• Class 2—Dust and combustible dust that can form explosivemixtures• Class 3—Fibers or flyings suspended in air that are easilyignitable

Divisions Within Class 1

• Class 1, Division 1• (Zone 0 and Zone 1)—Ignitable concentration of flammable

mixtures exist most of the time.• Class 1, Division 2

• (Zone 2)—Ignitable concentrations of flammable mixturesare not likely to exist under normal conditions.

Most oil and gas applications require Class 1, Division 2. Someoffshore and refinery applications require Class 1, Division 1.NFPA 70 (2002) applies principally to the following turbocompressorpackage items:

• On-skid electrical• Wiring and cables• Electric motors (pumps, fans, starter)• Instruments and alarms• Junction boxes• Lights and heaters (if applicable)

• On-skid fire systems• Ultraviolet/infrared (UV/IR) flash sensors• Flammable mixture detectors• Temperature, temperature rise detectors• Smoke detectors (if used)• CO2 (or other) fire fighting system

• Air inlet and lube oil electrical elements

NFPA 70 (2002) does not apply to nonelectrical components andexception to NFPA 70 (2002) can often be taken for:

• Off-skid control system• Battery charger and batteries• Starter motor variable frequency drive• Synchronization panel

There are many requirements in NFPA 70 (2002) for Class 1,Divisions 1 and 2 (details are in NFPA 70, 2002, Sections 500 and501), but some of the most critical requirements are:

• Grounded explosion-proof conduit for all wiring• Conduit and conduit connections are copper free• Gas sealed conduit connections• Instruments, connectors, terminals in explosion-proof boxes• Motors rated explosion-proof• Fire detection system (UV/IR, thermal gradient, flash detectors)• Gas (flammable mixture) detectors (LEL)• Fire fighting system (inert gas or water mist)

API 617—Axial and Centrifugal Compressors and Expander-Compressors for Petroleum, Chemical, and Gas Industry Services

API 617 (2002) is a code that is applied to the centrifugalcompressor used in GT packages. This specification is prevalent foroffshore and refinery applications. It is usually less prevalent inpipeline applications. API 617 (2002) is often referenced withinuser-company specific purchase specifications, but all manufacturershave various amounts of comments and exceptions. API 617 (2002)has similar technical definitions as API 616 (1998), and these are,therefore, not repeated here. However, in general API 617 (2002)covers the following equipment:

• Centrifugal and axial compressors• Integrally geared compressors• Expander compressors

SCOPE OF API 617

API 617 (2002) covers the minimum requirements for axial andcentrifugal compressors, single-shaft and integrally geared process


centrifugal compressors and expander-compressors for use in thepetroleum, chemical, and gas industry services that handle air orgas. This standard does not apply to fans (covered by API 673,2002) or blowers that develop less than 34 kPa (5 psi) pressure riseabove atmospheric pressure. This standard also does not apply topackaged, integrally-geared centrifugal plant, and instrumentair compressors, which are covered by API 672 (1996).Furthermore, hot gas expanders over 300ºC (570ºF) are notcovered in this standard.

As with all other API codes, the equipment vendor mayoffer alternative designs if these designs improve the safety orperformance of the equipment. Otherwise all designs shouldcomply with this standard. If exceptions to the standard are taken,they must be clearly stated in the proposal.

SUMMARY

This tutorial provides an overview of the applicable codesrelated to typical industrial gas turbine packages, including API616 and the supporting API Standards 614 (1999), 617 (2002), 670(2000), 671 (1998), and 677 (1997), as well as NFPA 70 (2002). Asthe authors have presented, one size does not fit all in the selectionand procurement of gas turbines and many exceptions to API aremade. However, the API standards represent sound engineeringpractice based on many years of experience. Therefore, exceptionsto these standards should be kept to a reasonable minimum.

REFERENCES

API Standard 613, 2003, “Special-Purpose Gear Units forPetroleum, Chemical and Gas Industry Services,” FifthEdition, American Petroleum Institute, Washington, D.C.

API Standard 614, 1999, “Lubrication Shaft-Sealing and Control-Oil Systems for Special-Purpose Applications,” FourthEdition, American Petroleum Institute, Washington, D.C.

API Standard 616, 1998, “Gas Turbines for Refinery Service,”Fourth Edition, American Petroleum Institute, Washington, D.C.


API Standard 670, 2000, “Vibration, Axial-Position, and Bearing-Temperature Monitoring Systems,” Fourth Edition, AmericanPetroleum Institute, Washington, D.C.

API Standard 671, 1998, “Special Purpose Couplings forPetroleum, Chemical, and Gas Industry Services,” ThirdEdition, American Petroleum Institute, Washington, D.C.

API Standard 672, 1996, “Packaged, Integrally Geared CentrifugalAir Compressors for Petroleum, Chemical, and Gas IndustryServices,” Third Edition, American Petroleum Institute,Washington, D.C.

API Standard 673, 2002, “Centrifugal Fans for Petroleum,Chemical, and Gas Industry Services,” Second Edition,American Petroleum Institute, Washington, D.C.

API Standard 677, 1997, “General-Purpose Gear Units forPetroleum, Chemical, and Gas Industry Services,” SecondEdition, American Petroleum Institute, Washington, D.C.

API Standard 686, 1996, “Machinery Installation and InstallationDesign,” American Petroleum Institute, Washington, D.C.

ASME B133, A Series of Standards for Gas Turbines, AmericanSociety of Mechanical Engineers, New York, New York.

ASME PTC-10, 1997, “Performance Test Code on Compressorsand Exhausters,” American Society of Mechanical Engineers,New York, New York.

ASME PTC-22, 1997, “Gas Turbine Power Plants,” AmericanSociety of Mechanical Engineers, New York, New York.

ASME Boiler & Pressure Vessel Code - Section VIII - PressureVessels, 2004, American Society of Mechanical Engineers,New York, New York.

ISO 3977, 1995-2002, “Gas Turbines—Procurement,” A Series ofNine Amendments, International Organization forStandardization, Geneva, Switzerland.

NFPA 70, 2002, National Electrical Code, National Fire ProtectionAgency, Quincy, Massachusetts.


John A. Kocur, Jr., is a MachineryEngineer in the Plant Engineering Divisionat ExxonMobil Research & Engineering,in Fairfax, Virginia. He has worked in theturbomachinery field for 20 years. Heprovides support to the downstreambusiness within ExxonMobil with expertiseon vibrations, rotor/aerodynamics, andhealth monitoring of rotating equipment.Prior to joining EMRE, he held the position

of Manager of Product Engineering and Testing at SiemensDemag Delaval Turbomachinery. There Dr. Kocur directed thedevelopment, research, engineering, and testing of compressor andsteam turbine product lines.Dr. Kocur received his BSME (1978), MSME (1982), and Ph.D.

(1991) from the University of Virginia and an MBA (1981) fromTulane University. He has authored papers on rotor instabilityand bearing dynamics, lectured on hydrostatic bearings, has beena committee chairman for NASA Lewis, and is a member ofASME. Currently, he holds postions of API 617 Vice-Chair andAPI 684 Co-Chair.

John C. Nicholas is owner and Presidentof Rotating Machinery Technology,Incorporated, in Wellsville, New York.His company repairs and servicesturbomachinery, and manufactures bearingsand seals. Dr. Nicholas has worked in theturbomachinery industry for 30 years inthe rotor and bearing dynamics areas,including five years at Ingersoll-Randand five years as the Supervisor of the

Rotordynamics Group at the SteamTurbine Division of Dresser-Rand.Dr. Nicholas, a member of ASME, STLE, and the Vibration

Institute, has authored over 40 technical papers concerningrotordynamics and tilting pad journal bearing design andapplication. He received his B.S. degree from the University ofPittsburgh (Mechanical Engineering, 1968) and his Ph.D. degree

from the University of Virginia (1977) in rotor and bearingdynamics. Dr. Nicholas holds several patents including one for aspray-bar blocker design for tilting pad journal bearings andanother concerning bypass cooling technology for journal andthrust bearings.

Chester C. Lee is the Group Manager of Rotordynamics Designand Test in the Gas Compressor Engineering Department of SolarTurbines Inc., in San Diego, California. He has been with thisgroup since he joined Solar 17 years ago. His major responsibilityis to support rotordynamic analysis in design, manufacturing,testing, and field operation.Dr. Lee received his Ph. D. degree (Mechanical Engineering)

from the University of Virginia, specializing in rotating machinery.Before joining Solar, he worked for Mechanical Technology Inc., inLatham, New York, on various rotating equipment.

ABSTRACT

The evaluation of rotor system stability, which has become anessential part of rotordynamic analyses and rotating machinerydesign, relies heavily on the bearing and seal dynamic coefficientmodeling to obtain an accurate prediction of the turbomachinerybehavior. Lacking experimental validation, analytical predictionscan be widely varied and even divergent as more complexprocedures and models are created. To measure the variabilityof bearing and labyrinth seal coefficient predictions, a surveyof 60 turbomachinery users, manufacturers, consultants, andacademicians was conducted under the auspices of the AmericanPetroleum Institute (API). Coefficients received from therespondents were incorporated into a common rotordynamic modelto determine the impact on the predicted rotor stability. In addition,several of the most popular analytical codes for the prediction oftilt pad journal bearings are compared. Starting with an iso-viscousprediction and proceeding through more complicated thermal andstructural deformation solutions, the authors compare the variabilityand divergent nature of these codes. The measured variability ofthe data collected clearly illustrates the need for the resolution

1

SURVEYING TILTING PAD JOURNAL BEARINGANDGAS LABYRINTH SEAL COEFFICIENTSAND THEIR EFFECT ON ROTOR STABILITY

byJohnA. Kocur, Jr.Machinery Engineer

Plant Engineering Division

ExxonMobil Research & Engineering

Fairfax, Virginia

John C. NicholasOwner and President

Rotating Machinery Technology, Inc.

Wellsville, NewYork

andChester C. Lee

Manager, Rotordynamics Design and Test

Gas Compressor Engineering Department

Solar Turbines, Inc.

San Diego, California

of fundamental bearing issues (i.e., synchronously versusnonsynchronously reduced bearing coefficients) and labyrinth sealpredictions based on repeatable experimental data.

INTRODUCTION

Centrifugal compressor instability remains a major concerncausing reduced unit availability and project commissioningdelays, leading to lost revenue for both users and vendors of theequipment. The evaluation of rotor stability has become anessential part of rotordynamic analyses and rotating machinerydesign. In the latest edition of American Petroleum Institute (API)617, Seventh Edition (2003), specifications for performingstability analyses of centrifugal compressors were added for thefirst time. As a continued effort to improve API specifications, astudy of the current state of bearing and labyrinth seal dynamiccoefficient predictions is undertaken.Tilting pad bearing analysis started over 40 years ago spawned

by work by Lund (1964). API 684 (2005), Section 2.5.4, andNicholas (2003) provide an excellent historical perspective of thedevelopment of tilt pad bearing analysis. While work has advancedto include flexibility of the pad and pivot, nonsynchronousbehavior, thermal effects of the fluid, and pad deformation,surprisingly little experimental work has been published on thedynamic behavior of tilting pad bearings. A series of paperswas published in the 1990s on the measurement of coefficientsconcluding with Wygant, et al. (1999). However, bearing operatingspeeds were not representative of the high speed compressorapplications that typically suffer stability problems. In addition, theuse of synchronous versus nonsynchronous coefficients in stabilityanalysis was not addressed. Cloud (2006) has investigated thisargument from a rotor system standpoint. While the publishedresearch provides valuable insight into the appropriate coefficientsto use in a stability analysis, it still relies on the assumed accuracyof the bearing prediction methods employed.Gas labyrinth seal research seems to follow a different path.

While significant efforts to understand the dynamic behavior ofthese seals has been undertaken (Iwatsubo, et al., 1982; Childs andScharrer, 1986; Kirk, 1990), the validity of these approaches isoften called into question by comparison to experimental work,summarized by Childs (1993). Poor matching of the data producedby analytical methods has raised questions regarding continuedmodeling efforts of labyrinth seals from the single volumeapproach of Childs and Scharrer (1986), to the three volumeapproach of Nordmann and Weiser (1990). Current experimentalefforts seem to indicate the more complicated analytic models arenot providing better results. For the most part, these discussionsfocus on the tangential coefficients of the labyrinth seal. Radialcoefficient predictions suffer such a wide variation to experimentaldata as to be ignored in their use.Testing of short labyrinth seals has also proved to be a

challenge. Short seals with smaller pressure drops produce smallforces that have resulted in high uncertainty levels during testing.While the individual impeller eye seal may not provide the sameforce magnitude compared to the balance piston (BP), themultiple stages usually result in a total force approachingthe balance piston. In addition, the importance of data takennear application conditions was emphasized by Childs andRamsey (1991), Elrod, et al. (1995), and Wagner and Steff(1996). The later, while extending test conditions, held much ofthe data proprietary.The main objective of this paper is to determine the magnitude

of variations in tilting pad journal bearing and gas labyrinth sealdynamic coefficient predictions. This is approached in twofashions: first, as a survey sent to industry wide participants usinga common data set; and second, by comparing coefficientsobtained from several widely used tilting pad bearing analysiscodes. The survey is intended to measure the existing variations inthe industry when supplied with bearing and seal dimensions from

an unstable compressor. The code comparison effort attempts toidentify the source of the tilting pad bearing coefficient variationsas found in the survey.At the conclusion of this work, answers to the following

questions are sought:

• Is an API Level I type analysis still justified given the currentstate of analysis technology?

• Has sufficient experimental work been performed to act as thebasis for comparison of the existing analysis packages?

• Do the available analysis codes give reasonable agreement whensupplied with similar input?

• Why do different tiling pad journal bearing codes providedifferent bearing coefficients?

• Do the different journal bearing codes converge to identicalresults after stripping out all of the parameters that cause thecoefficient variations?

API SURVEYA survey was conducted under the auspices of the American

Petroleum Institute with the intent to determine the conformitylevel of bearing and labyrinth seal dynamic behavior predictions.Direct comparison of the coefficients and impact on rotor stabilityare the two methods used to determine the variability of thepredicted coefficients. The working hypothesis of the survey canbe stated as follows: “To date, there remain significant differencesacross the industry in the prediction of dynamic coefficients forfluid film tilt pad bearings and labyrinth seals.”The survey focuses on the prediction of tilt pad bearings and

impeller eye and balance piston labyrinth seals as related tocentrifugal compressor applications. Consequently, the goals of thesurvey are threefold:

• First, to improve the recently introduced specifications in API617 (2003) and 684 (2005) regarding rotordynamic stability. Thiscan be realized by examining the necessity and/or appropriatenessof the steps within the stability methodology.

• Second, highlight and communicate the disparity of thepredictions within the industry. Hopefully this can be used assupporting evidence for the continued need of research fundingfrom industry and government agencies.

• Finally, determine the need for experimental data to act as thegold standard.

While some groups have started measuring tilt pad bearingcoefficients, there remains a lack of verifiable data that can be useduniversally to determine the accuracy of analytical methods foreither bearings or labyrinth seals.The survey was proposed and supported by API. Information

relevant to the prediction of the tilt pad bearing and labyrinth sealswas sent to 60 respondents directly from API. The respondentsincluded industrial users and vendors of rotating equipment (bothat a component and turbomachinery level), consultants, andeducators. While the list of respondents was known (anddeveloped) by the authors of this paper, the responses were keptanonymous by API. From the 60 requests for information sent, 16respondents supplied bearing coefficients, 12 of which also sentseal coefficients, and several responded that they did not have theability to calculate the requested information. Approximatelyone-third of the group replied in some fashion.

SURVEY INFORMATION

As preferred by the API Subcommittee on MechanicalEquipment (SOME), an actual compressor application was soughtas the basis for the survey. The compressor used in the surveyrepresents a multistage nitrogen compressor intentionally designedby the vendor to be unstable for stability testing. Extensive testing

PROCEEDINGS OF THE THIRTY-SIXTH TURBOMACHINERY SYMPOSIUM • 20072

by the vendor determined the instability onset speed for severaloperating conditions. One such condition was selected for thesurvey representing the highest �P across the compressor obtainedduring the testing program.To minimize efforts requested from the respondents and to

eliminate modeling differences between respondents, a commonrotor model and damped eigenvalue solution algorithm (Vázquezand Barrett, 1998; Vázquez, et al., 2001), was proposed for thestudy. By eliminating these differences, the variations in thepredicted damped eigenvalues can be attributed to the suppliedbearing and seal coefficients. Respondents were asked tosupply bearing coefficients, the first and last impeller eye sealcoefficients, and balance piston coefficients for a single operatingpoint representing the onset of instability. Each was instructed totreat the information supplied as they would a new designcompressor or one with a known stability problem. The appropriatelevel of analysis to use (including thermal analysis, pivot stiffness,etc.) was left to the respondent to determine.As noted, a single rotor model was employed throughout

the study. Figures 1 and 2 display the mass (top) and stiffness(bottom) and 3D rotordynamic model of the survey compressor.The compressor is a five-stage tie-bolt design compressor withnitrogen as the working fluid. General information regarding thecompressor and operating point in question can be found in Table1. The operating point selected for the survey consists of anoperating speed of 21,662 rpm with the compressor producing 209bara at a pressure ratio of 2.53 in N2.

Figure 1. Mass and Stiffness Model of the Survey Compressor.

Figure 2. 3D Model of the Survey Compressor.

Table 1. General Information, API Survey Compressor.

The tilt pad bearings consisted of five pads with the gravity loadlocated between the pad pivots. The steel backed babbitted pad arclength was 60 degrees. Bearing clearance ratio was on the order of1.5 mils per inch. Other details of the bearing can be found inTable 2. Survey information also included oil type, oil inlettemperature and flow, oil viscosity at referenced temperatures, andpivot stiffness.

Table 2. Bearing Information, API Survey Compressor.

Labyrinth seals were described as tooth-on-rotor for all impellereyes and the balance piston. Geometry dimensions included thetooth spacing, tooth tip width, tooth height, operating clearance,and diameter as shown in Figure 3. The impeller eye sealsconsisted of four tooth seals at a diameter of 133.35 mm (5.25inches). The balance piston incorporated 11 teeth at a diameter of127 mm (5.0 inches). The gas preswirl was also given due tothe complexity of information needed to accurately compute thisparameter. Preswirl was set at 70 percent for all seals. Onlycoefficients for the first and fifth (last) impeller eye seals wererequested (also to minimize efforts). A simple linear variation wasused to derive the eye seal coefficients for the middle three impellers.

Figure 3. Dimensions Supplied in Survey for the Impeller Eye andBalance Piston Labyrinth Seals.

Gas conditions for each impeller eye seal were supplied in termsof suction and discharge pressure and temperature for the stage andfor the compressor overall. Recognizing that some respondentschose methods similar to the modified Alford’s method employedin the API 617, Seventh Edition (2003) Level I stability analysis,impeller diameter and discharge tip width and diffuser minimumwidth were also supplied in the survey. To further eliminateunintended variations, the gas properties of nitrogen were suppliedat key operating points within the compressor. Figure 4 displays atypical set of data for nitrogen.

Figure 4. Typical Nitrogen Gas Properties Supplied in Survey.

SURVEYING TILTING PAD JOURNAL BEARINGANDGAS LABYRINTH SEAL COEFFICIENTSAND THEIR EFFECT ON ROTOR STABILITY 3

Finally, an approximate location of the first critical speed wassupplied at 6700 cpm. This was intended to be used by thoserespondents supplying bearing and seal coefficients for the whirlfrequency rather than synchronous frequency. Currently, API statesthat synchronous reduction should be used in the Level I analysisfor the bearing coefficients. It should be noted however that nearlyhalf of the participants in that effort to develop the API stabilityspecifications did use nonsynchronous reduction. No suchstatement is made for the calculation of labyrinth seal coefficients.

SURVEY COEFFICIENT RESULTS

Sixteen sets of bearing data were received from the respondents.The data were normalized using the minimum stiffness suppliedfor the principle stiffness in the X-direction. The same respondent’sprinciple damping in the X-direction was used to normalize thedamping coefficients. Coefficients are presented on a graphplotting Kxx versus Kyy and Cxx versus Cyy. This display wasselected to show variations in the coefficients due to loading.However, the survey compressor’s bearing was very lightly loadedproducing almost identical coefficients in the horizontal andvertical directions. Figure 5 displays the variations in stiffness fromthe survey. Considering the effects of reduction, almost an order ofmagnitude of difference exists in the supplied coefficients. Usingthe information supplied by respondent #8, frequency reducedcoefficients are assumed to occupy the higher range of thenormalized coefficients. Figure 6 plots the variations in damping.As with the stiffness, the variation in damping from the smallestto largest is also nearly an order of magnitude. While usingnonsynchronous reduction increases the bearing stiffness, dampingis predicted to decrease with the reduction (Figure 6).

Figure 5. Normalized Principle Stiffness Coefficients.

Figure 6. Normalized Principle Damping Coefficients.

It should be noted that the approximate location of the firstcritical speed was erroneously supplied at 6700 cpm rather than theintended 9700 cpm. Unfortunately, this error was identified afterthe information was received from the respondents. While thisinformation has no effect on the respondents’ using synchronousvalues for the bearing and labyrinth seal coefficients, it willadd some unintended variation to those supplying frequencydependent coefficients.The frequency impact and reduction schemes are well known

(Parsell, et al., 1983, and Barrett, et al., 1988). Examining theeffect of this parameter on tilt pad bearings, a set of coefficientswas predicted using a solution scheme developed by Branagan(1988). Comparing the frequency reduced coefficients using 6700cpm, 9700 cpm, and 21,662 rpm, the change in the principlestiffness and damping coefficients are shown in Table 3. As noted,reducing the full coefficient matrix to the lower frequency,increases the stiffness and reduces the damping an additional 10percent from the synchronously reduced coefficients whencompared to the reduction to the intended frequency of 9700 cpm.

Table 3. Predicted Impact of Frequency on Principle BearingCoefficients.

Figure 7 illustrates the expected change in respondent #8’sbearing coefficients if a frequency of 9700 cpm is used in thereduction process instead of 6700 cpm. The difference betweensynchronous and nonsynchronous reduction is decreased by ≈25percent assuming the same variation as shown in Table 3. Theimpact on rotor stability is shown later.

Figure 7. Approximate Impact of Frequency on BearingCoefficients from Respondent #8.

The coefficient variations for the labyrinth seals are presentedon graphs of the tangential force components, Kxy versus Cxx. Allrespondents supplied symmetric coefficients for the seals. Aspreviously mentioned, coefficients for impeller #1 and #5 eye sealsand the balance piston were requested. The variations for theseseals are plotted in Figures 8, 9, and 10. For the most part, therelative variations in Kxy are matched by the variations in Cxx. Oneexception is noted for a respondent supplying Alford type forcesfor the labyrinth seals.The bulk of the respondents’ impeller eye seal coefficients falls

in the normalized range from 5 to 20. One respondent was less (at


Figure 8. Normalized Seal Coefficients for Impeller #1 Eye Labyrinth.

Figure 9. Normalized Seal Coefficients for Impeller #5 Eye Labyrinth.

Figure 10. Normalized Seal Coefficients for Balance PistonLabyrinth (Expanded Range Shown in Inset).

1.0) and one higher (at 42.) For the balance piston, the majority fellin the normalized range from 1 to 12 (an order of magnitude) withone at 37 and another at 137 (greater than two orders ofmagnitude.) (Confirmation of the values supplied by severalrespondents was requested but not received at the time of thispaper.) Since the variations in cross-coupled stiffness and principledamping were nearly equal for all coefficients, it comes as nosurprise that the destabilizing force, defined by Equation (1), alsoshows a similar range in normalized values, Figure 11.

Figure 11. Normalized Destabilizing Force for the Balance PistonLabyrinth.

Using a solution algorithm developed by Kirk (1990), the impactof the whirl frequency ratio (WFR) on the BP coefficients wasstudied. For the conditions specified, balance piston coefficientswere calculated for 0.31 (6700 cpm), 0.45 (9700 cpm), and 1.0(synchronous) whirl frequency ratios. Both cross-coupled stiffnessand direct damping were smaller for the lower whirl frequencyratio. If the destabilizing force is defined as:

Then calculating the Qa using the tangential coefficients derivedfrom the three whirl frequency ratios produces the results shown inTable 4. As noted, Qa is only 10 percent greater for the highersubsynchronous WFR. The impact on stability of using 6700 cpmrather than 9700 cpm would be a secondary effect. Also notice thesize of the destabilizing force if a WFR of 1.0 is used to predict theseal coefficients. This represents a 2× increase over the coefficientsderived from a WFR of 0.45.

Table 4. Impact of WFR on the Destabilizing Force of theBalance Piston.

STABILITY RESULTS

Coefficients supplied for the bearings and seals wereincorporated into a common model of the compressor rotor. Inaddition, a single complex eigenvalue solution algorithm was usedfor each model. The bearings were represented by the principleterms only. Cross-coupling terms were consistently three orders ofmagnitude smaller than the principle terms and could thus besafely ignored. Labyrinth seal behavior was modeled using onlythe tangential terms supplied by the respondents due to thevariations in the radial terms. Figure 12 plots the variation in theradial terms of impeller #1 eye seal. The principle stiffness variesby two orders of absolute magnitude and the cross-coupleddamping by nearly three orders of magnitude, both ranging frompositive to negative values. In addition, the relative size of eachwas large enough to significantly change the predicted frequencyof the first forward damped eigenvalue.


Q k Ca wf= − ω

Figure 12. Variations in Radial Coefficients of Impeller #1 Eye Seal.

The predicted compressor stability considering a rotor/bearingsonly model with the respondents’ coefficients is displayed inFigure 13. On the plot, the coefficients supplied by respondent #8are highlighted for the two reduction schemes. A third point (inred) is plotted for an anticipated reduction to 9700 cpm. The impacton the base stability level of the rotor is secondary even with the 25percent reduction in coefficient difference noted earlier in thepaper. As is expected given the impact on the bearing coefficients,synchronous reduction of the bearing coefficients produces a lowerfirst natural frequency (due to the softer stiffness values) and ahigher logarithmic decrement (log dec) (due to the higher predicteddamping.) The general trend can be seen on the plot as higherlog dec values are produced at lower natural frequencies. Thisillustrates both the effect of reduction schemes and the increase inrelative shaft-to-bearing stiffness making the bearing dampingmore effective in stabilizing the first mode.

Figure 13. Predicted Compressor Stability with Rotor/Bearings Only.

Adding the labyrinth seals to the stability model impacts thepredicted log dec as shown in Figure 14. As before, the tworeduction schemes supplied by respondent #8 are highlightedwith their seal coefficients. A third point is added that includesboth the bearing and seals coefficients altered representing a9700 cpm location of the first damped forward mode. Thechange in log dec remains secondary to the overall variations inpredicted rotor stability. In addition, a second respondent’s sealcoefficients were changed by the amount indicated in Table 4.The 10 percent increase in destabilizing force was added torespondent #13’s coefficients and the change indicated by thearrow in Figure 14.

Figure 14. Predicted Compressor Stability with Rotor/Bearings/Labyrinth Seals.

Two calculated points are not presented in the plotted range inFigure 14, one at approximately a log dec of 1.8 and the other ata log dec exceeding �2.0. These are omitted to more clearlydefine the influence of respondent #8’s coefficients and due to theuncertainty of the validity of these points.An engineer analyzing this compressor at the onset of the design

with the supplied coefficients would conclude a stable compressordesign with aWFR less than tested in the majority of the cases. Thesubsynchronous frequency of the first mode as witnessed on thetest stand was 10,700 cpm. This is plotted in Figure 14 against thepredicted results. While the change is slight from 9700 cpm to10,700 cpm, the change for most results would be in the wrongdirection, to the left and up in Figure 14. Finally, the impact of thelabyrinth seals on the rotor stability (log dec) is shown in Figure 15.The distribution seems skewed to the higher side ranging betweena �0.2 to �1.5 change in log dec.

Figure 15. Labyrinth Seal Impact on Predicted Log Dec.

TILTING PAD JOURNAL BEARINGSTIFFNESS AND DAMPING VARIANCE

From the previous sections, it is clear that there is a relativelywide variation in the stiffness and damping (K and C) propertiesprovided by various computer codes for this stability study. In anattempt to investigate the source of this K and C variation, fivedifferent tilting pad journal bearing codes are utilized. Referencesfor the five codes are Nicholas, et al. (1979), Branagan (1988), SanAndres (1995), He (2003), and Chen and Gunter (2005).The analysts were given the instructions to run the various codes

for the tilting pad bearing from Table 5 as if they were about toperform an API stability analysis. Additional instructions included


utilizing whatever thermal solution that was available within thecode and to use a pad pivot stiffness of Kpiv = 8.0 × 105 lbf/in.Finally, synchronous coefficients were requested.

Table 5. Stability Results Using Variable Viscosity DerivedCoefficients from Four Bearing Codes at 21,662 RPM.

The results are illustrated in Figures 16, 17, 18, and 19 for Kxx,Kyy, Cxx, and Cyy, respectively. Note that only codes #1 through #4are shown as code #5 did not have a variable viscosity capability(or the operator did not know how to use the capability).Additionally, code #2 had to be run with infinite pivot stiffness.Clearly, some K and C variation is evident.

Figure 16. Kxx - Variable Viscosity, Kpiv = 8.0×105 lbf/in.

Figure 17. Kyy - Variable Viscosity, Kpiv = 8.0×105 lbf/in.

At 21,662 rpm, the maximum pad metal temperature (Tmax)predictions from the four codes varied from 173�F (#1) to 221�F(#4). Ignoring #2 as it was run with infinite pivot stiffness, thelowest predicted temperature resulted in the highest predicted Kand C values (#1). Likewise, the highest predicted temperatureresulted in the lowest predicted K and C values (#4).

Figure 18. Cxx - Variable Viscosity, Kpiv = 8.0×105 lbf/in.

Figure 19. Cyy - Variable Viscosity, Kpiv = 8.0×105 lbf/in.

Stability results at 21,662 rpm using the variable viscosity K and Ccoefficients from the four codes discussed above are summarized inTable 5. Code #2 is listed twice for synchronous and nonsynchronouscoefficients.All other results are for the synchronous coefficients fromFigures 16, 17, 18, and 19. For synchronous coefficients, the variationin log dec is between 0.552 and 0.796. Code #2’s nonsynchronouscoefficients produce a log dec value of 0.128.Eliminating this variation in temperature predictions and thus in

viscosity variation, the codes were rerun assuming a constantviscosity. The iso-viscous results are shown in Figures 20, 21, 22,and 23. While the damping variation is negligible (Figures 22 and23), the stiffness values still show some variation. Closer examinationof Figures 20 and 21 reveal that codes #1 and #5 predict the samevalues while codes #3 and #4 predict the same values. Codes #1and #5 use the pad assembly solution technique (Lund, 1964) whilecodes #3 and #4 utilize the full pad solution technique.

Figure 20. Kxx - Iso-viscous, Kpiv = 8.0×105 lbf/in.


Figure 21. Kyy - Iso-viscous, Kpiv = 8.0×105 lbf/in.

Figure 22. Cxx - Iso-viscous, Kpiv = 8.0×105 lbf/in.

Figure 23. Cyy - Iso-viscous, Kpiv = 8.0×105 lbf/in.

Eliminating the pad pivot stiffness as a possible source ofvariation, the codes were rerun assuming infinite pivot stiffnessand constant viscosity. The iso-viscous, infinite pivot stiffnessresults are shown in Figures 24, 25, 26, and 27. Now, all five codespredict essentially the same results for speeds above 4,000 rpm.Below 4,000 rpm, four out of the five codes predict essentiallyidentical results.

Figure 24. Kxx - Iso-viscous, Kpiv = Infinite.

Figure 25. Kyy - Iso-viscous, Kpiv = Infinite.

Figure 26. Cxx - Iso-viscous, Kpiv = Infinite.

Figure 27. Cyy - Iso-viscous, Kpiv = Infinite.


CONCLUSIONS

This work presents surveyed bearing and seal coefficients fromvarious industry sources including academia, manufacturers,users, and consultants. The variation in those coefficients andtheir impact on an example compressor rotor stability predictionwas highlighted. In an effort to explain the origin of the bearingcoefficient differences, five widely used bearing codes werecompared. Various input options affecting the thermal solution andpivot stiffness were investigated for their effect on the predictedstiffness and damping of the tilting pad bearing.The industry survey supplied sufficient (but maybe not

exhaustive) data to confirm the working hypothesis that significantdifferences exist in the prediction of dynamic coefficients fortilting pad journal bearings and gas labyrinth seals. The differencesarise from several sources, some due to the solution algorithm andsome traced to the user. Whatever the source, sufficient evidenceexists to validate the following conclusions:

• A gold standard of experimental data is needed for both tiltingpad journal bearings and gas labyrinth seal dynamic coefficients.The data should be used to validate analytical prediction methods.Preferably, the experimental data should be obtained for identicalcomponents from several sources. Due to variations in testingprocedures, test equipment, etc., multiple sources should beencouraged and funded to obtain the data. Emphasis should beplaced on obtaining the component information. While systembehavior may provide insight into overall component behavior, i.e.,nonsynchronous versus synchronous reduction (Cloud, 2006), itdoes not provide enough information to validate componentcoefficient predictions.

• The Level I analysis in theAPI specifications is still needed. Theoriginal intent of that analysis was to provide a screening toolto identify rotors requiring only a simplified analysis and toprovide a common methodology for stability calculations. Thewide variations in bearing and seal coefficients and continueddebate on synchronous versus nonsynchronous reduction of tiltingpad bearing coefficients, still necessitate a common methodologyto permit valid comparisons across the industry.

• For the survey compressor, nonsynchronous reduction ofbearing coefficients appears to represent the rotor support situationmore accurately. Synchronous coefficients tended to underpredictthe frequency and overpredict the stability level of the first forwardmode. However, strong caution is advised in drawing widespreadconclusions without addressing the significant variations inbearing and seal coefficients shown in the study. Further research,both on a component and system level, is needed in this area.

The predicted stability of the survey compressor was greatlyaffected by the variations in the bearing and seals coefficients.Frequency predictions for the first forward mode ranged from6000 to 11,300 cpm. Log dec magnitudes from +1.0 to �1.0even after ignoring the extremes. While the authors understandthat not all compressors will show this sensitivity, the surveycompressor shares traits with rotors typically showing instabilityproblems, namely, lightweight rotors operating at high speedsand pressures.

• Analytical predictions of labyrinth seals, both short and long, arestill incomplete and, in some cases, insufficient. As noted in thestudy, the radial force coefficients of the seals were ignored due tothe extreme variations (approaching three orders of magnitude) inthe coefficients. The study also showed that for the balance pistonseal, the destabilizing force, Qa, produced from synchronouslyderived coefficients to be 2× greater than that produced fromcoefficients derived from a WFR of 0.45.

• The impact of the subsynchronous frequency used in solving foreither the bearing or seal coefficients was also studied using twospecific solution algorithms. A change in WFR from 0.33 to 0.45

decreases by 25 percent the difference between synchronously andsubsynchronously reduced bearing coefficients and increases thedestabilizing force by 10 percent for the balance piston seal. Bothwere shown to produce only secondary effects.

From the computer code survey, code-to-code variations in thetilting pad journal bearing stiffness and damping coefficients werefound to be due to several sources:

• Synchronous versus nonsynchronous coefficients• This is the major source of the K and C variation.

• The temperature solution technique and the resulting pad metaltemperature prediction and, thus, the viscosity variation

• The methodology of the inclusion of pivot stiffness• The pad assembly codes handle the inclusion of pivot

stiffness differently compared to the full solution codes.

Stripping out the above three variations, all five codes produceessentially identical synchronous coefficients for the infinite pivotstiffness, iso-viscous case.With regard to the industry survey, variations in the bearing

coefficients stem from user assumptions affecting pivot stiffnessand the frequency reduction of the coefficients. These variationsappear to be related more to the user controlled analysis optionsrather than the analytical methods employed.The authors hope the presented work can be used to stimulate

funded research in the areas of tilting pad and labyrinth sealcoefficient predictions/measurements and the impact of frequencydependency on both. As compressor development continues toexpand in size and power with an accompanied increase in the costof unexpected downtime or project delays, accurate prediction ofthe compressor dynamic behavior is essential.

NOMENCLATURE

Cs = Seal diametral clearance, mm (mils)C, Cxx = Principle damping, N-s/mm (lbf-s/in)D = Labyrinth seal diameter, mm (in)Hc = Minimum width of the impeller or discharge volute,

mm (in)K, Kxx = Principle damping, N-s/mm (lbf/in)k, Kxy = Cross-coupled stiffness, N/mm (lbf/in)Qa = Destabilizing force, N/mm (lbf/in)Tw = Labyrinth seal tooth tip width, mm (in)Th = Labyrinth seal tooth height, mm (in)Ts = Labyrinth seal tooth spacing, mm (in)WFR = Whirl frequency ratio = wwf/operating speedwwf = Whirl frequency of first natural frequency, rad/sec

REFERENCES

American Petroleum Institute Standard 617, 2003, “Axial andCentrifugal Compressors and Expander-Compressors forPetroleum, Chemical and Gas Industry Services,” SeventhEdition, American Petroleum Institute, Washington, D.C.

American Petroleum Institute RP 684, 2005, “API StandardParagraphs Rotordynamic Tutorial: Lateral Critical Speeds,Unbalance Response, Stability, Train Torsionals, and RotorBalancing,” Second Edition, American Petroleum Institute,Washington, D.C.

Barrett, L. E., Allaire, P. E. and Wilson, B. W., 1988, “TheEigenvalue Dependence of Reduced Tilting Pad BearingStiffness and Damping Coefficients,” ASLE Transactions, 31,pp. 411 - 419.

Branagan, L. A., 1988, “Thermal Analysis of Fixed and Tilting PadJournal Bearings Including Cross-Film Viscosity Variationsand Deformation,” Ph.D. Dissertation, University of Virginia,Charlottesville, Virginia.


Camatti, M., Vannini, G., Fulton, J., and Hopenwasser, F., 2003,“Instability of a High Pressure Compressor Equipped withHoneycomb Seals,” Proceedings of the Thirty-SecondTurbomachinery Symposium, Turbomachinery Laboratory,Texas A&M University, College Station, Texas, pp. 39-48.

Chen, W. J. and Gunter, E. J., 2005, “Introduction to Dynamics ofRotor-Bearing Systems,” Victoria, British Columbia, Canada:Trafford Publishing.

Childs, D. W., 1993, Turbomachinery Rotordynamics: Phenomena,Modeling, and Analysis, New York, New York: John Wiley &Sons, Inc.

Childs, D. W. and Scharrer, J. K., 1986, “An Iwatsubo-BasedSolution for Labyrinth Seals: Comparison to ExperimentalResults,” ASME Journal of Engineering for Gas Turbines andPower, 108, (2), pp. 325-331.

Childs, D. W. and Ramsey, C., 1991, “Seal Rotordynamic-Coefficient Test Results for a Model SSME ATD-HPFTPTurbine Interstage Seal With and Without a Swirl Brake,”ASME Journal of Tribology, 113, pp. 113-203.

Cloud, C. H., 2006, “Stability of a Rotor Supported on Tilting PadJournal Bearings,” Ph.D. Dissertation, University of Virginia,Charlottesville, Virginia.

Elrod, D. A., Pelletti, J. M., and Childs, D. W., 1995, “TheoryVersus Experiment for the Rotordynamic Coefficients of anInterlocking Labyrinth Gas Seal,” ASME Paper 95-GT-432,Presented at the International Gas Turbine and AeroengineCongress and Exposition, Houston, Texas.

He, M., 2003, “Thermoelastohydrodynamic Analysis of Fluid FilmJournal Bearings,” Ph.D. Dissertation, University of Virginia,Charlottesville, Virginia.

Iwatsubo, T., Matooka, N., and Kawai, R., 1982, “Spring andDamping Coefficients of the Labyrinth Seal,” NASA CP-2250,pp. 205-222.

Kirk, R. G., 1990, “A Method for Calculating Labyrinth Seal InletSwirl Velocity,” ASME Journal of Vibration and Acoustics,112, (3), pp. 380-383.

Kocur, J. A., Jr. and Hayles, G. C., Jr., 2004, “Low FrequencyInstability in a Process Compressor,” Proceedings of theThirty-Third Turbomachinery Symposium, TurbomachineryLaboratory, Texas A&M University, College Station, Texas,pp. 25-32.

Lund, J. W., 1964, “Spring and Damping Coefficients for theTilting-Pad Journal Bearing,”ASLETransactions, 7, pp. 342-352.

Nicholas, J. C., 2003, “Lund’s Tilting Pad Journal Bearing PadAssembly Method,” ASME Journal of Vibrations andAcoustics, 125, (4), pp. 448-454.

Nicholas, J. C., Gunter, E. J., andAllaire, P. E., 1979, “Stiffness andDamping Coefficients for the Five Pad Tilting Pad Bearing,”ASLE Transactions, 22, (2), pp. 112-124.

Nordmann, R. and Weiser, H., 1990, “Evaluation of RotordynamicCoefficients of Look-Through Labyrinths by Means ofa Three Volume Bulk Model,” Rotordynamic InstabilityProblems in High-Performance Turbomachinery, NASACP-3122, pp. 141-157.

Parsell, J. K., Allaire, P. E., and Barrett, L. E., 1983, “FrequencyEffects in Tilting-Pad Journal Bearing Dynamic Coefficients,”ASLE Transactions, 26, pp. 222-227.

San Andres, L. A., 1995, “Bulk-Flow Analysis of Flexure andTilting Pad Fluid Film Bearings,” TRC-B&C-3-95,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas.

Vázquez, J. A. and Barrett, L. E., 1998, “Representing FlexibleSupports by Polynomial Transfer Functions,” ASME Paper98-GT-27.

Vázquez, J. A., Barrett, L. E., and Flack, R. D., 2001, “A FlexibleRotor on Flexible Bearing Supports: Stability and UnbalanceResponse,” Journal of Vibration and Acoustics, ASMETransactions, 123, (2), pp. 137-144.

Wagner, N. G. and Steff, K., 1996, “Dynamic LabyrinthCoefficients from a High-Pressure Full-Scale Test Rig UsingMagnetic Bearings,” Rotordynamic Instability Problems inHigh-PerformanceTurbomachinery,NASACP-3344, pp. 95-111.

Wygant, K. D., Barrett, L. E., and Flack, R. D., 1999, “Influence ofPad Pivot Friction on Tilting-Pad Journal BearingMeasurements—Part II: Dynamic Coefficients,” STLETribology Transactions, 42, (1), pp. 250-256.

ACKNOWLEDGEMENT

The authors recognize API (Roland Goodman), ExxonMobil,Solar Turbines, and RMT, Inc., for their assistance and support.


Leonardo Baldassarre is currently theEngineering Manager & Principal Engineerfor Centrifugal Compressors with GeneralElectric Oil & Gas Company, in Florence,Italy. He is responsible for all requisition,standardization, and CAD automationactivities as well as for detailed design ofnew products for centrifugal compressorsboth in Florence (Nuovo Pignone) and LeCreusot (Thermodyn). Dr. Baldassarre

began his career with General Electric Nuovo Pignone in 1997. Hehas worked as Design Engineer, R&D Team Leader for centrifugalcompressors in Florence, Product Leader for centrifugal and axialcompressors, and Requisition Manager for centrifugal compressorsboth for Florence and Le Creusot teams.Dr. Baldassarre received a B.S. degree (Mechanical Engineering,

1993) and Ph.D. degree (Mechanical Engineering/TurbomachineryFluid Dynamics, 1998) from the University of Florence. He hasauthored or coauthored 20 technical papers, mostly in the area offluid dynamic design of 3D transonic impellers, rotating stall, androtordynamics. He presently holds three patents.

John W. Fulton is a Senior EngineeringAdvisor with Exxon Mobil Research andEngineering Company, in Fairfax, Virginia.In his 35 years with Exxon, he has worked inall phases of machinery engineering and inresearch and development. Mr. Fultonenjoyed years of assignments in Libya,Venezuela, Alaska, London, and KualaLumpur. He is co-inventor of six U.S. Patents.Mr. Fulton has a B.S. degree (Mechanical

Engineering) from New Jersey Institute of Technology.

ABSTRACT

The calculation for unbalance response of a rotor starts bycalculating the bearing load to provide the basis for the bearingstiffness and damping characteristics. Measurements in testrigs at a major Texas university laboratory have shown thathoneycomb-stator/drum-rotor annular seals can produce negativestiffness, in particular at zero to low whirl frequencies, whichtends to pull the rotor off-center. Data given in this paper, fromhigh-pressure factory tests of compressors using a honeycomb sealat the balance piston, have shown the rotor can be displaced from

the usual position in the lower part of its journal bearing, degradingthe unbalance response. In this paper the authors show that thehoneycomb seal can produce a large disturbance in the equilibriumposition of the rotor for certain values of negative stiffness,resulting in high bearing loads in unusual directions. It is alsoshown that aerodynamic forces on the rotor from the volute need tobe considered.

INTRODUCTION

In the oil and gas industry, the typical centrifugal compressor forreinjection duty has its impellers placed in the casing between twobearings, with the shaft horizontal with respect to gravity. Tocalculate the vibration response of the rotor to unbalance, orthe damped critical speeds (to evaluate rotordynamic instability)one has to know the bearing characteristics, which depend onthe bearing loading. Most rotordynamic suites include a simplecalculation that automatically finds the rotor weight and center ofgravity. This calculation of static equilibrium then finds the load ateach bearing by solving two equations, one found by setting thesum of forces equal to zero, and the other found by setting the sumof moments around one bearing equal to zero. This calculation willbe shown in detail below.

This paper is motivated by experience with several compressorson full-load test. The specific compressors involved, and theoperating conditions on test are given in Table 1.

Table 1. Specific Compressors and Operating Conditions.

These compressors use a honeycomb seal running against adrum rotor for the balance piston. In contrast to labyrinth sealswith teeth, or in contrast to rotors with labyrinth teeth runningagainst honeycomb, the honeycomb/drum type seals havesignificant direct stiffness (defined more completely below).“Hole pattern” type seals running against drum rotors are similar.Because the test experience related in this paper was withhoneycomb seals, not hole pattern type, this paper will discusshoneycomb seals only.

Figure 1 shows a photograph of the bore of a honeycomb seal.There are about 10,000 cells in this half of the seal. Each cell hasa hexagonal width of about 2 mm (.08 inch) and a depth of about2.3 mm (.09 inch). This seal is formed from a monolithic billet ofaluminum, for all examples. In these examples the honeycomb sealacts on the balance piston that is just behind the last impeller.

11

ROTOR BEARING LOADSWITH HONEYCOMB SEALSANDVOLUTE FORCES IN REINJECTION COMPRESSORS

byLeonardo Baldassarre

Engineering Manager & Principal Engineer for Centrifugal Compressors

General Electric Oil & Gas Company

Florence, Italy

andJohnW. Fulton

Senior Engineering Advisor

ExxonMobil Research and Engineering Company

Fairfax, Virginia

Figure 1. Photo of a Honeycomb Seal.

These compressors also have a volute at the last impeller. Thevolute can produce asymmetric pressure gradients around the rotor.If the gas pressure is sufficiently high, then the asymmetricgradient may produce a radial force on the rotor that exceeds therotor weight, as shown later.

EXPERIENCE ON FULL LOAD TEST

During full-load tests of compressor Example A, erraticresponse to unbalance was noted. Running at constant speed, thesynchronous vibration varied from 6 to 25 microns peak-to-peak,as pressure and flow were varied. Clearly the high pressure gas wasaffecting the unbalance response of the rotor, either directly, orby affecting the bearing load, and thus the bearing dynamiccharacteristics. In an attempt to identify the cause, the pressure andflow were varied and repeated, over an appreciable range, as shownin Figure 2 and 3, respectively.

Figure 2. Synchronous Response (Filtered at 1 × RPM) ofCompressor Example A at the Discharge End X-Probe as aFunction of Discharge Temperature.

Figure 3. Synchronous Response (Filtered at 1 × RPM) ofCompressor Example A at the Discharge End X-Probe as aFunction of Suction Volume Flow.

From these figures it is clear that the change in unbalanceresponse was well demonstrated, repeatable, and significantlyaffected. The three points at lower flow correspond to the threepoints at higher temperature. The separation of points is not asdistinct when plotted against pressure.

From previous work by Camatti, et al. (2003), it is known that theforces produced by a honeycomb seal in high pressure gas aresensitive to the clearance of the leakage annulus. From finite elementanalysis of the honeycomb seal in this compressor, it is known thattemperature changes the clearance and causes a taper in the clearanceas well. Therefore it is not surprising that the rotor response couldvary with discharge temperature, which sets the temperature of theseal, distorting the taper, and thus changing the honeycomb sealforces on the rotor. These changes apply to both the static force andthe dynamic stiffness and damping coefficients acting synchronously.

It will be shown below, by calculation for the test stand conditionsof Example C, that the volute force can exceed the rotor weight. Theforce and direction of the radial load produced by the volute dependson the ratio of volume flow to the design flow. Thus changing the flowcan produce substantial additional load on the bearings, and thus affecttheir dynamic characteristics. Thus it is not surprising that theunbalance response might vary with volute-induced bearing loads.However, no practical method was obvious to distinguish between thebearing loads caused by the honeycomb seal versus the volute.

The rotordynamic stability of the Example A (and Example B,discussed below) compressors were excellent as tested. Theresponse-to-unbalance was also within contract vibration limits.Therefore, an extensive investigation during full-load testing wasnot warranted, and an exact comparison of the rotor position on testand the position calculated below was not made. However, theeffects on unbalance response in Example A were significant, andcould have caused a problem, if the rotor response were not so wellbehaved in the base case. This gave motivation to find an analyticsolution to understand the lifted rotor position that was observedand estimate the bearing loads that might occur.

In the case of Example B (and in other cases [Camatti, et al.,2003]) it was found that the proximity probes at the bearing journalsshowed the rotor journal was not resting in the bottom of the bearingwhen running, as assumed in the bearing load calculation that wasoutlined above. Instead the journal was often found in the top of thebearing, and the journal could be seen to lift as the compressor wasstarted and loaded. Figure 4 shows the bearing journal position,measured by proximity probes, when the compressor was started andloaded. The left plot is for the balance piston end, and the right is thethrust end. The large circles represent the bearing clearance circle(nominal cold dimensions.) The parameter written by each point isrevolutions per minute (rpm). The suction volume flow 76 percent ofrated at full speed. From the calculated behavior of a tilt-pad bearing,one could expect that the journal should rise toward the center of thebearing, but not as high as the bearing center. Clearly, the journals inExample B rise above center, and more so, on the balance piston end.

Figure 4. Bearing Centerline Position as Compressor Example B isStarted and Loaded. The Plot on Left is the Balance Piston Endand the Plot on the Right is the Thrust End.


The tilt-pad bearings used on these compressors havethermocouples installed to measure the Babbitt temperature of thetwo bottom pads. On Example B, as a check on the validity of thejournal center positions shown in Figure 4, one pad was movedfrom the bottom to the top of one bearing, with the second padremaining in normal position. As expected from Figure 4, the toppad showed higher temperature than the bottom, confirming theload on the bearing was directed upward, and confirming thebehavior shown there. The maximum temperatures measured wereas shown in Table 2.

Table 2. Maximum Babbitt Temperatures.

Thus the top pad (72-TE-29031A) of the bearing near the balancepiston is 17�C (62.6�F) hotter than the bottom pad, indicating asubstantially larger load on the top pad. Just to give a completepicture, these temperature records are plotted versus time in Figure 5.

Figure 5. Trend Plot of the Bearing Pad Temperatures (Top FourTraces) on Example B, Bottom Trace is Speed in RPM.

CALCULATING THE EFFECTOF HONEYCOMB SEAL ONSTATIC EQUILIBRIUM OF THE ROTOR

Because both the potential honeycomb seal force and thepotential volute force could be responsible for the anomalousjournal position and bearing loads discussed above, the two effectswere investigated by calculations after the testing was complete.The bearings themselves were not suspected of causing theanomalous position, as multiple disassemblies to change dry gasseals during some of the testing did not implicate them. On oneinstance with Example B, an increase once-per-revolution vibrationprompted a bearing inspection that showed faulty assembly.However correcting this did not eliminate the anomalous position.The remainder of this paper will discuss the results of the calculation.

As a basis for understanding the calculations, it is necessary tocompare and contrast the general behavior of the honeycomb sealdirect stiffness and the volute force as follows:

• The balance drum/honeycomb seal was next to the volute (asusual) in the above cases, so that it was not obvious by comparingthe behavior of the bearing on one end of the casing to the other,whether the honeycomb seal or the volute was responsible for theanomalous journal position in the bearing, as might be expectedfrom the static equilibrium calculations.

The honeycomb seal forces are effectively proportional to thedisplacement of the drum within the honeycomb seal runningclearance, That is, they have the characteristic of a spring rate thatcan be measured in Newtons per meter (pounds per inch).

• The volute forces change as the flow rate changes with respectto the design (best efficiency) flow. Of course their magnitudechanges with pressure. However, the volute forces are independentof small changes of the rotor position within its running clearance.That is, the volute force does not have the characteristic of spring.

• The honeycomb seal spring rates are extremely sensitivevariation of the running clearance along the length of the drum. Awell-considered finite element analysis of the drum and of thehoneycomb seal is necessary to define the clearance along thelength as a function of temperature, pressure, shaft speed, andmounting conditions. Changes in these variables during testingmay change the static equilibrium of the rotor position.

• The honeycomb seal spring rates in both the direct andcross-coupled directions are strong functions of the whirlfrequency of the drum orbital motion. For calculation of staticequilibrium the displacement occurs at zero whirl frequency.

• Honeycomb seal spring rate, K, in the direction of drumdisplacement may affect the first bending frequency of the rotor.This reduced whirl frequency can fall into the range where thehoneycomb has negative damping. This can cause rotordynamicinstability, resulting in catastrophic vibration. Such a problem isreported in a previous paper (Camati, et al., 2003) but is not ofconcern here as the subject compressors were very stable.

The honeycomb seal spring rates were calculated using the sealcode developed by Kleynhans and Childs (1997). The voluteforces were calculated using a computational fluid dynamics code.Figure 6 shows the coordinate system used. In a free body diagramfor static equilibrium all forces are considered to be acting on therotor. Lhc is the distance from Bearing 1 to the center of thehoneycomb seal.

Figure 6. Rotor Coordinate System.

The simple calculation of the bearing load in the typicalrotordynamic code uses two equations. The sum of the forces, and thesum of the moments, must be equal to zero, for static equilibrium.For the simple calculation without the honeycomb seal, it is notnecessary to know the spring rates of the bearings as one can find thetwo bearing loads with the two equations. The static equilibrium canbe written by summing the forces to zero (Equation 1) plus summingthe moments to zero (Equation 2). Only the vertical plane need beconsidered, as there are no horizontal forces.

ROTOR BEARING LOADSWITH HONEYCOMB SEALSANDVOLUTE FORCES IN REINJECTION COMPRESSORS 13

•

fb fb Wr1 2 0+ − =

Lcg Wr Lcl fb⋅ − ⋅ =2 0

The standard model for forces acting on the rotor is given inEquation (3) (Kleynhans and Childs, 1997).

For static equilibrium, the velocities Xdot and Ydot are zero, whicheliminates the damping coefficients, C and c, from consideration. Thisequation can be applied to the static force caused by the honeycombseal acting on a rotor, as expressed by Equations (4) and (5):

In Equations (4) and (5) the direct stiffness is K and thecross-coupled stiffness is k. Note that the cross-coupled stiffnessrequires that the horizontal displacement be included in theequations for vertical forces and vice versa. The subscript y indicatesthe force from the honeycomb acts in the y (vertical) direction andthe subscript x indicates action in the horizontal direction.

When the honeycomb seal is added, there are now threeunknown force vectors, the two force vectors of the bearings on therotor and now the force vector of the honeycomb seal on the rotor.However, there are only two force equations (one in the horizontaldirection and one vertical) and two moment equations. Thus thecalculation becomes a “statically indeterminate” problem (Popov,1968), because the force caused by the honeycomb seal depends onits drum displacement.

Six equations are required to solve for the six displacements.Two force balances are provided by Equations (6) and (7) and twomoment balances are provided by Equations (8) and (9).

To account for the cross coupling, the forces are written asfunctions of both horizontal (x) and vertical (y) displacements. Theforce balance equation is normalized by rotor weight, Wr, to suitthe tolerance limit of the numerical method used. The momentequations are normalized by moment Wr × Lcl.

Figure 7 shows the displacements in the vertical direction asdefined by Equation (10).

That equation gives the displacement of the rotor drum at thehoneycomb seal, y3. The first term, y1, is the displacement ofthe bearing at the coordinate origin. The second term uses thedisplacement of the second bearing, y2, to find the centerlinedisplacement at the honeycomb seal location. The third termrepresents the bending of the rotor due to gravity. The fourth term

gives the displacement due to rotor bending under force from thehoneycomb seal. The displacements in the horizontal plane, shownin Equation 11, are similar but of course do not include bendingdue to gravity.

Figure 7. Definition of the Rotor Displacements in theVertical Plane.

The spring coefficient representing the rotor bending stiffnessis easily found by using a rotor response code to calculate anasynchronous response at very low frequency (1 Hertz) to find thedisplacement at the honeycomb seal caused by an asynchronousforce at the honeycomb location

As mentioned, Equation (10) also includes the displacement ofthe rotor at the honeycomb seal location, yhcstatic due to rotorweight. It is handled separately as it remains constant while theother displacements in Figure 6 vary. Handling it separately avoidscomplication due to the weight action not being at the honeycombseal location. For the examples given, yhcstatic is a significant term.The authors estimated it from the first bending frequency of therotor, using the concept of the resting displacement of a singlespring-mass oscillator. The yhcstatic is adjusted for the actualbending curve of the rotor, using the concept of the Raleigh naturalfrequency method, using the first bending mode shape as calculatedby a rotordynamic code.

This system can be solved for the six particular values ofdisplacement, using a standard numerical solution method.Because the force equations are written as functions of displacement,no further algebra is required when using the solver in a populartechnical calculation software. Figure 8 shows the result graphicallyin the same format as used in Figure 4. Please note that thedirection of rotation as viewed in Figure 4 is clockwise, while itis counterclockwise as viewed in Figure 8. (This reverses thehorizontal displacements between the two figures, as the view isfrom opposite ends of the rotor.) The displacements are normalizedby their respective clearance circles to indicate the limit of validityof the solutions, due to contact if a displacement exceeds theclearances. The honeycomb seal clearance is larger than thebearing clearance. The displacement of the seal drum is marked bythe solid triangle symbol, bearing 1 (near the honeycomb seal) bythe circle, and bearing 2 by the square.

When the honeycomb seal bore is not concentric with thecenterline between the two bearing bores, a variable called “offset”can be introduced to represent the honeycomb bore position. Toinclude this in the above calculation, the offset must be included inthe right-hand side of Equations (4) and (5) by adding it to thevariable displacements.

The honeycomb seal stiffness and damping, calculated by the sealcode developed by Kleynhans and Childs (1997) are �0.150 E9 N/m(�0.857 E6 lbf/in) and 0.200 E9 N/m (1.142 E6 lbf/in), respectively.These values are for a honeycomb seal with diverging clearance.They are for the compressor of Example B, but with slightly different


Fx

Fy

K k

k K

X

Y

C c

c C

Xdot

Ydot

= − ⋅

−

⋅

+

−

⋅

1

( )Fh x y kh x Kh yy , := ⋅ − ⋅

( )Fh x y Kh x kh yx , := − ⋅ − ⋅

( ) ( ) ( )Fb x y Fb x y Fh x y Wr

Wry y y1 1 2 2 3 3

0, , ,+ + −

=

( ) ( ) ( )Fb x y Fb x y Fh x y

Wrx x x1 1 2 2 3 3

0, , ,+ +

=

( ) ( )Lcg Wr Lhc Fh x y Lcl Fb x y

Wr Lcly y⋅ − ⋅ − ⋅

⋅=

3 3 2 20

, ,

( ) ( )Lhc Fh x y Lcl Fb x y

Wr Lclx x⋅ + ⋅

⋅=

3 3 2 20

, ,

( ) ( )y y y yhc y

HCradclear

LhcLcl

Fh x y

Krotor staticy1 2 1 3

0

3 3+ ⋅ − + + −

=,

( ) ( )x x x x

HCradclear

LhcLcl

Fh x y

Krotorx1 2 1 3

0

3 3+ ⋅ − + −

=,

internal parts, and slightly lower pressure and speed than tested forFigure 4. The cross-coupled stiffness causes the rotor center to bedisplaced to one side, instead of remaining under the center of thetilt-pad bearing (which has no cross-coupled stiffness).

Figure 8. Journal and Drum Positions in their Clearance Circles—Example B. Alternate Bundle, Diverging Clearance in Honeycomb,9965 RPM.

Figure 8 represents only one value of honeycomb seal direct andcross-coupled stiffness. The behavior of the static equilibriumcalculation varies dramatically with direct stiffness of thehoneycomb seal, when negative direct stiffness is considered.Figure 9 shows this vertical displacement of the honeycomb sealdrum as a function of honeycomb direct stiffness, over a range ofnegative seal stiffness, with all other input values held constant. Inthis figure, the drum displacement is normalized by its radialclearance while the honeycomb seal direct stiffness is normalizedby the bending stiffness of the rotor calculated at the seal location.The expected value of the direct stiffness is marked by the linelabeled “expected.” At the expected stiffness, the rotor is displacedupward to the level marked by the line labeled “Root2,” thusexplaining how the bearings may run against their top pads asshown in Figure 4.

Figure 9. Vertical Force on Bearing Journals as a Function ofArbitrary Direct Stiffness of the Honeycomb Seal.

As the honeycomb seal stiffness becomes more negative, theupward displacement increases at a larger rate, reaching 100percent of its clearance. If it were not limited by rubbing, thecalculated displacement would increase without bound, to reachthe line labeled “asymptote.” Going further left to more negative

values and the displacement suddenly changes directiondownward to negative values. This system of equations indicatesvery large forces on the bearings near a particular value ofhoneycomb seal stiffness. Mathematically, the force isunbounded and approaches an asymptote (defined in Thomas,1968) located at a particular magnitude of direct stiffness ofthe honeycomb.

A GRAPHICAL ILLUSTRATIONOF THE ASYMPTOTIC BEHAVIOR

The following explanation is intended to aid visualization of howthe asymptotic behavior occurs. It is based on the concept of solvingtwo simultaneous equations graphically. To show the concept of thegraphical solution, consider two simultaneous Equations (12) and(13) (where a, b, c, and d are numerical constants):

Each equation can be plotted as a straight line on x-y coordinates.If the two equations are independent and consistent equations thatapply simultaneously, then their solution (a particular value of xand of y that satisfies the equations) is found where the two linescross (Ayres, 1958).

To form the first line of the graphical solution, conduct a“thought experiment” on the rotor. Take the rotor running on itsbearings. Apply an arbitrary force in the positive vertical directionon the rotor at the location were the honeycomb seal acts. Measurethe displacement of the rotor at that location. Plot the arbitraryforce on the Y axis and the resulting displacement on the X axis.This is done in Figure 10, forming the red line,Y1. The slope of thered line represents the stiffness of the rotor-bearing system. Thisslope is not constant here because the static bearing stiffness actsonly in the direct axis and is not cross-coupled for the tilt-padbearings used. The nonlinearity is included in the analysis toaccurately portray the bearing characteristic at high eccentricity.Figure 11 shows the values used.

Figure 10. Graphical Solution with Positive Honeycomb Seal Stiffness.


y a x b= +*

y c x d= +*

Figure 11. Bearing Force and Bearing Static Stiffness, as aFunction of Eccentricity.

If the force at the honeycomb seal position is zero, the equilibriumposition occurs where the red line crosses the X axis (force is zerothere). In this instance, one can see in Figure 10 that equilibriumoccurs at about 25 percent of the radial clearance below the center ofthe honeycomb seal. (The X axis of the plot represents the honeycombseal drum vertical displacement normalized by radial clearance of theseal. The Y axis is force on drum normalized by rotor weight, Wr.)

Now do another experiment. Center the rotor drum in thehoneycomb seal and lift the rotor up. For a honeycomb seal withpositive stiffness the honeycomb seal will resist with a downward(negative) force on the rotor. Do this for a series of points and plot asthe blue line (Y2) as done in Figure 10. The stiffness of the honeycombseal in this plot is minus one times the slope of the blue line.

The two lines represent two equations that are solved where thelines cross. The static equilibrium occurs where the red and blueline cross, which is about 15 percent of the radial clearance abovethe center of the honeycomb seal, at the vertical line marked“Root1.” This is expected, as the positive stiffness of thishoneycomb seal is helping to support the rotor, lifting it up fromthe 25 percent position (below center) found without consideringthe honeycomb seal support. Knowing the force and displacementat the seal, the moment balance equation can be used to solve forthe force at bearing two, and then the force balance equation can beused to solve for the force at bearing one.

To demonstrate how the bearing load reaches the asymptotewhen the negative stiffness of the honeycomb increases, the aboveplot will be repeated with a honeycomb seal stiffness that has aslope near the slope of the rotor characteristic (red). This representsa honeycomb seal having negative stiffness. That is, as the drum isdisplaced away from the center, the honeycomb seal tends to pullthe drum further off center. Figure 12 shows this graphical solutionfor rotor equilibrium in the vertical plane with negative honeycombdirect stiffness.

Figure 12. Graphical Solution for Rotor Static Equilibrium forNegative Stiffness of the Honeycomb Seal.

The graphical solution is also useful to show honeycomb sealoffset from the bearing centerline. In Figures 10 and 12 the blueline is offset from the center representing an upward displacementof the honeycomb bore. The green line is offset by an equal andopposite amount showing a low honeycomb bore. The possiblerange of solutions for this offset falls between the blue and greenlines, representing an acute sensitivity to concentricity.

In Figure 12 the red and blue lines cross near 100 percent of theradial clearance below the honeycomb sea, at the displacementmarked “Root 1.” This intersection is the equilibrium position.Note that with negative stiffness small changes in the slope of theblue line can move the crossing point with the red line to very largepositive or negative values, thus representing large forces on thebearings. Of course, checking this graphical solution against thealgebraic solution, shown earlier, gives identical results. However,the graphical solution gives more insight.From this graph, the learning is to avoid negative stiffness whose

absolute value is near the stiffness of the rotor bearing system, thatis, near the asymptote. Note that the position of the asymptotedepends on the stiffness of the rotor bearing system as well as thestiffness of the honeycomb seal. Even larger negative stiffness, tothe left of the asymptote may create a larger problem, as it is likelyto depress the whirl frequency (typically the first bending mode)possibly causing the honeycomb to produce negative damping, andthus causing rotordynamic instability.

Actually plotting the graph is not necessary to find the solution,as the intersection of the rotor and honeycomb seal characteristiclines is easily found by setting the equations for the two lines equaland finding the displacement of the rotor in the honeycomb sealthat satisfies the two equations. However plotting the graph isuseful to visualize if the solution is near the asymptote.

For a rigid rotor solution it is possible to solve for the asymptotein closed form based on the lengths of the components along therotor. Given two identical bearings of stiffness Kyy, the criticalhoneycomb negative stiffness where the asymptote occurs is shownin Equation (14).

This equation only applies for zero offset.

BEHAVIOR DISPLAYEDBY THE GRAPHICAL SOLUTION

For practical application, rotor bending, rotor sag due to gravity,honeycomb seal offset from the bearing centerline, and bearingnonlinearity are usually highly significant. The honeycomb seal offsetneeds to be controlled by tolerances, or by adjustment on assembly.

For the compressor of Example B, all these effects were includedin the calculation of static equilibrium. Figure 12 gives the graphicalsolution for this case. The red line shows the force-displacementcharacteristic of the rotor-bearing system. It shows a hardeningspring rate due to the tilt-pad bearing reaching high eccentricity, asfound by a Reynolds code for laminar flow in bearings.

In Figure 12, the dashed green line shows a honeycomb seal withnegative stiffness, whose center is displaced 60 microns (2.3 mils)below the bearing centerline, shown on the graph as a displacementto the left on the horizontal axis, at about 40 percent of the radialclearance of the honeycomb seal. The intersection of the red andgreen lines shows the equilibrium position in honeycomb seal justabove center (Root2). The dashed blue line shows a honeycombseal with negative stiffness, whose center is displaced 60 microns(2.3 mils) above the bearing centerline, shown as a displacement tothe right on the horizontal axis. The intersection of the red and bluelines shows the calculated equilibrium position of the drum inthe honeycomb seal would heavily depress the rotor below itscenterline. The displacement normalized by the honeycomb seal’s


( )Kcrit KyyLb

Lb Lhc Lhc Lb:= − ⋅

+ ⋅ − ⋅ ⋅

2

2 22 2

radial clearance is near minus one, indicating the drum is close torubbing the honeycomb surface at the calculated equilibrium.

This behavior may be self-limiting, because if the rotor drumrubs the honeycomb seal surface, then, given a rotor whirlamplitude approaching the honeycomb seal clearance, the surfacewill be worn to give an axial leakage path of constant clearance.Based on calculations using the seal code developed by Kleynhansand Childs (1997), such a clearance does not usually have significantnegative direct stiffness.

Note that the red line is nearly parallel with the blue andgreen ones, which are offset by the concentricity tolerance on theposition of the honeycomb seal. The practical result is that thestatic equilibrium cannot be accurately calculated in this case,because the intersections within the tolerance band cover a rangefrom where the drum is near the bottom of the honeycomb to beinglifted above the centerline of the two bearings.

The behavior of the rotor when it is near the asymptote can beunderstood by the concept of “indifferent equilibrium.” Thisconcept applies to the case were all spring rates in the system arelinear. Indifferent equilibrium is most easily seen by an examplewhere the honeycomb seal (with negative direct stiffness) is at thecenter span of a symmetric rigid-rotor bearing assembly.

In that case the asymptote will occur when the honeycombstiffness is equal to minus the sum of the bearing stiffness (twicethe stiffness of the two identical bearings.). Thus the rotor will havezero stiffness to parallel translation, because the negativehoneycomb seal stiffness will cancel the positive bearing stiffness,as the three spring rates are additive because the springs act inparallel. (The resistance to angular displacement will remain.)

At this asymptote, if the rotor moves a honeycomb seal withnegative stiffness will pull it up, and the bearings will push down,giving a zero net force resisting the displacement. Therefore thisrotor can be moved to any lateral position without applying anexternal force, assuming perfectly linear stiffness. (Nonlinearstiffness of the bearing or honeycomb seal may not cancel at alldisplacements.) When the solution for static equilibrium is at theasymptote, a rotor is indifferent to its lateral position, and can bemoved to different positions with little or no force, provided therotor drum does not hit the honeycomb bore. However, the forceson the bearings can be very large.

INFLUENCE OF VOLUTE FORCESON ROTOR STATIC EQUILIBRIUM

In contrast to the honeycomb seal, which is characterized as aspring rate, the volute forces can be represented as a static force onthe rotor, independent of small displacements of the rotor. They canthen be incorporated into the above static equilibrium model in thesame manner as the weight of the rotor. As a simplification, thevolute force can be applied at nearly the same location as thehoneycomb seal force. However, obtaining a good estimate of thevolute force is laborious.

In order to calculate the force generated by the volute a completecomputational fluid dynamics (CFD) model of the last impellerplus diffuser plus discharge volute has been developed. Theanalysis has been carried out using the well-known mixing planeapproach. According to this technique the complete model hasbeen divided in two parts: first the impeller plus the first part ofthe diffuser, second the diffuser plus the discharge volute. Firstanalysis is done on the impeller by imposing the boundary conditionsat the inlet plus the mass flow. Once this calculation is completedtotal pressure and total temperature distributions as well as yawangle are extracted at the interface plane and are used as boundaryconditions for the CFD analysis of the volute. This analysis alsorequires imposing the mass flow (the same value applied for theCFD analysis of the impeller).

The analysis is able to capture the static pressure gradients in thetangential directions close to the outlet of the impeller. By knowingthe pressure distribution at the outlet of the impeller as well as the

geometry of the impellers, diaphragm-impeller cavities and seals itis possible to calculate the resulting radial force by means of astatic equilibrium analysis. Figure 13 shows a cross section of lastwheel stage of a reinjection compressor, Example C.

Figure 13. Cross Section of Last Wheel of a Tested Compressor.

The control volume used to make the static equilibrium of theimpeller is highlighted in red color. For the analysis the authorsassumed that the pressure gradient extends down to the diameter ofinlet eye seal on the impeller cover, and down to the hub at the footof the impeller disk. Pressure gradients have been consideredconstant along the two cavities. This assumption leads to make anoverestimation of the forces.

Figure 14 shows the result of the above analysis applied to a veryhigh pressure compressor, Example, C for the full density fullpressure conditions in the authors’ testing facility. In particularcircumferential pressure gradients are evident.

Figure 14. Static Pressure Distribution Around the Last Impeller.

Figure 15 is a graph showing the static pressure distributionimmediately at the discharge of the impeller and used in order tocompute the radial force created on the rotor. For this case thepressure equilibrium of the rotor control volume leads to a horizontalforce of �9561 N (�2149 lbf) and vertical force of 9663 N (2172lbf). The sum of these forces acts at an angle of 45 degrees fromvertical, as shown in Figure 16. The weight of the rotor is 350 kg(771.62 lb), i.e., the vertical force from the volute is around threetimes the weight. The negative value means that the force is in thesame direction of the weight.


Figure 15. Pressure Profile around Circumference of Last Impeller.

Figure 16. Direction of Net Force Due to Pressure.

Although this particular example is not correlated to the testdata, is serves to demonstrate that the volute forces can dominaterotor weight for compressors in this size and pressure range.

COMPARISON OF CALCULATIONSWITH TEST DATA

The honeycomb seal stiffness values used in this paper were allcalculated with the seal code developed by Kleynhans and Childs(1997). For Examples A and B, using a diverging clearance(leakage area increasing going downstream) in this seal code givesnegative direct stiffness. A converging clearance gives positivedirect stiffness of the honeycomb seal.

The honeycomb seal for Example B was modified after themeasurements in Figures 4 and 5 were made. The modification wasmade when an opportunity arose to limit the possibility of highbearing loads. Swirl brakes were used on the honeycomb seal to avoidsubsynchronous instability. Using the operating (hot and spinning)clearance of the honeycomb seal for Example B, the calculated directstiffness at zero frequency before and after changing the clearance isshown in Table 3 for the respective clearances.

Table 3. Calculated Direct Stiffness at Zero Frequency Before andAfter Changing the Clearance.

After the honeycomb seal was modified to have positivestiffness the compressor was retested. The measured bearingjournal positions are shown in Figure 17. The journal in thenondrive-end (NDE) bearing did not lift high but is pushed to theside. The journal in the drive-end (DE) bearing (next to thehoneycomb seal) still lifted to the top pads. The volute forces werenot calculated, as the effort was not justified by the good behaviorof the compressor.

Figure 17. Calculated Static Force by Bearings for Example B,before Modification of the Honeycomb Bore.

Figure 18 shows the calculated static force on the rotor by bearingsfor Example B before modification of the honeycomb bore. Theoperating conditions for Figure 18 are estimated to correspond toFigure 4 at maximum speed. The offset used corresponds to the greenline in Figure 12. The vertical line labeled “expected” in Figure 18represents the Keff shown in the above table for the rated conditions.At the expected Keff, the forces at both bearings are negative,pushing down on the rotor, in the same direction as shown in Figures4 and 5. Note that the bearing force is large, being nearly twice therotor weight on bearing number one. Because the volute forces arenot considered, the calculation is not empirically confirmed.However, the calculated and measured journal conditions are notcontradicted either.

Figure 18. Journal Positions in the Bearings of Example B afterRevising the Honeycomb Seal.

Figure 19 shows the calculated force on the rotor by the bearingsfor Example B as modified. The offset used corresponds to the blueline in Figure 20.The vertical line labeled “expected” representsanother value of the Keff = +0.259 E9 N/m (+1.43 E6 lbf/in) forthe as-modified conditions. At the expected Keff, the force fromBearing 1 (on the DE of the rotor) is pushing down, as shown on


the right-hand side of Figure 19. At the expected Keff, the forcefrom Bearing 2 (on the NDE of the rotor) is pushing down slightly,as shown in the right-hand side of Figure 19. Note that the Bearing2 force is small, with its absolute value being less than half therotor weight. Without the honeycomb seal force or volute forces,the bearing force would be about half the rotor weight, pushing up.The graphical solution predicting the rotor equilibrium as modifiedis shown in Figure 20.

Figure 19. Calculated Static Force by Bearings for Example B,after Modification of the Honeycomb Bore.

Figure 20. Graphical Solution Corresponding to Figure 19.

Figures 19 and 20 are consistent between the calculation andtest, assuming a known offset (tolerance) of the honeycomb bore.However, the actual offset is not known. Note that the offsets areopposite between Figures 19 and 18. The actual offsets wereunknown. Alternate compressor internal parts (rotor and bundle)were installed for the test shown in Figure 17 than were used in thetest shown in Figure 4. Therefore, the offsets may have changedbetween Figure 18 and 19.

DISCUSSION

The basic assertion of this paper is that honeycomb seal andvolute forces in reinjection compressors can invalidate the bearingload calculations typically used in rotordynamics analysis to findthe bearing characteristic. A practical solution is to use high

preload and tight clearances on tilt-pad bearings to make the bearingless sensitive to load. The disadvantage of high preload and tightclearances is loss of opportunity to maximize the bearing dampingand thus optimize unbalance response and rotordynamic instability.

The measured rotor position in the bearings of the examples wasoften against the top pads, in opposition to the standard assumptionof gravity load. A more unexpected conclusion, from the equationsof static equilibrium, is that a honeycomb seal having a negativestatic stiffness in the same range as the rotor-bearing systemstiffness (that is, at the asymptote) can produce bearing loads ofmany times the rotor weight. These calculations suggest that sucha load would only be bounded by contact of the rotor against thehoneycomb (based on a linear stiffness characterization of thehoneycomb). However, the contact force will be small due to theindifferent equilibrium effect discussed above. (The large force onthe rotor in the honeycomb is carried by the gas pressure.) In thecases presented, the calculations predict the asymptote was notreached. Only moderate contact was noted on disassembly.

The solution of the six equations of equilibrium (Equations 6through 11) is straightforward and well based on the theoryof statically indeterminate problems (Popov, 1968.) However,conceptually graphic solution gives more insight into the behaviorof the system.

The concept of static equilibrium could also be applied tohoneycomb seal test rigs used for determining the dynamiccoefficients. In the case of test rigs where the seal carrier moveswith respect to a stiffly supported rotor, the direct stiffness of thetie-rod assembly could be used in the calculation in place of therotorbearing system stiffness. When testing seals with negativedirect stiffness contact with the rotor may occur. The graphicalsolution presented in this paper shows such contact will occurwhen the absolute value of negative stiffness of the honeycombapproaches the positive stiffness of the mechanical parts of thesystem (when approaching the asymptote). This is conceptuallydifferent from reaching contact only at very large values ofnegative stiffness.

The test results presented prove that the bearing loads are notdue to gravity alone, as it is customary to assume. The calculationspresented here confirm that negative stiffness from the honeycombseal, and volute loads should have a dramatic effect on the staticequilibrium of the rotor on its bearings. Due to strong influence ofthe unknown magnitude and direction of offset between thehoneycomb bore and the bearing centerline, the calculationssuggest the bearing loads are not calculable, given typical offsettolerances. Furthermore, the aerodynamic force on the rotor canvary so that the estimated bearing loads may vary over a widerange. “Indifferent equilibrium” further confounds the problem.

Nevertheless, the analysis presented requires that the compressordesign must take the variability of the possible bearing loads intoconsideration. For oil film bearings this is easily considered in thedesign stage by calculating the loads at the extremes of offset.Magnetic bearings and bearings supported on squeeze filmdampers typically have direct stiffness that is an order ofmagnitude softer than the tilt-pad journal bearings used inExamples A and B. If magnetic bearings or damper bearings areused with a honeycomb seal, especially one having negativestiffness, the calculated asymptote would occur at a much smallernegative stiffness of the honeycomb seal than with stiffer bearings.Such behavior could disrupt the functioning of a magnetic bearingor damper bearing even if the honeycomb negative stiffness wererelatively small in absolute value.

CONCLUSIONS

Calculation of the bearing load based on only the rotor weight (forhorizontal rotors) is not valid for high pressure centrifugal compressorshaving honeycomb/drum-rotor seals, or single-volute or collectortype discharge hardware. The forces caused by these components areshown for the above examples to exceed the rotor weight.


Introducing the honeycomb/drum type seal into the bearing loadcalculation as direct and cross-coupled spring coefficients makesthe problem statically indeterminate and thus requires considerationof displacement of the journal in its bearing, the drum in thehoneycomb bore, and the rotor as a beam in bending. In this casethe bearing loads can be solved by the standard methods of themechanics of solids.

When the spring coefficient of the honeycomb seal is negative,the indeterminate beam calculation can show asymptotic behavior.In this case the drum displacement will only be limited by contactwith the honeycomb seal bore. At the asymptote, the rotor can beeither lifted or depressed.

This paper introduces a conceptually graphical solution tothe indeterminate beam calculation for the purpose of visualizingthe asymptotic behavior (here shown in a vertical plane.)This solution also makes clear the importance of small deviationsof the honeycomb bore from the centerline between thetwo bearings.

The asymptotic behavior does not depend merely on themagnitude of the negative stiffness of the honeycomb seal, butinstead depends on the ratio of that stiffness to the stiffness of therotor bearing system. In the graphical solution this occurs when theslope of the honeycomb seal displacement versus force line is nearlyparallel with the slope of the rotor-bearing system displacementversus force line.

This solution predicts the rotor can be lifted in its bearings by thenegative stiffness of the honeycomb seal, at sufficiently high gaspressures. This solution also predicts that the rotor can be liftedin its bearings by positive stiffness of the honeycomb seal, whenthe honeycomb bore is offset above the centerline between thetwo bearings.

For examples given in this paper, these predictions are notinconsistent with the observed behavior. However, the unknownvalues of honeycomb bore concentricity defeat an exact calculation,because the bearing position is acutely sensitive to the honeycombbore offset. The volute forces compound this problem.

NOMENCLATURE

fb1 N (lbf) = Vertical load on bearing number 1fb2 N (lbf) = Vertical load on bearing number 2Wr N (lbf) = Vertical force due to rotor weightLcg mm (inch) = Horizontal distance from bearing

1 to the rotor center of gravityLhc mm (inch) = Horizontal distance from bearing

1 to the rotor center of the sealLcl mm (inch) = Distance between horizontal centers

of the two bearingsFbx(x1,y1) N (lbf) = Force by the bearing on the rotor

in the horizontal direction due todisplacements x1 (direct) and y1(cross-coupled)

Fby(x1,y1) N (lbf) = Force by the bearing on the rotorin the vertical direction due todisplacements y1 (direct) and x1(cross-coupled)

Fhx(x3,y3) N (lbf) = Force by the honeycomb sealon the rotor in the horizontaldirection due to displacementsx3 (direct) and y3 (cross-coupled)

Fhy(x3,y3) N (lbf) = Force by the honeycomb seal onthe rotor in the vertical directiondue to displacements y3 (direct)and x3 (cross-coupled)

K N/m (lbf/in) = Radial spring coefficient oflinearized restoring direct force

k N/m (lbf/in) = Tangential spring coefficient oflinearized cross-coupled force

C N*s/m (lbf*s/in) = Damping coefficient—directc N*s/m (lbf*s/in) = Damping coefficient cross-coupledx1, x2 microns (mil) = Horizontal displacement of rotor

in bearing 1 and 2, respectivelyy1, y2 microns (mil) = Vertical displacement of rotor in

bearing 1 and 2, respectivelyXdot mm/sec (in/s) = Horizontal velocity of whirl orbitYdot mm/sec (in/s) = Vertical velocity of whirl orbitKyy = K N/m (lbf/in) = Bearing direct stiffness in the

vertical directionLb mm (inch) = LclKcrit N/m (lbf/in) = Honeycomb seal stiffness at

asymptote for the rigid rotor,linear bearing case

Keff N/m (lbf/in) = Direct stiffness of the honeycombseal at zero frequency

Subscripts

1 = Bearing 12 = Bearing 23 = Drum of honeycomb seal

REFERENCES

Ayres, F. Jr., 1958, FirstYear College Mathematics,NewYork, NewYork: McGraw-Hill Book Company, p. 28.

Camatti, M., Vannini, G, Fulton, J., and Hopenwasser, F, 2003,“Instability of a High Pressure Compressor Equipped withHoneycomb Seals,” Proceedings of the Thirty-SecondTurbomachinery Symposium, Turbomachinery Laboratory,Texas A&M University, College Station, Texas, pp. 39-48.

Kleynhans, G. W. and Childs, D. W., 1997, “TheAcoustic Influenceof Cell Depth on Rotordynamic Characteristics of Smooth-Rotor/Honeycomb-Stator Annular Gas Seals,” Journal ofEngineering for Gas Turbines and Power, 119, pp. 949-957.

Popov, E. P., 1968, Introduction to Mechanics of Solids, EnglewoodCliffs, New Jersey: Prentice-Hall, Inc., Chapter 12, “StaticallyIndeterminate Problems.”

Thomas, G. B., Jr., 1968, Calculus and Analytic Geometry, FourthEdition, Reading, Massachusetts: Addison-Wesley PublishingCompany, p. 317.

BIBLIOGRAPHY

Mathcad, 2007, Version 14, Parametric Technology Corporation,Needham, Massachusetts.

Wagner, N. G., 1999, “Reliable Rotor Dynamic Design of HighPressure Compressor Based on Test Rig Data,” Transactions oftheASME, Journal of Engineering for Gas Turbines and Power,October 2001, 123, pp. 849-856.

ACKNOWLEDGEMENTS

The authors thank General Electric Oil & Gas, and ExxonMobilResearch & Engineering for permission to publish this paper.Many thanks also to Luca Poli for his support and contributions,and also to Vittorio Michelassi in particular for the contributiongiven in the aero CFD analysis.


Steven M. Schultheis is a SeniorEngineer in Shell Global Solutions(US) Inc., in Houston, Texas. His primaryresponsibilities are to support Shellworldwide in regard to condition monitoringand vibration analysis of rotatingmachinery. In addition, his specialties arein the areas of machinery and pipingdynamics, rotordynamics, structuralvibration, and pulsation.

Mr. Schultheis received a B.S. degree (Mechanical Engineering,1984) from New Mexico State University, is a registeredProfessional Engineer in the State of Texas, and is certifiedCategory 4 vibration specialist through the Vibration Institute. Hehas published and presented at meetings of a number of technicalsocieties and symposium including the Texas A&MTurbomachinery Symposium, theVibration Institute, the NPRA, theASME, and others.

Charles A. (Chuck) Lickteig is a SeniorEngineer in Shell Global Solutions (US)Inc., in Houston, Texas. He is responsiblefor providing technical support for rotatingand reciprocating machinery at Shellaffiliated company’s worldwide as wellas commercial customers. Mr. Lickteigis currently developing advancedperformance and condition monitoringapplications. Since joining Shell in 1990,

he has had assignments in projects involving specification,evaluation, systems integration, installation, commissioning, andstartup of rotating equipment for existing and new process plants,as well as extensive field troubleshooting in performance andvibration analysis.Mr. Lickteig has a B.S. degree (Mechanical Engineering, 1990)

from the University of Kansas, and is a registered ProfessionalEngineer in the State of Louisiana.

Robert Parchewsky is a Principal Engineer in Shell GlobalSolutions International, in Calgary, Alberta, Canada. He is

responsible for managing the implementation of global standardsand systems on condition monitoring of rotating equipment forremote diagnostics. He has worked in a variety of business units atShell including Chemicals, Oilsands, Refining and Exploration,and Production. Mr. Parchewsky is currently developing advancedperformance and condition monitoring applications. He hasprovided support on new project design, troubleshooting, failureanalysis, condition monitoring, performance assessments,standards’ development, and field support.Mr. Parchewsky received a B.Sc. degree (Mechanical Engineering,

1988) from the University of Saskatchewan, and is a registeredProfessional Engineer in the Province of Alberta in Canada.

ABSTRACT

Technology for reciprocating compressor condition monitoringhas been around since the 1950s. However until the last 15 years orso it seemed that only the pipeline companies spent much effort onthis activity. Technology has advanced, and there are very effectiveapproaches to monitoring and protecting reciprocating compressorson the market today. While pipeline operations are pulling out theirreciprocating compressors, this machine is still the workhorse ofrefineries, chemical plants, and oil production facilities. As a resulta new generation of interest has developed in effective conditionmonitoring of reciprocating compressors. This paper will discussrisk-based decision making in regard to measurements andprotective functions, online versus periodic monitoring, proven andeffective measurement techniques, along with a review of bothmechanical- and performance-based measurements for assessingmachine condition. Case histories will also be presented todemonstrate some of the concepts.

INTRODUCTION

Each year at the Turbomachinery Symposium in the reciprocatingcompressor discussion group the focus of the discussion isprimarily on condition monitoring. With all the other technicalissues related to reciprocating compressors that could be discussed,this is usually the topic that generates the most interest. Pasttopics have included vibration monitoring, rod drop monitoring,pressure-volume analysis, and temperature measurements. Inthese discussions there are several thematic questions that havecome out:

107

RECIPROCATING COMPRESSOR CONDITION MONITORING

bySteven M. Schultheis

Senior Engineer

Charles A. LickteigSenior Engineer

Shell Global Solutions (US) Inc.

Houston, Texas

andRobert Parchewsky

Principal Engineer

Shell Global Solutions International

Calgary,Alberta, Canada

How is condition monitoring of reciprocating compressorsjustified?

• What measurement techniques are really effective?

• Should these parameters be used as protective functions or not?

• Should the measurements be online or periodic?

• What can I do easily to improve the monitoring on my existingmachine, especially without doing a major retrofit project?

• Is it really worth it?

All legitimate questions, but unfortunately there is not oneanswer for all equipment. Based on risk, a compressor in arefinery hydrotreater service may require a very differentcondition monitoring approach than a gas lift machine inan oil field application, or a nitrogen compressor in achemical plant. Sorting out the risk and applying the rightmonitoring approach is the primary objective of any condition-monitoring project.

AN OVERVIEW OF PROVENMEASUREMENT TECHNIQUES

For centrifugal compressor trains there is good agreementacross industry on how to effectively monitor machine condition.These approaches are summarized in American PetroleumInstitute (API) Standard 670 (2000), and there is little question asto what suite of instrumentation will be used to monitor centrifugalmachines. When it comes to reciprocating compressors inprocess plants, there is much less agreement on which monitoringtechniques should be standard but at least API 618 (1995)contains some basic requirements. For an ISO 13631 high speedreciprocating compressor, particularly in oil field service, thereis even less agreement in the industry on applicable monitoringsystems. API Standard 618 (1995) monitoring and protectionrequirements include high discharge temperature, low framelube-oil pressures and level, cylinder lubricator system failure,high oil filter differential pressure, high frame vibration, highlevel in the separator, and jacket water system failure. APIStandard 670 (2000) describes the requirements for installingproximity and casing transducers on reciprocating compressors,but the details of what measurements to make, and how toapply those measurements are left to the original equipmentmanufacturer (OEM) and the purchaser to decide. The followingare some techniques that have been proven effective across awide range of applications.

Vibration

There are two primary vibration measurements that have beenproven effective; measurement on the crankcase, and measurementon the crosshead/distance piece. This is due to the way forces areapplied in reciprocating machines. The most common machinedesign is a balanced opposed configuration and in this configurationthe reaction forces in the cylinders are balanced across the machineby the opposing cylinder. Since the cylinders are offset, momentsare also set up in the crankcase, and these moments are balanced asmuch as possible by cylinder placement and cylinder timing. Thebalance of forces and moments is fine-tuned by designing theweight of reciprocating parts and by applying counterweights onthe crankshaft. But it is rare for the balance to be perfect and if amalfunction occurs that upsets this balance of forces and moments,the result is high vibration at 1× or 2× running speed. The flipside is that if the machine support stiffness is reduced, forinstance due to grout deterioration or the loosening of foundationbolts, then even in the presence of normal forces and momentsthe vibration will increase. Catastrophic events such as breakinga piston rod or losing a counterweight result in a sudden increasein unbalanced forces and moments, and may result in very highcrankcase vibration.

Figure 1. Crankcase Vibration Transducers.

Crankcase vibration has been used as a basic protectionparameter for decades, usually with a mechanical “earthquake”switch. API 618 (1995) has specifically eliminated the mechanicalswitch as an option because of instrument reliability issues. Thismeasurement is now mostly made with solid-state electronicdevices. Most compressor OEMs specify a crankcase velocityalarm/shutdown level and the most common configuration is tomount the transducers on each end of the crankcase about halfwayup from the baseplate in line with a main bearing (Figure 1). Thisconfiguration enables the transducers to see the effects of bothunbalanced forces and moments while minimizing the number ofmeasurements. It is then also possible to monitor changes inrelative vibration amplitude and phase in the event that groutbreaks down or forces and moments become unbalanced in themachine. An alternate approach is to install a transducer at eachmain bearing, but this is much less common. The advantage of thisapproach is that information is available online regarding theoperating deflection shape of the crankcase.

Since the purpose of these transducers is to measure runningspeed related vibration, a low frequency transducer is required, andvelocity is the normal measurement parameter. The crankcasevibration acceptance criterion used by most OEMs is also invelocity. An option is to use a low frequency accelerometer andrun the signal to a dual path monitor, allowing both themeasurement of low frequency running speed related vibration aswell as the high g excursions associated with a major impact eventin the crankcase.

The other primary vibration measurement that has proven tobe effective is to measure acceleration on the crosshead ordistance piece of each cylinder (Figure 2). In this case the idea isto measure the mechanical response of the assembly to impactevents. Malfunctions such as liquid carryover, loose piston nuts,loose crosshead attachment valve problems, clearance problems,and many others can be identified with this measurement. Formore than 100 years mechanics have been listening to the knocksand rattles of reciprocating compressors to assess the health oftheir machines. An accelerometer on the distance piece of acylinder is in essence a microphone that allows those sounds tobe recorded, and allows for consistent alarms and shutdowns. Ofall the vibration measurements that could be made, this isprobably the most effective vibration protection measurementavailable. If a machine is undergoing catastrophic distress, it willtypically be picked up on the crosshead accelerometers. As aresult, even for very small, spared, or noncritical compressorsin hydrocarbon service a simple (and cheap) accelerometermeasurement on the crosshead or distance piece of the machineis easily justified.


•

Figure 2. Crosshead Accelerometer Installation. (Courtesy ofBently Nevada)

An extension of vibration technology is ultrasound analysis.Ultrasound has proven to be the preferred approach to analysis ofvalve condition. Ultrasonic energy is most often associated withgas leaks, so a valve that leaks is a strong generator of ultrasonicenergy. Ultrasound measurements are usually taken in conjunctionwith compressor pressure-volume analysis, which will bediscussed later (Figure 3).

Figure 3. Pressure and Ultrasound Traces for Valve Analysis.

Critical to successful implementation is properly set trip levelsthat are just high enough over the normal operating level to react tomechanical failures, but not so high as to miss the failure prior tocatastrophic release. Unfortunately, even when systems areproperly configured, catastrophic failures often still occur due tooperation’s restarting the machine to confirm that the trip wasreal (Case History 1). In these cases when restarted the failureis escalated, leading to a large consequence failure. Properoperating procedures need to be in place to ensure that reciprocatingcompressors are not restarted without proper inspection andengineering assessment.

Temperature

Machine temperatures are a valuable indication of machinecondition and are a primary tool for reciprocating compressorcondition monitoring. The primary temperature measurementsinclude cylinder discharge temperature, valve temperature, packingtemperature, crosshead pin/big end bearing temperature, and mainbearing temperature. Cylinder discharge temperature is one of theprotection parameters recommended byAPI 618 (1995) since leaksin rings and valves result in recompression of gas that will raise the

discharge temperature. Valve temperatures have proven to bevaluable in identifying individual valve problems, but are mosteffective if the measurement is made in a thermowell in the valvecover (Figure 4) so that the reading is taken close to the valve platesor poppets, rather than measuring contact temperature of the valvecovers. Packing case temperature or packing leak off temperaturecan give an indication of packing leakage, while main bearingtemperature measurement has been proven effective in preventingmajor damage due to main bearing failures. Many machines havebeen saved by shutdowns associated with eutectic or “turkeypopper” temperature devices in wrist pin and big end bearings, butthere are technology developments currently underway to makethis a wireless measurement allowing for a temperature trend.Fossen and Gemdjian (2006) describe one such technology in theirpaper on radar-based sensors. This and other similar technologiesare a major improvement in the protection and condition assessmentof connecting rod bearings.

Figure 4. Valve Temperature Measurement, Thermo Wells in theValve Covers.

Rod Drop/Rod Runout

In their paper, “Rod Drop Monitoring, Does it Really Work,”Schultheis and Howard (2000) discuss the significant problemsassociated with rod drop monitoring. While advances have beenmade to this technology since then, it is still a difficult measurementwith many sources of error. At the same time, rider band wear andfailure is still a significant maintenance issue with many machinesand as a result rod drop monitoring still has a place in thereciprocating compressor condition monitoring toolbox (Figure 5).

Figure 5. Rider Band Failure.

RECIPROCATING COMPRESSOR CONDITION MONITORING109

An underutilized parameter however is rod runout monitoring.Rod drop measurement is accomplished by using the directcurrent (DC) component of the proximity probe signal, which isproportional to position. This locates the average distancebetween the probe tip and the target. In a reciprocating compressorproximity probes are typically located under the piston rods, andare used to measure the rod position, which can be converted torider band wear (Figure 6). There is also an alternating current(AC) component of the probe signal, which is representative ofoperating rod runout. Cold runouts are usually held to about2 mils peak-to-peak, and operating runouts under normalconditions are typically somewhat greater than that, on the orderof 2 to 6 mils peak-to-peak. In the event of a malfunction such asa cracked piston rod attachment, a broken crosshead shoe, oreven a liquid carryover to a cylinder, the operating rod runoutwill increase significantly. These are data that have historicallybeen ignored, since the rod drop monitors on the market wereonly measuring the DC component of the signal. Parallelingthe signal from rod drop probes into a vibration channel, orconfiguring the rod drop monitors to display rod runout inaddition to rod drop, adds tremendous value to the condition dataset on a reciprocating machine.

Figure 6. Rod Drop Probe Installation. (Courtesy of Bently Nevada)

PV Analysis

Pressure velocity (PV) analysis is a technique that has provento be very effective in assessing the condition of reciprocatingmachinery and has been used for more than 50 years. Personalcomputer technology has significantly reduced the cost of thiskind of measurement, and improvements in transducer technologyhave overcome the technical obstacles such that PV analysis isnow available online. Dynamic pressure transducers are used tomeasure the pressure inside the cylinder over the course of thestroke (Figure 3). This allows the analyst to evaluate thecondition of the rings, valves, and packing, while at the sametime calculating the dynamic rod load, which is the source of theforces and moments described in the discussion on vibration.This requires that a pressure transducer be installed in thecylinder, either on a temporary basis using a valved port (Figure7) or for an online measurement using a permanent transducerinstallation (Figure 8). As described in the next section, it is oftenpossible to predict the PV diagram from existing instrumentationand modeling.

Figure 7. PV Measurement Valves on the Side of a Cylinder.

Figure 8. Permanent Pressure Transducer Installation Through aValve Cover. (Courtesy of Bently Nevada)

EFFECTIVELY USING THEDATA ALREADY AVAILABLE

In many reciprocating compressor applications, a tremendousamount of data is gathered as a matter of course, and while they aredisplayed to the operator through a distributed control system(DCS) or other control system it is otherwise wasted. In many ifnot most applications, those data are stored, most often in a datahistorian, but this is functionally like the black box on an aircraft;data gathered and stored so that they are available in the event of aproblem. Those data are in essence underutilized and since they areavailable they should be used as a part of the condition monitoringfor a compressor, rather than leaving them to collect dust untila problem arises. One approach to utilizing these data that isproving successful is to use a modeling program that comparesthe theoretical performance of the compressor to the actualperformance in order to determine if the compressor performanceis deviating from predictions. Such a model can easily reside on thesame computer with the data historian, gathering its input from thatdatabase, and writing results back to the historian. The modelingprogram is first populated with design data and physical propertiesof the compressor, e.g., piston diameter, stroke, rod diameter, etc.Then the model is validated against predicted performanceinformation. Next the model is integrated into the network wherethe data reside and data on the actual process conditions are fed


into the model to determine the actual performance of the machine.For a given set of process conditions the theoretical and actualperformance of several compressor performance variables can becalculated. Finally performance and deviation from theoretical arewritten back to the data historian for trending. Software alarms onthe calculated values can also be generated to alert operations ormaintenance of developing performance issues.

The theoretical model uses first principal type equations tocalculate: volumetric flow rate, volumetric efficiencies, power, rodloads, interstage pressure, outlet temperatures, and rod reversal.Each of these values for each cylinder or stage is then evaluatedagainst the design or actual values to determine if the performanceof the compressor is deviating from prediction. Deviations generallyindicate degradation in cylinder wear parts or off-design operatingconditions. There may be accuracy issues associated with the fieldinstrumentation. However trending performance deviations from abaseline can be as valuable as absolutely accurate performance data.

While it is true that a direct measurement of the compressorpressure volume trace utilizing a dynamic pressure transducergives the most accurate picture of cylinder condition, a performancecalculation based on the pressures, temperatures, and flows alreadyavailable in the process control system provides a valuableoverview of the condition of the machine. Establishing a baselineof performance when it is known that there are no problems withthe machine makes it possible to identify deviations from normal,alerting operations and maintenance to a developing problem, andperhaps giving operations a chance to take action to mitigate theproblem before it causes a machine failure.

A RISK-BASED APPROACHTO CONDITION MONITORING

The most common objections to condition monitoring onreciprocating compressors include: “We never needed it before,why now? We don’t have problems with these machines, whymonitor them? Monitoring is too expensive.” All of this maybe true, it may not be necessary, the machine may not haveproblems, and a complete system may be expensive. However if thecompressor is moving hydrocarbon gas, it becomes easy to justifyat least a simple accelerometer-based vibration system costing onthe order of $1000 USD, to shut down the machine in the event ofa catastrophic event and to install a performance calculationtool using existing instrumentation and databases. Unfortunatelypurchasing a condition monitoring system is much like buying aninsurance policy. If there is never a failure then there is really noneed to have it. One thought to consider is that at least 10 percentto 20 percent of all reciprocating compressors (based on theauthor’s experience) suffer a catastrophic failure or a failure thatcould have been catastrophic if protection systems did not stop theevent from progressing. In buying insurance, or in buyingcondition monitoring, the first step is to assess the risk, and thenpurchase what is appropriate to mitigate the risk specific to themachine and the service. The outcome of the risk assessmentshould be a list of parameters that will be used to protect themachine, as well as parameters for determining machinerycondition. The approach to risk assessment is usually a risk matrixthat includes aspects of safety, business impact, environmentalimpact, and reputation impact. Figure 9 is an example of such arisk matrix. The outcome from this kind of analysis can be either alevel of criticality, a safety integrity level, or a standard monitoringapproach. One of the ways this type of analysis can be used is todetermine if monitoring should be online or periodic. Machineswith high criticality ranking will typically require online monitoringand protection. Machines that are spared, and thus representsignificantly less risk, may only require minimal protection withperiodic monitoring. Machines with high criticality ratings may notonly justify a complete set of monitoring and protection instruments,it might also make sense to establish the information technology (IT)infrastructure for remote monitoring and diagnostics.

Figure 9. Risk Assessment Matrix.

CASE HISTORIES

Case History 1

A 1500 kW compressor was installed in 2002 that wasdeemed critical enough to justify an advanced monitoring and dataacquisition system with remote monitoring and diagnostic capability.After 3000 running hours a failure occurred at the crosshead forcylinder 1. The machine had experienced a four-hour shutdownafter a power failure and when this was resolved a restart wasinitiated. Shortly after startup a trip occurred due tovibration/impact at the accelerometer sensor for cylinder 1. Theweather was cold and stormy (0�F), and after examination of thecrosshead the failure was obviously a matter of poor lubrication.This compressor had one common oil supply line to the top andbottom crosshead guides and based on the design there wascertainly a preference in the oil supply to the top. That issue, incombination with the low temperature and associated viscositychanges in the oil along with startup conditions, resulted in a lackof oil supply to the crosshead and the resultant failure (Figure 10).The compressor tripped on impact level set in the online monitoringsystem, which prevented the machine from more extensive damageand longer outage time. However, the system showed a changeearlier in both vibration and temperature. Unfortunately noresponse was taken to this change in vibration level and temperatureincrease until the failure occurred (Figures 11 and 12).

Figure 10. Damage to the Lower Crosshead Guide.

Figure 11. Vibration Trend Prior to Failure (About 1g).


Figure 12. Vibration Trip on 9 January, Spiked up to 10g or More.

Case History 2

At a Canadian gas plant, a series of reciprocating compressorfailures occurred due to yielding of the aluminum pistons fromhigh internal pressure loads resulting in rod failures (Figure 13).The compressors were in acid gas service with high molecularweight gas, which resulted in high valve losses. By institutingonline performance monitoring of the compressor through the useof a compressor performance model interfaced to the datahistorian, and by highlighting to operations the appropriateoperating envelope based on the results of the performancecalculations, repeat failures have been eliminated. The same viewerwas used for the data historian that the operators were alreadyusing for other tasks, so there was little training required to makethis tool an effective operator interface (Figure 14). Onlinemonitoring utilizing existing instrumentation and requiring noadditional capital expenditures or machine modifications resultedin multimillion-dollar savings as the failures were eliminated.

Figure 13. Piston Rod Failure.

Figure 14. Data Display-Rod Load Curves.

Case History 3

A broken crosshead slipper was identified in the periodic datacollection on the machine. During regular PV data gathering, theanalyst also routinely gathered vibration data on the crosshead.Figure 15 shows the vibration trace of the cracked crosshead slipperon the top, along with two historical traces. The vibration signal onthis crosshead has changed significantly, and the predominantfeature is the impact ring down seen in the waveform between 240and 300 degrees. The crack in the slipper is shown in Figure 16.

Figure 15. Routine Crosshead Vibration Trace Shows Developmentof the Vibration as the Crosshead Crack Progressed.

Figure 16. Broken Crosshead Slipper.

Case History 4

Analysis identified a leaking crank-end discharge valve as thereason for decreased performance. Inspection of the dischargevalve during replacement showed what looked like pieces of pistonring or rider band material in the valve. The piston was removedfrom the cylinder and it was confirmed that the piston rings werebroken into segments and the edges of the rings were broken(Figure 17). The pressure time (PT) trace (Figure 18) shows thecrank-end discharge valve, 1CD3. Ultrasonic data indicated thisvalve leaking during the suction portion of the stroke. Although thepiston rings were modified by the failure to have more than oneend gap, they stayed in the piston ring groove, between the pistonand the cylinder, and performed their intended sealing purpose.There was probably more bypass than intended, but it was notnoticeable in the flow rate.

Figure 17. Broken Piston Rings Laid out on the Cylinder SupportPedestal During Disassembly.


Figure 18. Pressure Time and Ultrasonic Trace Showing Crank EndDischarge Valve Leak.

Case History 5

Again crosshead acceleration measurement saves the day as ittracks the development of a loose piston nut. The accelerationtrace (Figure 19) changed significantly between the May and Junemeasurements, and there is a distinct impact/ring down patternpresent after top dead center (TDC) and bottom dead center(BDC) indicating an impact as the load shifts from tension tocompression and back.

Figure 19. Crosshead Acceleration Traces Showing theDevelopment of a Loose Rod Nut. Crosshead Knocks after TDCand BDC (Ring Down).

CONCLUSIONS

Reciprocating compressor condition monitoring has developedwell beyond theory to the point that there are a number of provenand practical approaches to monitoring and protection of thesemachines. The first step to effective condition monitoring is to takeadvantage of the information already available on the machine.From there a well thought out risk analysis should be used to guidethe application of protective functions and condition analysisparameters, as well as determining whether monitoring should beonline, periodic, or a combination of both. For normal wear parts

such as valves, rings, and packing, some form of performanceanalysis is key to the early identification of cylinder problems,allowing for maintenance cost reduction through proactivemaintenance planning.

When properly configured and utilized an online conditionmonitoring system that is tailored to the criticality of theequipment can significantly reduce the likelihood of catastrophicfailures, and has the capability to create proactive action fromoperations, while being cost effective and repeatable. Onlinemonitoring can be quite effective even if only utilizing existinginstrumentation and vibration transducers.

REFERENCES

API Standard 618, 1995, “Reciprocating Compressors forPetroleum, Chemical, and Gas Industry Services,” FourthEdition, American Petroleum Institute, Washington, D.C.

API Standard 670, 2000, “Vibration, Axial-Position, and Bearing-Temperature Monitoring Systems,” Fourth Edition, AmericanPetroleum Institute, Washington, D.C.

Fossen, S. and Gemdjian, E., 2006, “Radar Based Sensors—ANew Technology for Real-Time, Direct TemperatureMonitoring of Crank and Crosshead Bearings of Diesels andHazardous Media Reciprocating Compressors,” Proceedings ofthe Thirty-Fifth Turbomachinery Symposium, TurbomachineryLaboratory, Texas A&M University, College Station, Texas,pp. 97-102.

BIBLIOGRAPHY

Leonard, L., June 1996, “The Value of Piston Rod VibrationMeasurement in Reciprocating Compressors,” Orbit Magazine,pp. 17-19.

Schultheis, S., June 1996, “Vibration Analysis of ReciprocatingCompressors,” Orbit Magazine, pp. 7-9.

Schultheis, S. M. and Howard, B. F., 2000, “Rod Drop Monitoring,Does it Really Work,” Proceedings of the Twenty-NinthTurbomachinery Symposium, Turbomachinery Laboratory,Texas A&M University, College Station, Texas, pp. 11-20.

Smith, T. and Schultheis, S., 1997, “Protecting and ManagingYourReciprocating Compressors,” Proceedings of the 1997 NPRARefinery and Petrochemical Plant Maintenance Conference,pp. 85-98.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the following people forcontributing to the pictures and case histories:

Dwayne Roberts, Shell Norco Chemical, Norco, Louisiana; BillRobertson, Staff Machinery Engineer, Oilsands Projects, ShellCanada Limited; Bert van het Maalpad, Shell, NederlandseAardolie Maatschappij B.V., Netherlands; Brian Howard, GEOptimization and Control, Seattle, Washington; Steve Stoddard,Shell Chemical, Mobile, Alabama.


David L. Ransom is a Senior ResearchEngineer at Southwest Research Institute,in San Antonio, Texas. His professionalexperience over the last 10 years includesengineering and management responsibilitiesat Boeing, Turbocare, and Rocketdyne. Hisresearch interests include rotordynamics,structural dynamics, seals and bearings,finite element analysis, and root causefailure analysis. He has authored eight

technical papers in the field of rotordynamics and thermodynamics.Mr. Ransom received his B.S. degree (Engineering Technology,

1995) and M.S. degree (Mechanical Engineering, 1997) fromTexasA&MUniversity. He is also a licensed Professional Engineer in theState of Texas.

ABSTRACT

Root cause failure analysis is a process for identifying thetrue root cause of a particular failure and using that information toset a course for corrective/preventive action. From a technicalstandpoint, it is usually a multidisciplinary problem, typicallyfocused on the traditional engineering fields such as chemistry,physics, materials, statics, dynamics, fluids, etc. However, it seemsthat too often the analysis stops with the technical aspects that areeasily understood in an engineering environment, where the realroot cause may exist in the human organization. In this tutorial, apractical guide to root cause failure analysis will be provided,followed by case studies to demonstrate both the technical andorganizational nature of a typical root cause failure analysis.

INTRODUCTION

Despite the best efforts to avoid them, failures are still acommon occurrence in every industry. Of course, there arethe more obvious and well-publicized failures in the automotive,petrochemical, aerospace, and mining industries, just to name afew, but there are also many less catastrophic failures occurring atany point in time. Failures not only represent imperfection in thetechnical attempt to design and operate complicated systems, but

they also represent organizational inability to successfully managethe competing interests of time, quality, and money. Therefore, inthe interest of continuous improvement, it is in one’s best interestto learn all one can from these failures, allowing us to avoidmaking the same mistake twice.The objective of this tutorial is to provide the reader with a

practical guide for performing root cause failure analysis anddetermining the appropriate corrective/preventive action necessaryto avoid the same failure in the future. The root cause failureanalysis (RCFA) process begins with the collection phase, followedby the analysis phase, and concludes with the solution phase. Eachof these phases is shown in Figure 1.

Figure 1. RCFA Flow Chart.

149

A PRACTICAL GUIDELINE FORA SUCCESSFUL ROOT CAUSE FAILUREANALYSIS

byDavid L. Ransom

Senior Research Engineer

Mechanical and Materials Engineering Division

Southwest Research Institute

San Antonio, Texas

Within collection, there are several key steps including teamforming, problem definition, and of course data collection. Theanalysis phase is simply represented by determining theimmediate, contributing, and root causes of the defined problem.The solution phase consists of determining corrective/preventiveaction, and then testing and implementation, which of course isthe final step in RCFA. In each phase of the process, there arecritical steps and simple guidelines to consider that will keep theinvestigation focused and practical. There are also some practicalmethods for organizing the investigation, depending on the sizeof the system under review. Finally, two turbomachinery relatedcase studies are presented and discussed throughout the lengthof the tutorial to assist in demonstrating the overall processand guidelines.

CASE STUDIES

As stated in the introduction, two turbomachinery related casestudies are threaded throughout the length of the tutorial, and areused to demonstrate steps and guidelines along the way.The first case study (Case 1) is from Alpha Company, who

manufactures after-market parts for industrial turbomachinery.Just like every other day, an order comes in for the manufactureof a product for which drawings already exist in the engineeringfiles. The engineering department pulls the drawings, selects thematerials, and issues the work order to the shop. The shop, in turn,manufactures the hardware per the print and sends all the piecesto final inspection. During the course of final inspection, itis found that the pieces of the assembly do not fit. This isparticularly troublesome for two reasons. First, this is a priorityjob and must ship immediately. Second, this is a product that hasbeen manufactured in the past, so any defects in the design ormanufacturing process should have been worked out in the pastthrough the engineering change request (ECR) process.The second case study (Case 2) is from the Bravo Power

Company, owner of several power generation gas turbinesoperating in combined cycle. After completing the spring outage(just in time for the summer heat wave), one of the gas turbinesbegins to exhibit a higher than normal temperature spread in theturbine exhaust. The unit is shutdown and inspected, and is foundto have cracked crossfire tubes on one of the combustors. Thedamaged hardware is replaced, and the unit is returned to service.However, the temperature spread is still unacceptably high. Asecond unscheduled outage reveals that yet another combustorhas cracked crossfire tubes. This time, the full set of combustorsis replaced, and the unit is returned to operation withoutfurther anomalies.

STEPS TO ROOT CAUSE FAILURE ANALYSIS

As described in the introduction, RCFA is generally divided intothree major phases: collection, analysis, and solution. Each of thesesteps is described in detail below. It is best to proceed through thephases as they are presented (i.e., one should not consider solutionsuntil the analysis is complete), but it is not a one way path. Thereare plenty of reasons to possibly back up and repeat or revisit anystep in the process before proceeding further.

Collection

Collection is used to describe all of the work necessary toprepare for the analysis phase. Naturally, the first step is to form ateam that will participate in the RCFA. Team members should haveownership of the problem, and will therefore probably includeengineers, technicians, operators, sales, management, etc. Theseteam members are considered the natural team, as they have a firsthand interest in the results of the RCFA.There are two main reasons why other team members might be

added over the course of the investigation (Figure 2). First, it maybe necessary to bring expertise into the team to help resolve key

questions or assist in the development of viable solutions. These“expert” team members do not need to be permanent members,and can be released once their contribution is complete. Theirrole is to support the investigation so that it is not halted fortechnical reasons.

Figure 2. Team Expansion for Either Technical or OrganizationalLimitations.

The second reason to add team members is to increase the circleof influence of the team. As the investigation matures, it maybecome apparent that the real root cause lies outside of thecurrent team’s influence. For example, an engineering investigationmay point to an issue in manufacturing. In such a case, it isimportant to add a team member that has the desired influence (i.e.,manufacturing lead/manager), so that the investigation is notprematurely halted due to organizational boundaries.In Case 1 above, the natural team formed includes the

engineering manager, two designers, one sales staff, and thequality manager. In this case, the natural team seems to have theownership, technical knowledge, and the influence necessary forthe resolution of this problem.In Case 2, the natural team is smaller, consisting only of the

plant manager and the operations manager. Since neither of themhas the technical knowledge necessary for the investigation, a thirdparty consultant is also hired for the investigation.It is also possible to have an investigation team lead by an

outside, independent investigator. This is most likely to occur whenthere is suspicion of overall organizational failure, and is typicallyimposed by a superior authority, such as senior management,industry regulator, etc. In such a case, the investigation may be leadby another division manager or a recognized industry expert. Inthis scenario, the people who would have otherwise been likelycandidates for the natural team become technical team members,and will likely only be involved in the actual datacollection phase.The next step in the collection phase is to define the problem.

Defining the problem is a team activity, usually requiring someamount of brainstorming to come up with just the right definition.The quality of the investigation depends heavily on the quality ofthe problem definition.A good problem definition is short, simple, and easy to

understand. In fact, if a problem statement is complicated, itmerely reflects a poor understanding of the real problem. It isimportant that everyone on the team understands and agrees withthe problem statement.The problem statement must also not be biased toward a specific

solution. The consequence is the potential to either completely


miss the real root cause, or at a minimum, miss some importantcontributing causes.In Case 1, the problem statement is determined to be, “Why did

the product fail to pass final inspection?” On the contrary, if theteam jumped immediately to what they intuitively determined asthe root cause, the problem statement might be, “Why didn’tengineering update the drawings after the first manufacture?” Aswill be shown later, this second problem statement prevents theteam from identifying an important contributing cause.The final portion of the collection phase is the actual data

collection. Table 1 is a listing of the most common sets of datathat should be collected for an industrial turbomachineryfailure analysis. Generally speaking, there are three commontypes of data: physical evidence, recorded evidence, andpersonal testimony.

Table 1. Common Data Types.

The most critical aspect of collecting the physical evidence is toresist the urge to clean. Although it may seem desirable to provideclean, easy to handle samples to the various technical experts forreview, the odds are that valuable data will be lost in the cleaningprocess. Figure 3 is an example of just such an occasion. The aircompressor impeller and the stationary passage are contaminatedwith chlorides and sulphur, leading to the stress corrosion cracking(SCC) failure of the impeller blades. Cleaning these parts beforethe completion of the investigation will add uncertainty to themetallurgical analysis, as well as eliminate the evidence of thecorrosive source (contaminated air). There are times when it isnecessary to further damage evidence just to remove it from thescene. In this case, care should be taken to not impact the actualdamaged portions of the evidence.

Figure 3. SCC Failure Due to Chloride and Sulfur in Air Stream.

In addition to the failed components, it may be important tothe investigation to provide good, undamaged components forstudy as well. For example, in blade failure analysis, the mosteffective method for evaluating blade modes is to rap test a goodblade. Undamaged parts can also be important for extractinggeometric information to be used in a computer simulation(finite element analysis [FEA], computational fluid dynamics[CFD], etc.).Depending on the type of failure, it may also be important to

capture physical evidence such as lube oil samples, water samples,air filter samples, deposit samples, etc. Generally, it is experiencethat will determine what other physical evidence needs to beretained. If there is doubt, it is certainly better to retain the samples.They can always be discarded once the investigation is over.Recorded evidence is the next significant type of data to be

collected. Pictures are clearly necessary for the investigation. Thetendency is to take too few pictures, because at the time, it seemsimpossible to forget what is being witnessed. However, experiencewill show that there cannot be too many pictures.There are two good concepts to keep in mind when taking

pictures. First, for each detail picture, include a series of picturesthat start from a very large view, and then gradually (perhaps threesteps) zooms into the desired level of detail (Figure 4). Thistechnique is vital to maintaining perspective and orientation.

Figure 4. Photo Sequence Captures Orientation.

Another important concept is to take pictures in orthogonalviews, as if they are intended to be used as manufacturingdrawings. Although isometric views are handy for seeing theoverall layout, they are very difficult to scale. Orthographic viewscan easily be used as pseudo drawings, especially if there are atleast three views recorded (front, top, and side). In Figure 5, the topphotograph shows an isometric view of this small, auxiliary powerunit (APU) gas turbine. Although this view is helpful to seewhere the fuel lines are located, it is very difficult to extract linedimensions from this view. On the other hand, the lower twophotos provide the proper view, allowing for dimensions to bescaled, if necessary.

A PRACTICAL GUIDELINE FORA SUCCESSFUL ROOT CAUSE FAILUREANALYSIS 151

Figure 5. Orthographic Views Are Easier to Scale.

The other forms of recorded data (operator logs, supervisorycontrol and data acquisition (SCADA) logs, etc.) can be critical tothe complete understanding of operating conditions at the time offailure. Figure 6 is an example plot from a pump SCADA system.These data are used to assist in a failure analysis of an overheatingbearing. Since log data typically are dated, they are ideal forgenerating a timeline of events. However, due to the costs associatedwith data storage and management, high resolution data are oftenretained only for a short period, replaced by lower resolution datafor long term storage. Therefore, it is critical to capture theseelectronic data as soon as possible.

Figure 6. Example SCADA Plot.

The other important category of data to collect is the personaltestimony. Theoretically, since everyone involved is discussingthe same event, all of the various stories should converge. If theinformation from the various personnel does not agree, it may be asign of multiple failures. Obviously, there is significant potentialfor finger pointing, or at least perceived finger pointing duringthis phase of data collection. To minimize this perception, itis important that the interviews be conducted by a rational,coolheaded person. Sending in an irritated and irrational person to

collect personal testimony will definitely have adverse effects onthe quality of the testimony. Second, it is important to stay focusedonly on data collection, building a consistent timeline, etc. Anypremature discussion of the cause of failure will likely adverselyimpact the interview process.It is up to the investigating team to resolve all conflicts in the

data, whether it is in the personal testimony, in the operator logs,etc. Unfortunately, due to the human influence, none of the datasources will be pristine. But, by comparing all of the data, fillingin the gaps, and resolving the conflicts, a clear and consistentpicture of the failure can be obtained.

Analysis

The analysis phase is solely focused on using the collected datato build the cause chain and determine the immediate, contributing,and root causes of the failure. The immediate cause is typically thefirst one in the cause chain, thus directly leading to the failure. Theroot cause is the last one in the cause chain, while the contributingcauses are the ones in between the immediate and the root.Although the process is referred to as root cause failure analysis, itis important to identify all of the causes.There are several common structures used in the analysis phase.

The “why” chart is a simple series of questions that guides the teamto the root cause. This is generally applied to small systems, orproblems that do not span over to more than a couple of systems.This method is generally useful for most rotating machineryfailures. The chart begins with the first problem statement,followed by the first answer to the problem statement. Thequestions are answered in small steps, which help to preventmissing any contributing causes.For Case 2, the “why” chart starts with the event question, “Why

did the gas turbine register an increased T5 spread?” The rest ofthe chart is provided in Figure 7.

Figure 7. Case 2 Why Chart.

At the conclusion of this RCFA, the immediate cause isdetermined to be uneven combustion, and the root cause is


determined to be an assumption of service provider qualitypractices, not simply a technical flaw. This is important to note, asit drives the types of solutions that are considered in the nextphase. Notice also that the contributing cause (poorly welded partsdelivered by the service provider) lies outside of the current team’sinfluence. At this point, it is recommended that the service providerbe included in the team.The “why” chart is simple to apply, and will work for most of the

turbomachinery related failures found in industry. For much largersystems, it may be more practical to use a fault tree. Fault trees arecommonly used in the aerospace and nuclear power industries,since these systems are typically very complicated and much moredifficult to investigate.Basically, the fault tree method requires that the team start with

the fault, and then work backward to identify all possible causes ofthis fault. For large, difficult to understand systems, this provides amap for dividing up the investigation. As the investigationproceeds, teams gradually rule out each potential cause, and markit off on the tree. By the end, there should remain a short list ofpotential root causes. Figure 8 is an example of a fault tree for Case2. In this case, each of the events is preceded by either an AND orOR logic statement. For example, “Uneven Combustion” can becaused by either “Uneven Fuel Delivery” OR “Uneven AirDelivery.” On the other hand, the “Poor Weld Quality” is the resultof both “Improper Welding Technique” AND “Inadequate QualitySystem.” Although this fault tree is incomplete, it demonstrates thelevel of detail required when using this approach. Each branchmust be expanded and evaluated by the team. By closing out eachbranch of the tree, the actual failure cause chain becomes apparent.

Figure 8. Case 2 Fault Tree.

Since this method is more complex and relies on a systemof symbols (similar to a process flow chart), there are manycommercial software packages available to assist in the process.Due to the size and complexity of the systems for which faulttrees are used, the investigation is usually managed by anexperienced investigator.Another popular structure for the analysis phase is the cause

and effect diagram (also known as a “fishbone” diagram). Wherefault trees are useful for complex systems, cause and effectdiagrams are useful for incorporating cross-functional influences.As seen in Figure 9, the head of the “fish” is the problem to beinvestigated, and each of the main branches (bones) represents aspecific functional area. To complete the fishbone diagram, theinvestigation team continues to list all of the possible connectionseach functional area might have with the failure. This formatallows the team to see the overall picture and begin to focus theinvestigation as each of the functional branches is evaluated. Inthis case, it is clear from the beginning that the failure iscontained within the engineering function, eliminating the need tofurther investigate the other branches.

Figure 9. Case 1 Cause and Effect (Fishbone) Diagram.

There are some other important key points to remember duringthe analysis phase. It is helpful to keep these handy as theinvestigation proceeds, so that each team member is reminded ofthese guidelines.

• Follow the data—The most difficult aspect of the analysis phaseis avoiding preconceived notions regarding the root cause. It is upto the team members to protect each other from this trap. Theinvestigation team must stick to the data and exclude “gut feel”from the investigation.

• Consider both technical and organizational causes—Findingthe technical answer is often difficult, but the investigation shouldnot stop there. Organizational influences can be just as significantand must also be included in the investigation.

• Concentrate on analysis—Save the problem solving for the nextphase. The key at this point is to identify the immediate, contributing,and root causes.

• Really operator or maintenance error?—It is rarely actuallyan operator or maintenance craftsmen error. We all work inorganizations with norms, procedures, and external pressures.What appears to be operator error is most likely a broken process,missing check, or unclear expectations.

The analysis phase is complete once the immediate, contributing,and root causes are identified. Keep in mind that the root cause isdependent on the reach of the team. If the last contributing causeexists at a boundary that cannot be crossed (by either addingtechnical or organizational influence), then it is effectively the root


cause. In other words, this is where the solution phase shouldfocus. It is only of academic value to identify a root cause overwhich the team has no influence. For example, consider againthe cause chain for Case 2 (Figure 7). The contributing cause isidentified as “GT service provider delivered poorly weldedcombustors.” Clearly, this cause lies outside of the currentteam’s influence. If the gas turbine (GT) owner hopes to eliminatethis cause, they must convince the service provider to jointhe investigation.

Solution

Certainly, failure prevention starts with good design, manufacture,and operation practices, providing protection against the alreadyconsidered failure modes. For complicated systems, cause andeffect diagrams and fault trees are used to study potential failuremodes, allowing them to be incorporated into the design phase.However, it is just as important to learn from experience,preventing the recurrence of the same mistake, and this is wherethe solution phase of root cause failure analysis fits in.The fundamental objective of the solution phase is to break the

cause chain. Of course, this means that the quality of the solutiondepends heavily on the quality of the cause chain developed in theanalysis phase. To emphasize this point, consider again theproblem statement for Case 1. As shown in Figure 10, the correctinitial problem statement is “Why did the product fail to pass finalinspection?” Suppose the team instead started with “Why weren’tthe drawings corrected after the first manufacture?” This statementis located in the lower half of the why chart, below a critical split.Such a poorly posed problem statement would prevent theinvestigating team from identifying the real root cause, which is theassumption that all previously manufactured products haveupdated drawings.

Figure 10. Case 1 Why Chart.

Another important feature of the cause chain is that since mostfailures are the result of both root and contributing causes, thereare usually multiple areas that can be addressed. This is importantto recognize in the solution phase, as it helps to open up thenumber of possible solutions to the original problem. It is alsopossible that preventing some of the contributing causes can alsolead to improved reliability in other areas not presently considered.For example, in Case 1, one of the possible contributing causes is“Shop fixed the problem without engineering involvement.”Clearly, if this occurs, there is no feedback through the ECRprocess, and the design cannot possibly be corrected. Although thisdid not occur in this particular case, it is identified as a potentialfailure mode, and preventive action is taken to minimize thepossibility of this failure.

Therefore, using the well-developed cause chain as a startingpoint, the solution phase begins by identifying all the possible waysto break the chain. These solutions are referred to generically ascorrective/preventive actions. Each corrective/preventive actionmust be evaluated for effectiveness (i.e., does it reduce thelikelihood of the event recurring to an acceptable level) and realism(i.e., is it reasonable to implement with respect to cost, time,organizational influence, technical requirements, etc.). Table 2 is alist of corrective/preventive actions for both of the case studies,along with an assessment of effectiveness and realism for eachpotential action.

Table 2. Corrective/Preventive Actions.

For Case 1, perhaps the most obvious, or at least the initialresponse, is to thoroughly check each drawing before it is issuedto the shop. But, when this possible solution is evaluated foreffectiveness and realism, it becomes clear that it is a poor option.It is later determined that the best corrective/preventive actionplan is to eliminate the ECR back log (and discontinue practice ofadding to the back log), and begin to review previous job files foreach repeat order, thus significantly improving the probability ofeliminating repeat mistakes.For Case 2, although the root cause is an assumption of service

provider quality, it is also clear from the cause chain that there aretwo reasonable approaches to the corrective/preventive actionplan. First, it might make sense to begin the practice of inspectingcombustors for weld quality before they are installed. This providesthe GT owner with the direct control of the combustor weld quality.However, it is also in the GT owner’s interest to get involved in theservice providers quality system, making it possible to eliminatethe extra level of quality inspection. Obviously, this approachrequires teaming with the service provider, and therefore may notbe relied upon for the immediate correction that is necessarybefore the next outage.

SUMMARY

Failures in human-made systems reflect both technical andorganizational flaws. Although it is unreasonable to expect perfectperformance with perfect reliability from these systems, it is just asunreasonable to allow the same failure to occur multiple times.Therefore, the objective of this tutorial is to provide the reader witha practical guide for performing root cause failure analysis anddetermining the appropriate corrective/preventive action necessaryto avoid the same failure in the future.


RCFA starts with the collection phase, consisting of teamforming, problem definition, and of course data collection. Next isthe analysis phase, determining the immediate, contributing, androot causes of the defined problem. Finally, the solution phaseconsists of determining the appropriate corrective/preventiveaction plan that will effectively break the cause chain. In eachphase of the process, there are critical steps and simple guidelinesto consider that will keep the investigation focused and practical.These, of course, are the key characteristics of a successful rootcause failure analysis.

BIBLIOGRAPHY

AlliedSignal Aerospace FM&T, 1997, “Root Cause Analysis andCorrective/Preventive Action Workshop.”

Vesely, W. E., Goldberg, F. F., Roberts, N. H., and Haasl, D. F.,1981, “Fault Tree Handbook,” NUREG-0492, U.S. NuclearRegulatory Commission, Washington, D.C.


Charles R. (Charlie) Rutan is SeniorEngineering Advisor, Specialty Engineer-ing, with Lyondell Chemical Company, inAlvin, Texas. His expertise is in the field ofrotating equipment, hot tapping/plugging,and special problem resolution. He hasthree patents and has consulted on turbo-machinery, hot tapping, and pluggingproblems all over the world in chemical,petrochemical, power generation, andpolymer facilities.

Mr. Rutan received his B.S. degree (Mechanical Engineering,1973) from Texas Tech University. He is a member of the AdvisoryCommittee of the Turbomachinery Symposium, and has publishedand/or presented many articles.

ABSTRACT

This paper is an extended case history of a turbine catastrophicfailure due to stress corrosion and the reconstruction of the turbine,compressor, gearbox, and the drive motor after a fire. Wheelstresses and the design upgrades to minimize the potential forfailure while reducing the subsequent damage if a failure occurredof the complete system will be discussed.

INTRODUCTION

A Lyondell manufacturing facility has a recycle gas compressorthat continuously recycles ethylene and ethylene oxide to thereactors. On December 28, 1996, the drive turbine of the compres-sor machine train catastrophically failed.

General Description

The equipment consists of a compressor driven at one end by asteam turbine and at the other end by a motor through a speed-increasing gear. The driving motor, gear, compressor, and turbineare mounted on a common bedplate. The compressor train isequipped with nonlubricated diaphragm couplings, water sealsystem, control and lube oil systems, and the accessories necessaryfor the safe and efficient operation of the unit. Design data can befound in Table 1.

The operation of the compressor train is somewhat unique for achemical plant. The turbine is started and once it reaches 6000rpm, the motor is started and the compressor train comes up to8676 rpm. Figures 1 and 2 show the layout of the compressorplatform.

During the summer, at peak energy consumption, the turbine isused to augment power when electrical power shedding occurs forthe local electrical grid. This process lets steam down to a levelrequired by the process and uses the steam developed in theprocess reactors.

A major overhaul and an uprate of the turbomachinery train hadbeen performed in the spring of 1996. The new motor, couplings,and gearbox were installed at this time. However the turbine, com-pressor, and foundation did not require modifications as the uprate

Table 1. Design Data.

1

TURBINE FAILURE AND RECONSTRUCTION

byCharles. R. Rutan

Senior Engineering Advisor, Specialty Engineering

Lyondell Chemical Company

Alvin, Texas

Turbine

Number of stages 3

Rating, bhp (kW) 5050 (3768)

Speed, rpm Original 7680, New 8676

Maximum continuous speed, rpm Original 7757, New 8676

Tripping speed, rpm 9000

Normal inlet pressure, psig 760

Normal inlet steam temperature, �F 626

Normal exhaust pressure, psig 250

Wheel diameters 22.616

First rotor response speed (critical), rpm 4700

Rotor material ASTM A470 Class 8

Rim welds for the first and second

wheels

F8

Rim welds for the third wheel F7

Bucket material Original #1 AISI 422 SS, New TG-410AB001

Compressor

Number of stages 1

Speed, Rpm 8676

Gas handled Recycle mixture

Molecular weight 25.79

K 1.305

Z 0.964

Inlet pressure, psia 235

Inlet temperature, �F 85

Discharge pressure, psia 317

Discharge temperature, �F 135.5

Rating, bhp 13,390

Gear

Type Double helical, speed increaser

Rating, bhp 17,285

Input speed, rpm 1793

Output speed, rpm 8676

Gear ratio 1:4.8388

Service factor 1.4

Motor

Type Induction, premium efficiency

Rating, bhp 15,030

Speed, rpm 1793

Phase 3

Volts 12,470

Efficiency, % 97.3

Hertz, cycles per second 60

Service factor 1.25

Figure 1. Compressor Platform Layout Plan View.

Figure 2. Compressor Platform Layout.

was a speed increased from 7757 rpm to 8676 rpm and an increasein horsepower occurred to meet the new load requirements of thecompressor. The compressor manufacturer completed a thoroughdynamic and torsional analysis and found no issues. The turbineand compressor rotors were removed, inspected, and refurbishedlocally. Inspection included:

• A thorough cleaning.

• Visual inspection.

• Dimensional verification.

• Magnetic particle• Shafts.• Integral wheels.• Buckets.• Shroud bands.• Tenons.• Impellers.

• Complete electrical runouts.

• Complete mechanical runouts.

• Incoming low speed balance.

Minor wear of the turbine and compressor was reported, whichrequired some handwork. The rotors were low-speed balanced,then at-speed balanced, and returned to the facility where theywere installed in their respective cases. The nozzles, diaphragms,packing boxes, etc., were similarly inspected with the same result.New packing and bearings were installed. The plant had most ofthe normal spare parts including a new spare compressor rotor.Plant management did not see the value in a spare turbine rotor,spare gears, or a spare motor. The vibration monitoring system andturbine controls (completely manual) were not upgraded. The timelimitation for the installation was the time it took to dump thereactor catalyst and reload new catalyst.

Due to the process safety concerns the facility had, as per thestandard operating procedure (SOP), all of the automaticshutdowns of the entire compressor train were bypassed:

• Vibration

• Thrust

• Low lube oil pressure

• Low lube oil level

• Overspeed

• Low sealing water flow

• Low sealing water pressure

• Low sealing water flow

With the exception of the motor electrical overloads located in themotor control center, all the instrumentation was locally monitoredand controlled from a foul weather building located on the com-pressor platform.

When the turbine failed (Figures 3, 4, 5, and 6), the inboardbearing cover split into two pieces (Figure 7) and the outboardbearing cover rolled to one side of the turbine (Figure 8). Oil beganspewing out all over the platform and the atomized oil ignited. Theignition source was the molten metal caused by the heat generatedfrom the metal-to-metal contact rub in the bearing areas. Theplatform was engulfed in flames burning all the instrumentation,electrical wiring, and the foul weather building. The conduitmelted and the electrical insulation melted away exposing the wirethat came in contact with other wiring and the metal structure(Figure 9). At this point, the production operator arrived on theseen and was about to turn on the firewater monitor, located atgrade level on the platform. However, when he saw all the electri-cal arcing he decided that it would be unsafe and elected to let thefire burn itself out, which occurred when all the lube oil had beenconsumed. This was not a good idea. The compressor mechanicalwater end seals did not fail and the trip throttle valves located at theinlet and inner stage admission did close, but the motor continuedto turn at-speed until the motor current overloads tripped.

Figure 3. Inboard Compressor Turbine.

Figure 4. Valve Rack Support.

PROCEEDINGS OF THE THIRTY-FOURTH TURBOMACHINERY SYMPOSIUM • 20052

TurbineGear

Motor

Compressor

Figure 5. Turbine Outboard Wobble Foot.

Figure 6. Exhaust End Wobble Foot.

Figure 7. Turbine Inboard Bearing Cap Split.

Figure 8. Turbine Outboard Bearing Cap.

Figure 9. Typical Electrical Instrumentation Junction Box.

The turbine was supplied with steam from two sources. The inletor head end steam was purchased from another chemical companythat has a plant located near this plant. Low-pressure steam wasgenerated in the unit reactors, and then introduced between thesecond and third stages. At this location inside the turbine the sta-tionary nozzles are located in the upper half and a diaphragm in thelower half of the case. As the incident investigation progressed, anotification of a caustic excursion in the boiler feedwater makeuphad occurred on December 24, 1996. The source was the companythat supplied the high-pressure inlet steam to the turbine. No addi-tional information was supplied at that time. Operations personneltook a condensate sample and found that it had a 13 pH.Supervision did not realize that a pH of 13 was a serious issue andcould cause stress corrosion cracking of steels. The steam supplierhad sent, by mail, a formal notification of the caustic excursion tothe plant manager, but he did not receive it until December 30,1996. This notification was too late by six days.

On the morning of January 3, 1997, the compressor platformwas released to maintenance to begin the job of damage assess-ment and repair. Structural consultants, piping experts,instrumentation, electrical, piping, and rotating equipmentengineers were brought in to see what could be salvaged and whatitems had to be replaced. The motor, gearbox, and compressor case(with rotor) were returned to the original manufacturers for inspec-tion. The compressor rotor, turbine end, was severely damaged(Figure 10).

Bearing journal, labyrinth oil seals, and coupling areas wouldrequire submerge arc welding and machining to bring these areasback to the original specified tolerances. Additionally, the rotor had

TURBINE FAILURE AND RECONSTRUCTION 3

Figure 10. Compressor End.

a 5 mil kink, turbine side, starting at the compressor end seal areaextending to the shaft end. A decision was made that no effortwould be made to repair the compressor shaft at this time becausea new spare compressor rotor was available; so all the attentionwas focused on the turbine repairs.

The motor sustained minimal damage to the bearing journal andoil seal areas. After the repair of the journal areas and the installa-tion of a new bearing, the motor was tested and returned to theplant with a fresh coat of paint.

Minor damage was found upon inspection of the gear. Thebearing journal areas were hand dressed. The gear teeth requiredregrinding and new bearings were installed. The gearbox was thentested and returned to the plant with a fresh coat of paint.

While the above was transpiring, the decision was made thatrepairing the turbine rotor was more expedient than waiting for anew forging. A detailed list of all the components that could berepaired, replaced with stock inventory, and fabricated wasdeveloped. An outside consultant was contacted to determine whatcaused the turbine failure. The first two wheels suffered severedamage. It was learned that one complete segment of the thirdwheel was missing (Figures 11, 12 and 13).

Figure 11. Turbine First and Second Stages.

Figure 12. Turbine Third Stage.

Figure 13. Third Wheel.

When the remainder of the turbine rotor was magnetic particleinspected the Christmas trees (bucket tenons) of the third wheellooked like they had grown hair (Figure 14). The cracks weresectioned and inspected. Caustic was present and the wheel hadfailed due to stress corrosion cracking. A section of the wheel,identical to the section that had failed (an eight-bucket packet), wasremoved and weighed. A quick calculation with this weight, radiusfrom the center of the rotor and speed, was performed. It was deter-mined that when the section came off the rotor, the turbineexperienced an unbalanced force of 160,000 lb. If any of theshutdowns had been activated, and the compressor train had auto-matically been shut down, then the damage would have been bad,but not catastrophic. Fortunately, the compressor mechanical waterend seals did not leak and remained intact. If these seals had failedand the ethylene oxide gas had been released to the atmosphere,there was a very high probability that an additional event wouldhave occurred.

STEAM CONTAMINATION

Stress corrosion cracking is one failure mechanism that has beenwell documented. Please refer to the attached list of references atthe end of the paper for more information. Turbine manufacturersare all in agreement that the steam purity must be maintained at thelowest practical level of contaminants. It should not exceed 3.0 ppbNa, cation conductivity of 0.2 μmho/cm. Total suspended solidshall not exceed 0.1 ppm (100 ppb) and pH 8.0 to 10.0 (inclusive)during normal operation. The manufacturers’ recommendations areintended to limit sodium compounds, such as caustic (NaOH) andsodium chloride (NaCl). One manufacturer has published thefollowing limits during periods of abnormal operation: for short


Figure 14. Third Stage Sectioned.

periods, not to exceed 100 hours per incident and accumulating500 hours or less in a 12 month operating time, 6.0 ppb Na and 0.5μmho/cm should not be exceeded. During emergency conditions,for periods of 24 hours or less with accumulations not exceeding100 hours in a 12 month operating time, 10.0 ppb and cation con-ductivity of 1.0 μmho/cm should not be exceeded. The plant didnot, at the time of the incident, have the capability of measuringanything other than the pH of the steam condensate. Just as areminder, the plant had measured a 13 pH of the steam condensate.This situation was rectified during the period of time it took torebuild the compressor train. Typical specifications for steamquality are found in Tables 2, 3, 4, and 5.

Table 2. ABMA Guidelines—Boiler Feedwater, Boiler Water, andSteam Specifications Applicable for Steam Drums OperatingBetween 0 to 300 PSIG.

Water/steam related parameters are found in Tables 6, 7, and 8.The overall quality of the combined “fresh” and recoverable con-densate (vacuum and/or suspect) makeup ultimately impacts boilerefficiency and operating continuity and steam purity. Poor steampurity can lead to carryover related problems such as steam turbinefouling. Lastly, continuous and intermittent blow-down control canhelp ensure optimum boiler water cycles of concentration, andreduce unwanted impurities.

The following information covers online sodium analysis and itsrelationship to total solids/total-dissolved solids.

“A sodium analyzer is used to determine steam purityby measuring the sodium ion present in steam with theuse of specific ion electrodes. Specific sodium ion elec-trodes (Note: Samples streams must be cooled with asample cooler to prevent damage to various components

Table 3. ASME Guidelines—Boiler Feedwater, Boiler Water, andSteam Specifications Applicable for Steam Drums OperatingBetween 0 to 300 PSIG.

Table 4. ABMA Guidelines—Boiler Feedwater, Boiler Water, andSteam Specifications Applicable for Steam Drums OperatingBetween 301 to 450 PSIG.

of the sodium analyzer) have been developed which areextremely accurate (±2 percent) and reliable. As a resultof this and because sodium is the most common ionfound in boiler water, it is an excellent indicator of boilerwater carryover problems. When using the specificsodium ion technique, the approximate total solidspresent in steam is calculated by multiplying the sodiumion content by a factor of three. Thus, a sodium ionreading of 0.33 ppm would represent approximately 1.0ppm of total solids.”

“By running steam purity evaluations, determinationcan be made of the carryover TDS (total dissolvedsolids) in steam as well as evaluate methods for reducingTDS in the steam by making mechanical or chemicalchanges in boiler operation. Other applications forsodium analyzers include monitoring demineralizereffluent and condensate.” (Drew, 1994)


Total dissolved solids1 700-3500 ppm

Total alkalinity2 140-700 ppm as CaCO3

Suspended solids 15 ppm (max)

Total dissolved solids2,3 steam 0.2-1.0 ppm (max expected value)

Notes:1 Actual values within range reflect the total dissolved solids in feedwater.2 Actual values within the range are directly proportional to the actualvalue of total dissolved solids of boiler water.3 These values are exclusive of silica.

Boiler Feedwater

Iron 0.100ppm or 100 ppb as Fe

Copper 0.050 ppm or 50 ppb as Cu

Total hardness 0.300 ppm or 300 ppb as CaCO3

pH Report

Boiler Water

Silica 150 ppm as SiO2

Total alkalinity1 350 ppm2 as CaCO3

Specific conductance 3500 mmho

pH Report

Notes:1 Minimum level of hydroxide alkalinity in boiler below 1000psi must be individually specified with regard to silicasolubility and other components of internal treatment.2 Maximum total alkalinity consistent with acceptable steampurity. If necessary, the limitation on total alkalinity shouldoverride conductance as the control parameter.

The above parameters and limits should be reviewed with thesite water treatment provider.

Total dissolved solids1 600-3000 ppm

Total alkalinity2 120-600 ppm as CaCO3

Suspended solids 10 ppm

Total dissolved solids (steam)2,3 0.2-1.0 ppm

Notes:1 Actual values within range reflect the total dissolved solids infeedwater.2 Actual values within the range are directly proportional to theactual value of total dissolved solids of boiler water.3 These values are exclusive of silica.

Table 5. ASME Guidelines—Boiler Feedwater, Boiler Water, andSteam Specifications Applicable for Steam Drums OperatingBetween 301 to 450 PSIG.

Table 6. Water/Steam Related Parameters Specifically for 451 to600 PSIG Steam Generators (ASME Guidelines Unless OtherwiseStated).

Comment: It is assumed that the author considers total solidsand total dissolved solids as being synonymous. Therefore, if theTDS in steam concentration (ppm) is determined analytically, ortaken from ASME/ABMA steam purity guidelines, the theoreticalsodium concentration can be determined by multiplying the TDSconcentration by 0.33.

Table 7. Water/Steam Related Parameters Specifically for 901 to1000 PSIG Steam Generators (ASME Guidelines UnlessOtherwise Stated).

Table 8. Water/Steam Related Parameters Specifically for 1001 to1500 PSIG Steam Generators (ASME Guidelines UnlessOtherwise Stated).


Boiler Feedwater

Iron 0.050 ppm or 50 ppb as Fe



Boiler Water

Silica 90 ppm as SiO2

Total alkalinity1 300 ppm2 as CaCO3

Specific conductance 3000 mmho

Notes:1 Minimum level of hydroxide alkalinity in boiler below 1000psi must be individually specified with regard to silicasolubility and other components of internal treatment.2 Maximum total alkalinity consistent with acceptable steampurity. If necessary, the limitation on total alkalinity shouldoverride conductance as the control parameter.

The above parameters and limits should be reviewed with thesite water treatment provider.

Boiler Feedwater Limits

Organic Nondetectable (TC/TOC min. detection ~ 0.5 ppm)




Dissolved oxygen* 10 ppb as O2 (w/o oxygen scavenger)

Boiler Water

Silica** 8 ppm as SiO2

Total alkalinity 100 ppm as CaCO3

Specific conductance 1000 mmho/cm

Total solids 1250 ppm (ABMA max.)

Suspended solids 40 ppm (ABMA max.)

Steam Purity Limit

Total dissolved solids 0.1-0.5 ppm (max. expected value)

Silica 0.02-0.03 ppm, or 20-30 ppb as SiO2 (ABMAlimit)

Notes:* Well-designed and operated deaerators can reduce oxygen to as low as 7 ppb.** Silica limit based on limiting silica in steam.

ASME guidelines unless otherwise stated.The above parameters and limits should be reviewed with the site water treatmentprovider.





Total hardness None detectable


Boiler Water

Silica** 2 ppm as SiO2

Total alkalinity Not specified, dictated by boiler water treatment program




Steam Purity Limit

Total dissolved solids 0.1 ppm or 100 ppb (max. expected value)

Silica 0.02-0.03 ppm or 20-30 ppb as SiO2 (ABMA limit)

Notes:

* Well-designed and operated deaerators can reduce oxygen to as low as 7 ppb.

** Silica limit based on limiting silica in steam.

ASME guidelines unless otherwise stated.

The above parameters and limits should be reviewed with the site water treatment provider.







Boiler Water

Silica** 35 ppm as SiO2 (ABMA max.)

Total alkalinity 250 ppm as CaCO3




Steam Purity Limit

Total dissolved solids 0.2-1.0 ppm (max. expected value)

Silica 0.02-0.03 ppm, or 20-30 ppb as SiO2 (ABMA limit)

Notes:

* Well-designed and operated deaerators can reduce oxygen to as low as 7 ppb.

** Silica limit based on limiting silica in steam.

ASME guidelines unless otherwise stated.

The above parameters and limits should be reviewed with the site water treatment

provider.

Cleaning

Cleaning rotors after being exposed to steam contaminantsinvolves the removal of water-soluble and water-insolubledeposits. The original equipment manufacturer (OEM) should becontacted for recommendations concerning water washing and gritblast specifications and procedures. The author prefers to grit blastprior to water washing. For the general cleaning of steam turbinerotors, the recommended material is aluminum oxide with aparticle size of 220 mesh maximum. All the sensitive areas must beprotected by masking, such as the journal bearing areas, thrustdisk, threads, rotor shaft ends, vibration probe, etc. It is vital tohave an experienced individual performing this task. If the blastgun is held in one place too long, metal will be removed in additionto the contaminants. The only reason for water washing is toremove water-soluble deposits in the dovetail regions of the bucketand wheel assemblies and the tenon areas of the bucket and shroudband areas. In general water washing consists of three steps:

1. A general rinse and high-pressure water spray to remove themajority of the water-soluble deposits on the surface.

2. Completely submerging the turbine rotor assembly in a waterbath. The bath must consist of demineralized water with a con-ductivity of less than 1 micro siemens (μS) per centimeter andhave a pH range of 5.0 to 7.

3. A high-pressure water wash to remove seepage productsfollowing the soaking is the final step. The seepage must betested and the pH must be below 7 to 10. If the seepage is abovea pH of 10, then the rotor must be resubmerged in the cleaningbath.

Step 3 must be repeated as many times as required to lower or raisethe seepage pH into the recommended range.

Turbine rotors can be water washed at slow roll speeds. At-speedwater washing is not recommended by any of the turbine manu-facturers. This will remove water-soluble deposits such as salts(NaCl) and turbine blading if not executed very carefully. Thedeposits in the turbine that are not water-soluble, such as silica, canonly be removed by abrasive blasting of the rotor with the criticalareas, e.g., bearing journals and vibration probe, masked.

BACK TO THE STORY

The turbine case was sent to welding and after a thoroughcleaning the rotor was stress relieved then put into a lathe. Therotor was turned (machined) true. The wheel areas were actuallyundercut in the rotor shaft. The next step in the process was tobegin submerged arc welding. Weld metal was deposited about 3inches on the radius, then the rotor was turned to a roughdimension and the rotor was magnetic particle inspected. If therotor passed the magnetic particle inspection the weld material wasultrasonically tested to find any inclusions in the welds. When aninclusion was found, greater than 1/32 inch, the inclusion wasground out and repaired via welding (Figures 15 through 19). Thisprocedure was repeated many times until the rotor was sent to finalmachining and inspection (Figures 20 through 22).

During this period discussions were conducted to determine ifthere was any way to reduce the stresses on the third wheel. Anagreement was reached that the steel bucket and shroud bands wouldbe replaced with titanium Z lock buckets that would reduce theforces on the third wheel. A full engineering study was completedand no issues were found with the change of the bucket material.

As the rotor work was proceeding, the turbine case was beingweld repaired. As can be seen this was not an easy matter (Figures23 through 26). First, the split lines of the upper and lower halveswere weld repaired and rough machined. Then, the register fit areaswere weld repaired and rough machined. The turbine case halveswere then sent to stress relieving. Once this operation wascompleted, the case was returned to the machine shop and finalmachining was completed and the case was dimensionally checked.

Figure 15. Initial Weld Pass of Turbine Rotor.

Figure 16. Submerged Arc Welding.

Figure 17. Machining after Initial Weld Pass.

The original cast bearing bracket was fabricated while otherwork was performed. The valve lift beam had damaged supportsand had to be fabricated along with other “miscellaneous” compo-nents. The rotor was at-speed balanced to 0.5 mm/sec/bearing (theauthor’s specification), then installed in the repaired case andreturned to the plant (Figure 27). All the above work detailed tooksix weeks.

Since all the controls and vibration/temperature monitoringequipment were destroyed in the fire, this opportunity was taken toupgrade these systems. New state-of-the-art turbine controls andmonitoring systems were installed that allowed the startup of thetrain to go smoothly. The startup procedure was:


Figure 18. Turbine Rotor Stress Relieving.

Figure 19. Turbine Rotor Welding Completed.

Figure 20. Turbine Rotor Final Machining.

1. Line up and start the water to the compressor water seals.

2. Line up and start the lube/control oil system.

3. Test the shutdown turbine overspeed shutdown devices.

4. Perform three shutdowns:a. Low oil tripb. Thrust tripc. Overspeed trip

5. Couple the turbine to the compressor train.

6. Slow roll the turbine at 500 rpm for 1 hour to allow the turbinecase, turbine rotor, and oil systems to warm up close to theiroperating temperature.

Figure 21. Cutting Bucket Dove Tails in Turbine Wheels.

Figure 22. Loading Buckets into Turbine Rotor Wheels.

Figure 23. Upper Half of Case.

Figure 24. Case Split Line.


Figure 25. Bottom Half of Case.

Figure 26. Upper Half of Case after Welding Stress Relieving andMachining.

Figure 27. Finished Turbine Rotor in the At-Speed Balance Bunker.

7. Step the turbine speed up at 500 rpm per step until the speed setpoint, which is 500 rpm below the critical speed.

8. Press fast ramp on the electronic governor, the turbine ramps upat 250 rpm/second until 500 rpm above the first critical speed.

9. Set at this speed for 30 minutes, allowing all the turbinecomponent temperatures to become stable.

10. Once the minimum operating speed set point is reached,continue to jog up at 500 rpm steps until the high-speed stop(maximum continuous operating speed) is reached.

11. Bypass the high-speed stop and run the turbine up to theoverspeed trip set point.

Once the turbine was soloed it was shut down and coupled to thecompressor/gearbox/motor and restarted. The startup of the

compressor train steps is identical until step 10 (listed above),when 6000 rpm is reached, the motor is started, and the equipmentis brought up to the normal operating speed of 8676 rpm. This steptakes about four seconds, then the machine is turned over to oper-ations. The vibration levels were less than 1 mil in any plane.Bearing temperatures, thrust positions, compressor process tem-peratures, etc., were at their normal operating conditions and lifewas good!

Approximately an hour after startup, the compressor inboardand outboard bearing temperatures begin to rise, and the vibrationin both planes started to rise about 0.2 of a mil per hour. All theother operating parameters remained unchanged. The maximumradial shaft vibration level, in any plane, was considered to be 4.5mils. The orbit of the compressor shaft in the bearings wasperfectly round and at synchronous speed. A spectrum analysis ofthe vibration also indicated that the vibration was predominately atsynchronous speed. It appeared that an unbalance existed and wasslowly moving away from the compressor shaft centerline. Thecompressor was shut down and allowed to come to ambient tem-perature and then restarted. At speeds below the first critical speed,the vibration levels of the compressor rotor were slightly abovethose seen initially on the first startup. Once the critical speed ofthe compressor was gone through, the vibration levels rapidlyincreased to 4.5 mils and continued to climb. Analysis of thevibration indicated that at this point a rub was being picked up andthus the vibration climbed at a faster rate. Two additional attemptsto start up were made before the conclusion to split the compressorcase and determine the cause of the vibration.

The compressor case was split and some minor rubbing wasapparent, but it was not significant enough to cause the vibrationlevels to be seen until a dial indicator was put on the compressorlower half and the runout was checked at the center of the rotor. A45 mil bow was discovered. The rotor was removed from the caseand returned to the manufacturer’s machine shop. Dimensionalchecks were repeated. Runouts were repeated. Except for theinstance when the runout at the center of the rotor was performed,5 mils of runout were measured, which was acceptable for thisrotor. But this was not close to the 45 mils seen in the field. Sinceall readings were back to within tolerances, the rotor was returnedto the plant and reinstalled in the compressor. The thought was thatsomething must have shifted on the compressor rotor during thetransport by truck to the machine shop.

After the reinstallation of the rotor, the compressor was closedup and recoupled and the machine train started up. At this point,the repair was started on the compressor rotor that was originallyinstalled in the compressor during the initial wreck. The compres-sor train was restarted, as per procedure, and the exact samesequence of events occurred as detailed above.

The compressor case was shut down and split, once again, andthe condition of the rotor was as it had been left. Management wasnot happy and life was not very good. The rotor was again checked.There was a 45 mil bow in the center of the rotor, so the rotor wasreturned for a thorough inspection. The same results were seen inthe shop as before, rotor runout at the midspan was 4 mils. Therotor was unstacked and reassembled and everything was to spec-ifications. A decision was made to return the compressor rotor tothe facility. Once installed in the lower half of the compressor case,a runout check at midspan would be performed, and it was. Therunout at midspan was exactly as measured in the shop, 4 mils.Armed with this information a decision was made to complete theinstallation and startup.

As expected by some, the same sequence of events occurred.Life at this time was terrible! When the rotor was removed fromthe case this time, it was transported with the thrust disk and theend seals were left on the rotor. The rotor was set into v-blocks,seals supported, and runouts were retaken. The 45 mil runout atmidspan was found this time. As the rotor was disassembled, a dialindicator was placed at the midspan so any changes could be seen.


In the shop, with at least a dozen spectators, the thrust disk nut wasloosened and the midspan went from 45 mils to four mills. TheOEM representative and the author were watching the dialindicator when it changed. We would have called each other a liarif we had not both seen the change. Fortunately, the compressorrotor from the initial wreck was repaired and the rotor was readyfor installation. It was installed into the compressor and the com-pressor train was restarted, as per the operating procedure. Theplan was to get the unit up then go back to the new rotor andperform a very detailed inspection and determine what was causingthe 41 mil change in the rotor assembly. The machine train wasrestarted and vibration levels were less than 1 mil peak-to-peak atthe operating speed of 8676 rpm. Life was good again.

The problem compressor rotor was stripped (impellers removed)and every type of test to the 17-4ph shaft was performed. Nothingcould be found that would explain why the rotor would bow. Thefinal test was to hang the shaft vertically and stress relieve it. Oncethe stress relieving was completed it was allowed to cool, then itwas placed in a low speed balance stand and shaft runouts weretaken. The problem was immediately determined, a 45 mil bowwas found in the shaft at its midspan. The shaft had not been stressrelieved properly. The OEM bought us a new shaft and reassem-bled the rotor. Several years later during a major compressoroverhaul the new rotor was installed and it ran perfectly.

One would think that everyone lived happily ever after but theywould be wrong. The compressor train ran about three months,then the author received a call that the vibration levels at thebearing journal had pegged the vibration monitor at 5 mils and thetemperature of the turbine bearings was starting to climb.Operations was advised to shut down immediately and life hadturned again. The unit was shut down. The turbine bearings wereinspected and nothing was found. Digital data recordings of thevibration were reviewed and it appeared that a mass unbalancesuddenly occurred. The turbine case was split and four titaniumbuckets had broken off at the blade root and were lying in thebottom of the case. Now a way had to be found to start up the com-pressor train without the turbine. A modal and tensional analysiswas completed. The only way the compressor train could bebrought up was to leave the turbine side coupling that was mountedon the compressor rotor on the end of the rotor; that overhungmoment was needed to have a stable machine train. Startup proce-dures were rewritten to prevent the motor from tripping due to theextended rampup time that was expected for motor overloads. Thetime out was set from 8 seconds to 15 seconds and it actually took13 seconds to reach 8676 rpm with the compressor completelyunloaded. About three weeks later the motor tripped due tooverloads and this was a surprise to operation personnel becausethey were not aware this shutdown set point was not increased togive them any warning. This did prove that they could trip withoutwarning and the plant would not exceed a safety critical variable.All the safety systems functioned as designed; ethylene oxide gaswas directed to the flare.

Now the reason the titanium buckets failed could be investi-gated. The buckets were removed from the third wheel and therotor was inspected thoroughly and the wheel was in as-newcondition. It was finally determined that the steam at the inlet ofthe third stage was causing very high alternate loading of thebuckets because the purchased steam was still super heated and the“reactor” steam was just at the saturation temperature at this pointin the turbine. Options were defined and it was decided to addthickness, about fi inch to the third stage wheel and install thickerbuckets and these buckets would have shroud bands. The modifi-cations were made to the turbine rotor and the rotor was installedin the case on the compressor platform awaiting an opportunity tocouple up to the compressor. We did not have to wait long—twomonths before an electrical outage brought the unit down. Duringthe outage the turbine was coupled to the compressor and startedup, as per the operating procedures. The turbine rotor forging that

had been ordered was then received and machined to the new spec-ification. It was then put into a climate controlled storage hangingvertically as a spare.

ANOTHER TURBINE FAILURE

In mid-1999, after several years of smooth operation the turbinerotor vibration began to rise. At 2.5 mils the first vibration alarmsounded in the control room alerting the operation personnel thatthe turbine had a problem. The other equipment vibration levelsremained normal. Fortunately a field operator was in theimmediate area and reported a “strange noise” emitting from theturbine. The resident mechanical engineer was called and he toldoperations that if the vibration reached 4.5 mils to shut down thecompressor train. Within a minute, operations had to shut down thecompressor train.

Upon reviewing the vibration data, all indications were thatthere was a mass unbalance and it was getting worse. The decisionwas to uncouple the turbine and run the compressor with the motoralone. Maintenance forces were mobilized and the turbine was dis-mantled after a quick inspection of the bearings showed no signsof damage. The case was lifted; the problem was obvious. Theshroud band of one of the bucket packets was pealing off like an18-wheeler retread.

Metal pieces were recovered from the recycle gas drive com-pressor turbine after its failure in May. They were believed to bepart of the shroud assembly that covered the turbine buckets. Themetal pieces, all smaller than 1 inch in length and fi inch in width,were heavily banged and deformed. However, one round nipple-shaped piece, which is believed to be one of the tenons, appearedto have an undamaged fracture surface. The turbine componentshad been in service for about two years.

Analysis

SEM Fractography

The metal pieces were first ultrasonic cleaned in a concentratedchemical cleaner to remove grease and loose surface films, andthen the “nipple” and one of the larger pieces were examined undera scanning electron microscope (SEM).

• Tenon (or nipple) piece—It was quite fortunate that the fracturesurface was mostly undamaged except in areas near the circumfer-ence of the tenon (Figure 28). However, the fracture surface isseverely oxidized, which masked out the fine details on thesurface. Still, it could be determined to be a fatigue failure (Figure29) with the typical feature of fatigue striations (Figure 30).

Figure 28. Bucket Tenon.


Figure 29. Fatigue Failure.

Figure 30. Fatigue Striations.

On one end of the fracture, there was clear sign of final ductileoverload fracture (Figure 31). Also known as microvoid coales-cence (MVC). Another final fracture area is found about 90degrees to the first one. Here the microvoids appeared to bedistorted or elongated (Figure 32), which indicates the fracture isat a different orientation. Efforts to locate the crack initiation siteby tracing back from the final fracture areas to the opposite end ofthe fracture surface were not successful, as the area was toodamaged to be recognizable (Figure 33).

A feathery area (Figure 34) across from the second final fracturearea was determined to be a fatigue area (Figure 35). When viewedat higher magnification, all in all, the fatigue area was estimated tocover over 70 percent of the fracture surface.

• Shroud piece—There was not much to look at, because the“fracture surface” had been damaged (rubbed) and also heavilyoxidized (Figure 36). In a cracked area, near one end of the surface,it was also damaged, but it had some scale attached to it (Figure37). Even the surface inside the crack was found to be severelyoxidized (Figure 38). Heavy scaling was also found on the sidesurface of the shroud piece (Figure 39).

Chemical Analysis (by X-Ray Fluorescence)

• Based metal—The chemical compositions of the tenon andshroud pieces were found to be close to that of 422 stainless steels.They are listed below in Table 9.

Figure 31. Ductile Overload Failure.

Figure 32. Elongation of Microvoids.

Figure 33. Badly Damaged Area.

• Oxidized surfaces—The oxygen content ranged from 8 to 14 wt.percent or 24 to 35 atomic percent. In some areas the oxide filmswere so thick that only 2 to 3 percent of chromium was detected.Other than the basic elements like iron, nickel, and manganese,


Figure 34. Feathery Area.

Figure 35. Final Fracture Area.

Figure 36. Fracture Surface.

there were also silicon (0.4 to 1.5 percent), vanadium (0.8 to 1.9percent), aluminum (0.3 to 0.5 percent), zinc (up to 1 percent),copper, (up to 2.4 percent), and sulfur (0.3 percent).

Figure 37. Scale at One End.

Figure 38. Severe Oxidation.

Figure 39. Shroud Piece.

• Scale—The surfaces of the shroud and tenon pieces, includingthe fracture surfaces, were covered with scale. The scale was foundto be rich in calcium (3 to 4 percent), silicon (0.7 to 3.3 percent),


Table 9. Chemical Compositions.

phosphorus (1 to 3.2 percent), vanadium, (0.7 to 1.7 percent),copper (0.5 to 1 percent), aluminum (0.15 to 1.42 percent), andzinc (0.4 to 1.1 percent).

The failure of the tenon has been confirmed by SEM frac-tographs as fatigue, or more precisely corrosion assisted fatigue. Itis also likely a high cycle (or low stress) fatigue failure because:

• The ratio of fatigue area to final fracture area is over 2 to 1. Thismeans the fatigue crack propagated through a large portion of thecross-sectional area of the tenon before final fast fracture occurred.

• The fatigue area was heavily oxidized. In order to have such athick oxide film on the surface, the oxidation is believed to haveoccurred over a long period of time. The final fracture areas, on theother hand, were not heavily oxidized because the fracture tenonwas removed from the turbine shortly after the failure.

The shroud and tenon were confirmed to be 422 stainless steel.With the small amount of alloying element molybdenum andnickel, 422 SS offers better corrosion resistance and higher hard-enability than the lower alloy 400s martensitic SS like 410 or 416.The excess deformation and rubbing damages on the pieces, thelack of secondary cracking, and the MVC appearance of the finalfracture all indicate the material has good ductility and toughness.

The severe oxidation and scaling on the failed pieces are notbelieved to be beneficial to the service life of the turbine compo-nents. It was believed that the steam is of poor quality and iscorrosive to 422 SS. The steam also contains a high amount ofimpurities. The elements found in the scales like calcium, phos-phorous, silicon, and vanadium are believed to be water treatmentchemicals, while the zinc, copper, and aluminum are likely to becorrosion products that were carried over by the steam. Sincesteam is water vapor, it is not supposed to carry many impurities.The steam used in the turbine is thus believed to be very wet andcontained liquid entrainment. Besides oxidation, there was noother sign of environmental degradation like localized corrosion,stress corrosion cracking, or caustic embrittlement on the failedpieces.

The failure of the tenon is believed to be high cycle fatigue. Theactual fatigue duration is not known, but is estimated to be months.The quality of steam is questionable; it is believed to have con-tributed to the scaling and heavy oxidation on the failedcomponents. However, there is not enough evidence to indicate thelow quality steam, or the scaling and oxidation on the parts, are theroot causes of the failure. The materials of construction of thetenon and shroud are 422 SS, which has good corrosion resistanceand mechanical properties. So the failure is not believed to bematerial related.

With the above information engineers revised the shroud designby reducing the overall width of the band by 1/2 inch, thus reducingthe stress on the tenons.

LESSONS LEARNED

• Stress corrosion crack can occur very rapidly when all the rightconditions are present.

• Recommended steam quality

• Using today’s technology almost anything can be repaired.

• If the equipment is critical to the operation of the unit or to thefacility, spare rotors are essential.

• New rotors can have unseen problems.

• 15-5ph material is far better than 17-4ph, relative to stability.

• High thrust shutdowns should not be bypassed.

• The perception of danger is not always accurate or understood.

REFERENCE

Drew, 1994, “Principles of Industrial Water Treatment,” EleventhEdition, Published by Drew Chemical Corporation, Boonton,New Jersey, p. 310.

BIBLIOGRAPHY

“The ASME Handbook on Water Technology for Thermal PowerSystems,” Paul Cohen, Editor-in Chief, Sponsored by theASME Research and Technology Committee on Water andSteam in Thermal Power Systems, EPRI Research Project No.RP 1958-1.

ASTM A-380, 1999, “Standard Practice for Cleaning, Descaling,and Passivation of Stainless Steel Parts, Equipment, andSystems,” American Society for Testing and Materials, WestConshohocken, Pennsylvania.Elliott Specifications for SteamQuality.

General Electric Instructions GEK-63430, “Turbine Steam Purity.”

General Electric Instructions GEK-72281, “Steam Purity—StressCorrosion Cracking.”

General Electric Power Generation Global Product & TechnologySupport, TIL 1231-3, “Maintenance Recommendations forCleaning Deposits and Chemical Contamination from SteamTurbine Rotors,” August 15, 1997.

General Electric Power Systems GEK 98965, “Steam Turbines,Steam Purity for Industrial Turbine.”

Lindinger, R. J. and Curran, R. M., 1981, “Experience with StressCorrosion Cracking in Large Steam Turbines,” The NationalAssociation of Corrosion Engineers, The InternationalCorrosion Forum, Corrosion/81, Paper Number 7, Toronto,Ontario, Canada.

McIntyre, D. R. and Dillon, C. P., 1985, “Guidelines for PreventingStress Corrosion Cracking in the Chemical ProcessIndustries,” MTI Publication No. 15.


422 SS Tenon Shroud

Chromium 11.5 13.5 12.26 12.46Molybdenum 0.75 1.25 1.03 0.97Nickel 0.5 1.0 0.79 0.86Manganese 1.0 max. 0.78 0.75Silicon 0.75 max 0.62 0.46Vanadium 0.15 0.3 < 0.1 < 0.1Phosphorus 0.04 max. < 0.1 < 0.1Iron Balance 84.5 84.5

Manabu Saga is a Manager of theTurbine Design Section in the Turbo-machinery Engineering Department,Mitsubishi Heavy Industries, Ltd., inHiroshima, Japan. He has had experiencewith basic and detail engineering of steamturbines and gas turbines for power gener-ation and development.

Mr. Saga has a B.S. degree (MechanicalEngineering) from Keio University.

Shugo Iwasaki is Acting Manager of theMaterials Laboratory in the HiroshimaResearch and Development Center,Mitsubishi Heavy Industries, Ltd., inHiroshima, Japan. He has 12 years of expe-rience in the evaluation of materials, thedevelopment of new manufacturing pro-cesses, and troubleshooting for turbines,compressors, heat exchangers, and vessels.

Mr. Iwasaki has B.S. and M.S. degrees(Materials Engineering) from Kyushu

University.

Yuzo Tsurusaki is the MechanicalEngineer of the Turbine Design Section inthe Turbomachinery Engineering Depart-ment, Mitsubishi Heavy Industries, Ltd., inHiroshima, Japan. He is a specialist inrotor design and has five years of experi-ence with troubleshooting of steamturbines.

Mr. Tsurusaki graduated from KurumeCollege of Technology (MechanicalEngineering).

ABSTRACT

Mechanical drive steam turbines play an important role as coreequipment in petrochemical plants, and these turbines areprotected for safe operation by an antioverspeed trip device, as

well as other monitoring and protection systems. However, insome cases, a turbine will suffer severe mechanical damage due toimproper operation or failure to activate the protection system as aresult of human error. For urgent plant recovery and to minimizethe duration of risky operation with no spare rotor, a damagedturbine has to be repaired in as short a time as possible. This paperintroduces actual experiences in repairing and reviving catastroph-ically damaged turbine rotors through special welding procedures,based on element tests to find the optimized welding conditions,detailed strength calculations to confirm the integrity, and heattransfer analysis for proper heat treatment process conditions.These basic procedures are discussed to show useful data. Therevived rotor of the extraction condensing turbine was placed backinto the casing and operated. The turbine was uniquely modified inorder to balance the required amount of power and minimize therepair time, to restart the plant as quickly as possible. The casestudy for this optimization is discussed by showing thermody-namic calculations, performance, and repair schedules.

Root cause analysis for the process of the catastrophic failure isexplained, for the integrated control system governing the controland operation positioner systems.

INTRODUCTION

Mechanical drive steam turbines play important roles in petro-chemical plants. For safety in operation, these turbines areprotected by an overspeed trip device, as well as other monitoringand protection systems. However a turbine still can suffer severemechanical damage from improper operation or failed activation ofprotection systems, caused by human error and improper control ofsteam purity.

When rotor damage does occur, a rotor is normally replacedwith a spare rotor and the plant continues to operate with therunning risk of having no replacement rotor. For urgent plantrecovery and to minimize the operating time with no spare rotor, adamaged turbine has to be repaired in as short a time as possible.However, in cases where the damage is catastrophic and existingtechnology cannot repair the damaged rotor, it has to be scrappedand a new rotor must be fabricated. Consequently, the end user hasto wait a minimum of six months to one year for a new rotor.

In order to repair and revive a catastrophically damaged turbinerotor in emergencies in response to end user requirements, a newwelding technique has been developed based on risk analysis ofrepair and laboratory element tests. In addition, to optimizewelding conditions, the authors studied detailed strength calcula-

15

REPAIR TECHNOLOGIES OF MECHANICALDRIVE STEAM TURBINES FOR CATASTROPHIC DAMAGE

byManabu Saga

Manager, Turbine Design Section

Shugo IwasakiActing Manager, Materials Laboratory

Yuzo TsurusakiMechanical Engineer, Turbine Design Section

andSatoshi Hata

Manager, Turbine Design Section


Hiroshima, Japan

tions to confirm integrity, and performed heat transfer analysis forfinding proper heat treatment process conditions.

The authors introduce typical experiences in rotor repair by dis-cussing useful results of the above analysis and the test results.

FAILURE MODE OF STEAM TURBINES

The failure damage modes of mechanical steam turbines areclassified into main components for steam flow paths, rotatingcomponents such as rotors, grooves, disks, and blades, as well asstationary parts such as control valve bearings (Table 1). The basicroot causes for each damage mode are listed. Severe damagemodes include destruction of blades due to excessive centrifugalforce, blade corrosion fatigue failure, and rotor bowing due tointerference/rubbing in transient operations such as initial start up.

Table 1. Failure Damage Mode.

In Figure 1, a typical damage condition is shown in the steamturbine cross section. The first stage nozzle profiles are eroded atthe trailing edges, due to contact with hard solid particles. The flowpath of the diaphragm and blade profiles at the leading edges in thelow-pressure (LP) high moisture section are eroded by waterdroplets. In some cases, the welded zone of the nozzles are affectedby a combination of erosion and corrosion. Corrosion fatiguefailure will occur in the blades of the LP section at the wet and dryenrichment zone under corrosive conditions, if proper watertreatment is not performed during steady and transient operation,or if a corrosive chemical leakage occurs at a heat exchanger.Excessive contact friction at the rotor thrust collar will inducemelting damage of thrust bearing pads, due to excessive thrustforce from drain invasion. In other cases, lubricant oil additivesmay generate oil sludge contaminants on pad lubricating surfaces,and this will reduce bearing durability. If lubricant oil is notselected properly, in the worst case the rotor will rub against thebearings.

TYPICAL FAILURE EXPERIENCE

Rotational equipment manufacturers design, manufacture, anddeliver steam turbines to customers based on well-proven tech-nologies and supply experiences, and therefore they do notexperience catastrophic accidents. However, this paper introducestypical catastrophic damage in using steam turbines.

Figure 2 shows trends in rotation speed and key events from startto excess speed rotor damage, during the site-based solo runningtest. This extraction condensing turbine for driving a charge gascompressor is equipped with a position control backup system. Theturbine was coasted up smoothly to a maximum continuous speedof maximum continuous rate (MCR) 10,279 rpm. Rotating speedwas increased for a moment to electric overspeed trip speed(EOST); however, the trip interlock did not activate and speed

Figure 1. Failure Damage Map.

continued to increase, exceeding EOST speed. After the operatinggovernor function key and the turbine control were shifted to thebackup system, the governing valve suddenly opened from 12percent to 60 percent in a no load condition and the turbine rotorwas accelerated to 170 percent of MCR. Eventually, the turbinerotor disks and blades became damaged, resulting in oil leakage.

Figure 2. Typical Failure Experience (Overspeed Burst).

The control system for this turbine is shown in Figure 3. Thecontrol system has backup modules operating by position control,and this position control is used to supply the actuator output signalinstead of the electric governor in the case of a governor failure.The backup system monitors the actuator output signals from thegovernor and controls two signals to switch the actuator signalsource from governor to position control. Two backup modules areutilized for governing and extraction control valve actuators.

As a result of detailed investigation of the system and signals, itwas found that circulating current from the distributed controlsystem (DCS) into the signal system of the backup modules causedthe supply signal to the actuator to increase, thereby opening thevalve. This caused a large amount of steam flow leading to theabnormal increase in rpm, and the interlock that should haveperformed an emergency valve close did not operate properly.

Circulating current causes a 12 percent to 60 percent deviationof signals between the governor and governor backup module.

The actual damage conditions are shown in Figure 4 and Figure5. The second (high-pressure section) and third stage blades (low-pressure section) have broken off from the rotor grooves, and thedisks and grooves are deformed at the first stage and other stages


Components Damage Mode Root Causes

Stationary

Parts

Rotating

Parts

Flow Path

Parts

Control Valve

Nozzles (HP)

Blades (HP)

Nozzles (LP)

Blades (LP)

Blades (HP)

Blades (LP)

Rotor

Thrust Collar

Journal Shaft

Casing

Diaphragm

Bearings

Seals

Valve Stem

Bending Fatigue Failure

Solid Particle Erosion

Fouling


Fouling

Fouling

Drain Attack Erosion

Fouling

High Cycle Fatigue Failure

Centrifugal Force Failure

High Cycle Corrosion Fatigue

Centrifugal Force Failure

Rubbing & Shaft Bow

& High Vibration Disk SCC

Rubbing & Scratch/Wear

Creep Deformation/ Creep Rapture,

Erosion/Corrosion

Diaphragm Bending Deformation

Erosion & Corrosion

Rubbing & Melting

Rubbing & Erosion

High Steam Velocity

Fluid Excitation

Hard Foreign Materials

Inside of Pipes

Hard Foreign Materials Inside

Steam Impurity

Steam Impurity

High Moisture & High Velocity

Steam Impurity

Excessive Excitation Force

Excessive Over Speed

Improper Water Treatment

Excessive Over Speed

Improper Start Up

Drain Intake, Steam Impurity

Excessive Thrust Force

Improper Oil Supply (UPS)

Operation Over Allowance

Excessive Drain / Galvanic

Operation Over Allowance

Excessive Drain / Galvanic

Excessive Thrust Force

Improper Oil Supply (UPS)

Improper Start Up

Excessive Drain

TIME

500rpm

1st Critical4400rpm

SPEED

(rpm)

2000rpm

Critical Zone

4840rpm

3960rpm

GOV ControlTTV Control

3000rpm

7000rpm

16:30 17:00 17:30 19:00

MGS 8075rpm

Normal 9500rpm

MCR 10279rpm

5000rpm

Valve Opening Deviation was observed.Valve Opening Deviation was observed.

GV Open 12%, ECV Open 100%

However, Back Up System indicated

GV and ECV Open 60%

Preparations for Over Speed Test

EOST Trip 11307rpm

GOV Trip 11407rpm

OST Displayed on EOST Module and DCS

Customer Design EOST Lower than GOV Trip

Control shifted to Back Up System by GOV F Key

GV suddenly opened to 60% due to Bias Signal

EOST Module Recorded Max. 17557rpm

Oil Leakage due to Excessive Vibration

Solo Running Test

Cross Section of Extraction Condensing Turbine

Rotor Disk & Blades Broken at 170% of MCR

No Signal from Interlock Circuit of DCS at OST

Figure 3. Overspeed Burst and Back Up System Error.

of the low-pressure section. The rotor is bowed from a largeamount of friction and contact due to a seriously unbalancedcondition. The journal and thrust bearings are damaged and theirBabbitt metal pad has melted due to a huge friction loss heatingresulting from the excessive overspeed and high loads. Rotor shaftat bearing location had no damage and was protected by Babbittmetal melting promptly. As explained in Figure 5, damage of therotor was catastrophic.

Figure 4. Damaged Condition of Turbine.

Figure 5. Detail Damaged Condition of Turbine.

REPAIR RISK ANALYSIS

For urgent plant recovery and to minimize the duration ofoperation with no spare rotor, this damaged rotor had to be repaired

in as short a time as possible. In order to revive this catastrophi-cally damaged turbine rotor, a special welding procedure wasnecessary.

In order to analyze the applicability of overlay welding for thisturbine, technical risks are evaluated in a common procedure forrotor repair according to experiment data. In addition, experiencewith welding rotors and risk factors are categorized as well-proven, possible, practical, or not well-proven. Table 2 showsdetails of this test.

Table 2. Technical Risk Evaluation.

Elements of welding risk factors include mechanical properties,creep and SCC strength, residual stress, cold cracking, and residualdeformation of the rotor. Risks in operation after repair consist ofthe bonding strength in frictional areas and spiraling around thelabyrinth seal at the high temperature and pressure side.

For the risk analysis and evaluation, element tests wereperformed to find the optimum welding conditions, detailedstrength calculations were performed to confirm the integrity, andheat transfer analysis was performed to achieve proper heattreatment process conditions. These basic procedures are studiedand discussed in the next section.

As explained below, mechanical properties and strength can bemaintained at the same levels as that of a new rotor. However,residual rotor deformation is affected by residual stress in thesurface of the welded shaft section. This deformation must be lessthan 0.005 mm (.0002 inch) when considering required limitcriteria for balance. Though it is possible to calculate residualstress and shaft deformation, it is very difficult to predict such asmall amount of deformation precisely.

Because of this, residual warpage of the rotor still remains as arisk factor, and an actual full scale test is necessary to confirm thatresidual warpage is within allowances, and that there are nochanges when reheated.

TECHNIQUES FOR WELD REPAIR

Technical Issues and Solutions for Weld Repair

Table 3 summarizes the technical issues on weld repair ofdamaged rotor disks and solutions for those issues. The rotor shaftat the bearing has no damage and was protected by the Babbittmetal melting promptly. Damaged pads can be easily replaced withspare pads. However, there are risks in replacing forged rotormaterial with weld deposited material. Forging gives the rotormaterial proper mechanical properties and reduces defects causedby casting, but weld repairs have no such effects. Therefore, it isnecessary to apply welding material that will provide propermechanical properties even though there is no forging in the repairprocess. The proper welding material was selected as explained inthe following section.

REPAIR TECHNOLOGIES OF MECHANICAL DRIVE STEAM TURBINES FOR CATASTROPHIC DAMAGE 17

GV E/H

Actuator

Extraction Flow

TRIP

GOV TRIP

ALARM

OVER SPEED

TRIP Signal

DC 4-20mA

DC 4-20mA

Over Speed

Trip

2/3 Voting

Module

Suction Pressure

ESD

SEQ

TTV ESD

DCS

Trip

SWSOV

Electric

Governor

X

B/UP

Module

B/UP

Module

Speed

Pickup-NO.1

Speed

Pickup-No.2

Local PanelRaise/Lower

Speed Pickup

No.1, 2 & 3

ECV E/H

Actuator

12% Open

100% Open

60%

60%

Manually SW OFF

Circulating Current from DCS

X

��

��

��

��

��

��

��

��

��

Risk Factor

Element for

Welding

Mechanical

Properties

Deterioration

Creep Strength

SCC Strength

Cold Cracking

Stress Relief

Cracking

Blow Holes

DefectsResidual

Deformation

Spring Back

Element for

Operation

Rubbing Spiraling

vs Bonding Stress

Element for

ExperienceUnexpected

Defects

Welded

Position

Boundary

Condition

Evaluation Results/Countermeasures

2.25Cr-1Mo Welded Metal Lab. Test

Stress of Damaged PartsChange of Impact Value and others

8% Decrease of Tensile Strength

Proper Safety Factor of Yield Stress

Welding

Procedure

10% Decrease is Expected from Mech.

Property and Temp. Test

Proper Safety Factor of Creep Life

Expected 300 to 400

Confirmed by SCC Test.

Residual StressResidual Stress is Measured by Test

and Actual Rotor..

NO SCC

Within Allowable

Applied for Test and Actual Rotors

including Heat Treatment & NDE.WPS/PQR Issued

Well-Proven Heat Treatment Procedure

Heat Transfer Analysis for Disks

Circumferential Equal Temperature Profile

Issued Basic Procedure is OK

Heat Transfer Analysis is Required.


including Heat Treatment & NDE.WPS/PQR Issued

Applied for Actual Rotors and Within Criteria

0.005mm for Disks.

No Experience for HP Balance Piston

Basically Expected Within Criteria

Stress Analysis & Actual Size Test are Required.

No Impacts for Creep, SCC by Test and

For Corrosion Fatigue due to HP Sect.

Stress Relief and No Spring Back during

operation to be confirmed by Actual Size Test

No Experience for HP Balance Piston

Experience of HP and LP stage Disks


including Heat Treatment & NDE.

Actual Size Test are Required.

Expected 300 to 400

Heat Transfer Analysis is Required.

No Blowhole Inspected

Long Term

Capability

Confirmed Same Bonding Strength By Lab. Test

Deformation due to Stress Relief by Contact Heating

Stress Relief and No Spring Back during

operation to be confirmed by Actual Size Test

Eccentricity &

Unbalance

By Fine Setting and Machining After Welding &

Balancing, They Can Be Within Criteria.

Actual Size Test are Required

for Damaged Parts.

Table 3. Technical Issues and Solutions on Weld Repair ofDamaged Rotor Disk.

Proper welding methods and conditions should be provided toavoid welding defects that would initiate cracks. Proper weldingconditions were decided by welding laboratory scale test pieces.An ultrasonic test for the weld metal was performed with thesame criterion on minimum defect size as for the base rotormetal.

Temperature distribution and gravity during welding and postweld heat treatment (PWHT) cause unfavorable deformation in therotor. Therefore, the repaired disk was heat treated locally, and therotor was suspended vertically at the PWHT to prevent bowing ofthe shaft. The whole circumference of the repaired disk waswelded and heat treated uniformly.

The welded portion needs to be heat treated to reduce residualstresses due to welding and hardening at the heat affected zone(HAZ), which can cause stress corrosion cracks (SCC) under wetconditions. However, PWHT reduces the strength of base metaland weld metal. Therefore, the repaired disk was heat treatedlocally to avoid degradation of the base metal. An optimum PWHTtemperature satisfying both weld metal strength and HAZ hardnesswas determined.

Selecting Weld Metal

Ni-Mn-Mo steel (AWS:ER110S-G) and 2.25Cr-1Mo steel(AWS:ER90S-G) were proposed as the welding material becauseof their excellent welding characteristics, strength, and toughness.Table 4 shows the comparison of these material properties. Ni-Mn-Mo steel is suitable as weld metal for the disk that requires highstrength at lower temperatures. However, 2.25Cr-1Mo steel issuited for higher temperatures because of its creep strength. Thelatter is preferable because it can be utilized for the entire rotorstage. Thus 2.25Cr-1Mo steel was ultimately selected as the metalfor repair welding. Both materials were welded and evaluated intesting conducted on a laboratory scale test.

Table 4. Weld Metal for Repair.

Welding and Post Weld HeatTreatment on a Laboratory Scale Test Piece

Figure 6 shows the appearance of the test piece after welding.Both types of weld material were welded to a bar shaped rotormaterial (1.25Ni-Cr-Mo steel). The weld metal was formed with athickness of 75 mm (2.95 inches) on the bar by gas tungsten arcwelding (GTAW). Optimum welding conditions were selected toavoid defects and these were controlled constantly throughout thewelding process to obtain uniform heat input to the bar. Table 5shows the welding and PWHT conditions for the test pieces.

Figure 6. Appearance of Welded Test Piece in Laboratory ScaleTest.

Table 5. Welding and PWHT Conditions for Test Pieces.

Figure 7 shows the relationship between stress and time to SCCinitiation reported by Speidel and Bertilsson (1984). Based onthese data, it is estimated that the critical stress that causes SCCafter 105 hours of operation is 420 MPa in a normal steam envi-ronment. Considering the safety factor, the target residual stress onthe weld metal was set at 200 MPa.

The specimens for evaluation were taken in a welded conditionand the heat treated test pieces. Figure 8 indicates the locations ofeach specimen in the test piece. Tensile, impact, hardness meas-urement, microstructure observation, and SCC tests wereperformed.

Evaluation of Laboratory Scale Test Pieces

Welding Defects

Welded test pieces were inspected by ultrasonic testing. Defectsexceeding the criteria [1.6 mm (.06 inch) in diameter] were notfound. According to the analysis by fracture mechanics, the criticaldefect size for cracking is much larger then the criterion.Therefore, it can be concluded that the defect in weld was smallenough to prevent fracture.


G HG H•Same allowable defect size

for weld metal with base

metal

Welding•Welding defects

--•Proper material selection-•Properties of weld metal

PWHT

(Local)

PWHT

Welding

PWHT

Process

F-•Optimum heater layout

based on FE analysis

•Residual stress

CB C D E•Proper heating temperature

and time

•Reduction of strength

on base metal

•High hardness on HAZ

Real scaleLab. scale

A-•Uniform heat input

•Local heating

•Uniform heat input

•Vertical rotor layout

•Deformation

Check points on testSolutionTechnical issues

A: dimension measurement B: Tensile test C: Hardness measurement D: Impact test

E: SCC test F: Residual stress measurement G: Ultrasonic test H: Magnetic particle test

�

�

�(2.4-2.8 inch/min)

�o (302-482oF)

�o (482-572oF)

�o (36oF/h)

o (1112oF)o (1085oF)

o (1112oF)

�o (45oF/h)

o (482oF/h)

�

Figure 7. Estimated Critical Residual Stress for SCC. (CourtesySpeidel and Bertilsson, 1984)

Figure 8. Locations of Specimens for Evaluation.

Mechanical Properties

Table 6 shows results of the mechanical tests on test pieceswelded using 2.25Cr-1Mo steel and heat treated for 10 hours at585°C and 600°C (1085°F and 1112°F). Figure 9 and Figure 10indicate distribution of hardness on the welded test piece and theheat treated piece at 600°C (1112°F).

Hardness, tensile properties, and impact value on both heatingtemperatures satisfied the target for the weld metal and the require-ment for base metal, except for hardness heated at 585°C (1085°F).Figure 11 shows the relationship of maximum hardness andminimum tensile strength with PWHT temperature. Hardness andtensile strength of the weld metal tends to become weaker as thewelding temperature becomes higher. On the other hand, tensilestrength of base metal is almost constant in the temperature rangetested. This tendency comes from the high temperature propertiesof base metal.

The maximum hardness of HAZ satisfies the target at lowertemperatures. On the other hand, the minimum tensile strength ofthe weld metal satisfies the target at higher temperature.Therefore, temperature range that satisfies both targets is between595°C and 600°C (1103°F and 1112°F) when welding with2.25Cr-1Mo steel.

Table 7 shows the results of mechanical tests for a test piecewelded with Ni-Mn-Mo steel. The tensile strength of Ni-Mn-Moweld metal is higher than the one of 2.25Cr-1Mo heat treated at thesame time and at the same temperature [10 hours, 600°C(1112°F)]. This shows that the Ni-Mn-Mo weld metal has anadvantage in strength over 2.25Cr-1Mo at low temperatures.

Sensitivity for Stress Corrosion Cracking

An SCC test was performed in an accelerating NaCl solution forthe test piece, for the 2.25Cr-1Mo weld metal. Table 8 shows thetest conditions. Figure 8 shows the location of specimens on the

Table 6. Mechanical Properties of Specimens (Weld Metal: 2.25Cr-1Mo Steel).

Figure 9. Distribution of Hardness on Welded Test Piece.

Figure 10. Distribution of Hardness on Heat Treated Test Piece.

test piece. Stress around the yield point of material was applied byU-bending. The surface of the test pieces was observed up to 1000hours.

Figure 12 indicates the appearance of the HAZ specimen afterthe 1000 hour test. Although the surface became rough bycorrosion, cracks did not occur.


(60,900 psi)

σa σy(29,000 psi)

σy(87,000 psi)

σa σy5

��

��

��

��

��

A

A’

B

B’��

75

A - A’

B - B’

��

C’

C

C - C’

��

350286

243

370324

277

Vickers Max

hardness, HV Min

Charpy

2mm(0.08inch)

U Notch

J (ft-lbf)

Reduction of area

%

Elongation

%

Yield strength

MPa (psi)

Tensile strength

MPa (psi)

-87 (64)

98 (73)

67 (50)

-Thickness

direction

39 (29)

39 (29)

201 (148)

153 (113)

191 (141)

201 (148)

77 (57)

112 (83)

Longitudinal

direction

40

40

71.3

69.7

68.9

64.0

66.5

66.5

15

15

28.0

28.0

24.0

23.2

23.2

23.2

600 (87,000)

638 (92,500)

645 (93,500)

642 (93,100)

817 (118,500)

850 (123,300)

785 (113,900)

785 (113,900)

720 (104,400)

785 (113,900)

793 (115,000)

792 (114,900)

938 (136,000)

971 (140,800)

927 (134,500)

922 (133,700)

Weld

metal

HAZBase

metal

Target for weld metal

Requirement

for base metal

585oC (1085oF), 10h

350309

210

339297

262

Vickers Max

hardness, HV Min

-

91 (67)

143 (106)

66.5

66.5

22.5

22.5

791 (114,700)

812 (117,800)

932 (135,200)

951 (137,900)

Base

Metal

Charpy

2mm(0.08inch)

U Notch

J (ft-lbf)

Reduction of area

%

Elongation

%

Yield strength

MPa (psi)

Tensile strength

MPa (psi)

-83 (61)

94 (69)

99 (73)

Thickness

direction

39 (29)

39 (29)

229 (169)

197 (145)

83 (61)

99 (73)

Longitudinal

direction

40

40

74.3

74.3

69.7

68.2

15

15

28.5

28.5

24.5

24.0

600 (87,000)

638 (92,500)

602 (87,300)

602 (87,300)

795 (115,300)

808 (117,200)

720 (104,400)

785 (113,900)

722 (104,700)

725 (105,200)

913 (132,400)

921 (133,600)

Weld

metal

HAZ


Requirement

for base metal

600oC (1112oF), 10h

Figure 11. Relationship Between PWHT Temperature andMechanical Properties.

Table 7. Mechanical Properties of Specimens (Weld metal: Ni-Mn-Mo Steel).

Table 8. Conditions of SCC Test.

Welding and Heat Treatment for the Rotor Disk

An actual rotor was welded and heat treated for a real scale test.Figure 13 shows the appearance of a welded disk. 2.25Cr-1Mo waswelded 60 mm (2.36 inches) in height by GTAW. Welding speedwas kept constant to prevent bowing of the rotor shaft.

Local heating was applied for the PWHT. Layout of the heater wasdecided based on the result of a finite element analysis to reduce thethermal stress around the edge of the heating zone due to temperaturedistribution. Figure 14 shows the schematic view of the heaterlayout. The rotor was suspended vertically during heating to avoidbowing due to the influence of gravity on the horizontal rotor layout.

Figure 12. Surface of Specimens after 1000 Hours in NaClSolution.

Figure 13. Welded Disk on Real Scale Test.

Figure 14. Heater Layout on Local PWHT for Welded Disk.

Evaluation of the Welded Rotor

Deformation

Measurement was performed for bowing of the shaft and otherdeformations due to welding and PWHT. No unfavorable defor-mation was found.

Welding Defects

The welded disk was inspected by ultrasonic testing. Defectsexceeding the criterion [1.6 mm (.06 inch) in diameter] were notfound. According to the analysis of fracture mechanics, the criticaldefect size for cracking is much larger then the criteria. Therefore,it can be concluded that any defects in the weld were small enoughto prevent fractures.

Residual Stress

Figure 15 shows the appearance of the residual stress measure-ment. A strain gauge was attached to a square weld metal area, anda groove of 1 mm (.04 inch) in depth was cut around this area.


350315

280

340313

292

Vickers Max

hardness, HV Min

-

180 (133)

179 (132)

68.9

68.2

24.6

24.0

796(115,500)

799(115,900)

932 (135,200)

932 (135,200)

Base

metal

Charpy

2mm(0.08inch)

U Notch

J (ft-lbf)

Reduction of area

%

Elongation

%

Yield strength

MPa (psi)

Tensile strength

MPa (psi)

-96 (71)

97 (71)

77 (57)

Thickness

direction

39 (29)

39 (29)

130 (96)

138 (102)

196 (145)

178 (131)

Longitudinal

direction

40

40

67.3

69.7

64.8

68.2

15

15

27.8

28.4

24.2

23.0

600 (87,000)

638 (92,500)

766 (111,100)

776 (112,500)

828 (120,100)

826 (119,800)

720 (104,400)

785 (113,900)

858 (124,400)

852 (123,600)

921 (133,600)

928 (134,600)

Weld

metal

HAZ


Requirement

for base metal

600oC (1112oF), 10h

o

Tensile strength of Base metal

Tensile strength of Weld metal

Hardness of Weld metal

Weld metal: 2.25Cr-1Mo steel

Holding time: 10h

(87,000 psi)

(145,000 psi)

(1058oF) (1130oF)

o (353oF)

o (1112oF)

Figure 15. Residual Stress Measurement for Repaired Disk.

Measured strain after cutting was translated into the releasedresidual stress. Measured residual stress was around 200 MPa(tension) in a circumferential direction. This corresponds to thetarget value required to prevent SCC and the value estimatedthrough the finite element analysis.

Mechanical Properties

Hardness on the surface of the weld metal was measured. Figure16 indicates the measured hardness. The measured hardness of 240to 270 HV is almost the same as that of the final bead on the labo-ratory test piece heated at 600°C (1112°F) as shown in Figure 10.Therefore, it can be estimated that the hardness of HAZ was alsoreduced within 350 HV.

Figure 16. Hardness Distribution on Real Scale Disk.

NUMERICAL ANALYSIS

A finite element analysis was performed to set the optimumheater layout at local PWHT. Temperature distribution and thermalstress by elastic analysis were estimated through a heating processthat included heating and cooling. Heater layout was adjusted tomaintain a target thermal stress of less than 200 MPa. Analysis wasperformed for the heating of both the disk and the shaft.

Temperature and Stress Distribution on thePost Weld Heat Treatment Area of the Disk

Figure 17 shows the temperature distribution at the start of tem-perature hold after heating and the temperature change at each spotfor positioned heaters. The temperature at the root of the heateddisk and adjacent disks was less than 450°C (842°F). Therefore,temperature at the root of the disk should be maintained at morethan 450°C (842°F) to satisfy the target residual stress.

Figure 18 shows the distribution of circumferential stress aroundthe heated disk at the start of temperature hold. Circumferential

Figure 17. Estimated Temperature Distribution on Local PWHTfor Disk.

stress is around �200 MPa on the weld metal. This means that theweld metal is compressed by the disk during heating. This stress isturned into tensile stress after cooling down. It is presumed that thecompression stress during heating changes into tensile stress bycreep deformation.

Figure 18. Estimated Circumferential Stress Distribution on LocalPWHT for Disk.

Temperature and Stress Distribution ofPost Weld Heat Treatment for the Shaft Portion

Figure 19 shows the temperature distribution at the start of tem-perature hold and the temperature change at each spot forpositioned heaters that satisfy the uniform temperature on thewhole surface of the weld metal. The center of the shaft’s heatedportion is more than 500°C (932°F). This means there is a higherpossibility of unfavorable deformation by welding of the shaft andPWHT than that for the disk.

Figure 19. Estimated Temperature Distribution on Local PWHTfor Shaft.


0

100

200

300

400

500

600

700

0 10 20 30 40 50Time[h]

A

B C

D

E G, HF

o(662oF)

(1112oF)

(1292oF)

(-29,000 psi)

(14,500 psi)

(27,700 psi)

150

250

350

450

0 10 20 30 40 50 60

0o

180o

90o

270o

Target

o (1112oF)

60

60

Weld

metal

Measured

surface

(2.36 inch)

(2.36 inch)

(2.36 inch)

��

�

�

o(392oF)

(1112oF)

(1292oF)

Figure 20 shows the distribution of circumferential stress aroundthe heated shaft at the start of temperature hold. The stress aroundweld metal is low and the compressive stress is 75 MPa maximumin a circumferential direction around the edge of the heater.Therefore, it can be determined that the residual tensile stress onthe shaft is lower than that on the disk. Bending residual stressaround 80 MPa can be estimated on the disk adjacent to the heatedportion. This is almost the same as that on the shaft.

Figure 20. Estimated Circumferential Stress Distribution on LocalPWHT for Shaft.

TYPICAL REPAIR EXPERIENCE

Experience in overlay welding for repairs is listed in Table 9.These overlay welds were performed in emergency cases. In con-sidering rotor materials and the required strength, overlay weldingwas applied to a portion of blade root grooves and rotor disks byselecting suitable weld metals. In this paper, two typical repairexperiences for mechanical drive turbines are introduced.

Table 9. Application of Rotor Overlay Welding.

Welding Repair for a Large Rotor

First, Figure 21 shows the case for a large rotor in a charge gascompressor drive steam turbine at an ethylene plant. The ninthstage disk of this turbine had heavy frictional contact with thediaphragm, and the blades and one side of the disk were damaged.The damaged portion was cut off and repaired successfully basedon the results of the above laboratory test and analysis. Figure 22shows hardness check results in the repair process used to confirmthat the welding conditions are the same as in the test results.

Welding Repair for High-Speed Rotor

Second, as explained in the catastrophic damage of Figure 4 andFigure 5, the case for a high-speed machine is explained in detail.All stages of this rotor were damaged; thus it was necessary to

Figure 21. Welding of the Rotor Disk.

Figure 22. Hardness of the Welded Metal for the Actual Rotor.

repair all stages. However, it was also necessary to start the plantvery quickly, and the time for repair was limited. In discussion withthe customer regarding minimum required power in the plant,turbine performance and blade strengths of each stage were decidedupon with the customer, and as a result, the number of stages werereduced and heat drop distribution was changed for the LP section.Figure 23 shows the condition of welding repair on the rotor disks.Figure 24 shows rotor residual warpage check after welding wascompleted. Rotor residual warpage after welding and finalmachining were within the criteria of 0.005 mm (.0002 inch), andthe blades for each stage were assembled as shown in Figure 25.

Figure 23. Welding Repair of Rotor Disks.

Figure 26 shows the comparison between the turbine perform-ance curves at the original 9555 kW and after repairs at 6700 kW,with the number of stages reduced by cutting off the last fifth,


(-14,500 psi)

(7,300 psi)

(10,700 psi)

Removal ofDamaged Disks Overlay Welding

1.25Ni Cr Mo VRotor DiskHP and LP Section

(4 Stages)

MechanicalDriver

6

Removal ofDamaged Disk

Overlay Welding

1.25Ni Cr Mo VRotor DiskLP Section(1 Stage)

MechanicalDriver

5

Removal of theWhole Root Groove

Overlay Welding

Whole Root GrooveCircumference

PowerGeneration

4

Overlay Welding2.5Ni Cr Mo VDisk Side PlaneCircumference

PowerGeneration

3

Removal of Cracks Overlay Welding

Cr Mo VRoot Groove (PartiallyPowerGeneration

2

Ni Cr Mo V

RotorMaterials

Overlay WeldingRoot Groove Partially

Ship1

MethodWeldedPart

UseNo.

Figure 24. Rotor Residual Warpage Check.

Figure 25. Blade Assembly after Welding Repair.

sixth, and seventh stages. Blades and diaphragm strengths werechecked and exhaust vacuum pressure was limited. This repairedand revived the rotor of the extraction condensing turbine, and itwas actually placed back into the casing and operated successfullyat a condition with low vibration and full load for three monthsuntil the new rotor was prepared.

Figure 26. Actual Operation of Repaired Rotor.

CONCLUSIONS

This paper introduces typical steam turbine engine damagemodes and causes, and actual experiences in repairing and revivingcatastrophically damaged turbine rotors through special weldingprocedures, based on risk analysis for rotor repair from a system-atic viewpoint.

Root cause analysis in the catastrophic failure process isexplained, regarding the integrated control system for governorcontrol and backup system. The practical repair techniques aredeveloped based on element tests to find optimized welding condi-tions and detailed strength calculation to confirm the integrity, andheat transfer analysis for proper heat treatment process conditions.

These basic procedures are discussed to show useful data. Thecase study for this optimization is discussed by showing thermo-dynamic calculation, performance, and repair schedule. Finally,this turbine was uniquely modified in order to balance requiredpower and repair time, in order to quickly restart the plant in anemergency case, to acquire customer satisfaction. The revivedextraction condensing turbine rotor was placed back into the casingand operated successfully at full load.

REFERENCES

Speidel, M. O. and Bertilsson, J. E., 1984, International BrownBoveri Symposium on Corrosion in Power GeneratingEquipment, p. 331.

ACKNOWLEDGEMENT

The authors gratefully wish to acknowledge the following indi-viduals for their contribution and technical assistance in analyzingand experiments: O. Isumi and M. Wakai of Mitsubishi HeavyIndustries Ltd.


John C. Nicholas is the owner, Director,and Chief Engineer of Rotating MachineryTechnology, Incorporated, a company thatrepairs and services turbomachinery, andmanufactures bearings and seals. He hasworked in the turbomachinery industry for28 years in the rotor and bearing dynamicsareas, including five years at Ingersoll-Rand and five years as the Supervisor of theRotordynamics Group at the Steam TurbineDivision of Dresser-Rand.

Dr. Nicholas, a member of ASME, STLE, and the VibrationInstitute, has authored over 40 technical papers, concentrating hisefforts on tilting pad journal bearing design and application. Hereceived his B.S. degree from the University of Pittsburgh(Mechanical Engineering, 1968) and his Ph.D. degree from theUniversity of Virginia (1977) in rotor and bearing dynamics. Dr.Nicholas holds several patents including one for a spray-barblocker design for tilting pad journal bearings and another con-cerning bypass cooling technology for journal and thrust bearings.

John A. Kocur, Jr., is a Machinery Engi-neer in the Plant Engineering Division atExxonMobil Research & Engineering, inFairfax, Virginia. He has worked in the turbomachinery field for 20 years. He pro-vides support to the downstream businesswithin ExxonMobil with expertise on vibra-tions, rotor/aerodynamics, and healthmonitoring of rotating equipment. Prior tojoining EMRE, he held the position ofManager of Product Engineering and

Testing at Siemens Demag Delaval Turbomachinery. There Dr.Kocur directed the development, research, engineering, and testingof compressor and steam turbine product lines.

Dr. Kocur received his BSME (1978), MSME (1982), and Ph.D.(1991) from the University of Virginia and an MBA (1981) fromTulane University. He has authored papers on rotor instability andbearing dynamics, lectured on hydrostatic bearings, has been acommittee chairman for NASA Lewis, and is a member of ASME.Dr. Kocur holds a patent on angled supply injection of hydrostaticbearings.

ABSTRACT

The American Petroleum Institute (API) has recently imple-mented new rotordynamic stability specifications for centrifugalcompressors. The specifications consist of a Level I analysis thatapproximates the destabilizing effects of the labyrinth seals andaerodynamic excitations. A modified Alford’s equation is used toapproximate the destabilizing effects. If the compressor fails theLevel I specifications, a more sophisticated Level II analysis isrequired that includes a detailed labyrinth seal analysis.

Five modern high-pressure example centrifugal compressors areconsidered along with a classic instability case, Kaybob. Afterapplying API Level I and Level II stability analyses and reviewingthe results, design changes are made to stabilize the compressors,if necessary. For these cases, the API stability specifications areused to identify the component with the greatest impact on rotorstability. Specifically, the balance piston seal and impeller eyeseals are analyzed. The suitability of applying the modifiedAlford’s equation to compressors with multiple process stages isexamined and compared to the full labyrinth seal analysis.Important aspects of labyrinth seal analyses are discussed, such asseal clearance effects, inlet swirl effects, and converging, divergingclearance effects. Finally, a modal approach to applying thelabyrinth seal calculated cross-coupled forces is presented. For allfive example compressors, the modified Alford’s force was deter-mined to produce the worst case stability level compared tolabyrinth calculated forces.

INTRODUCTION

Centrifugal compressor instability became a major problem inthe 1960s due to increased speeds and power ratings. Unstablecompressors exhibited a high subsynchronous vibration whosevibration frequency coincided with the rotor’s first fundamentalnatural frequency. Two famous and classic centrifugal compressorinstability cases from the early 1970s are referred to as Kaybob(Smith, 1975; Fowlie and Miles, 1975) and Ekofisk (Geary, et al.,1976). Both problems occurred onsite and the solution involved acostly and time-consuming effort ultimately requiring rotorredesigns.

As a result of these experiences, the evaluation of rotor systemstability has become an essential part of rotordynamic analyses androtating machinery design. Most often, the lowest or first mode,corresponding to the rotor’s first fundamental natural frequency, isthe mode that is “reexcited” causing the subsynchronous vibrationand rotor instability. The primary results of a stability or damped

25

ROTORDYNAMIC DESIGN OF CENTRIFUGAL COMPRESSORSIN ACCORDANCE WITH THE NEW API STABILITY SPECIFICATIONS

byJohn C. Nicholas

Owner, Director, Chief Engineer


Wellsville, New York

andJohn A. Kocur

Machinery Engineer

ExxonMobil Research and Engineering Company

Fairfax, Virginia

natural frequency analysis are the stability prediction from the realpart and the predicted instability frequency from the imaginary partof the solution roots or eigenvalues.

Some of the earliest stability work includes two classic publica-tions by Gunter where basic stability methodology (Gunter, 1966)and internal friction excitation (Gunter, 1967) are discussed. Initialattempts at developing a stability computer code were made byLund (1965). Lund’s code was not easy to use and it was said thatif one knew the approximate answer the code would converge onthe solution. Another early computer code was by Ruhl and Booker(1972). They presented both finite element and transfer matrixsolution techniques using a Muller’s method solver. For lightlydamped systems, their transfer matrix solution analysis workedfine but for more heavily damped structures, such as fluid filmbearing supported turbomachinery, the program’s analysis method-ology produced incorrect and false modes.

These shortcomings were overcome in Lund’s landmarkstability paper (1974). Lund not only outlines a detailed transfermatrix solution procedure, but also describes how the stabilityresults may be presented to study machine design parameters. Histransfer matrix solution is able to search for the first several modesvery efficiently, solving for the most important lowest frequencymodes first, although in random order.

Ruhl’s transfer matrix analysis was updated to include flexiblesupports by Bansal and Kirk (1975), replacing Muller’s solver witha Cauchy-Rieman condition finite difference algorithm plus aNewton-Raphson search solution specified by Kirk (1980). Withminor differences, this was essentially the same solution as used byLund (1974). Both procedures work but will occasionally skipmodes especially when asymmetric flexible supports are includedat the bearing locations. Other transfer matrix computer programshave been developed based on Lund’s original analysis such asBarrett, et al. (1976).

More recent computer codes are based on a finite elementsolution that successfully extracts all of the correct modes. Thesefinite element code authors include Nelson and McVaugh (1975),Rouch and Kao (1979), Edney, et al. (1990), Chen (1996), andRamesh and Kirk (1993). One disadvantage of the finite elementanalysis was that the problem size increases dramatically with thenumber of elements used to model the rotor, resulting in longer runtimes compared to the transfer matrix method. However, this is nolonger an issue with the fast processing speed of modern personalcomputers. Additionally, some methods require that all roots befound, extracting the highest natural frequency first and endingwith the lowest, most important mode. Alternate solution tech-niques are now available that extract the lowest eigenvalues first(Murphy and Vance, 1983).

The most recent API Standard 617, Seventh Edition (2002), forcentrifugal compressors includes stability acceptance criteria alongwith analytical procedures. The stability specification is segmentedinto two parts: a simplified Level I analysis and a detailed Level IIanalysis. The Level I analysis is meant to be a screening process inwhich a quick and simple analysis can be conducted to filter outmachines that are well away from the instability threshold. Level Iutilizes a modified Alford’s equation (Alford, 1965) to estimate thedestabilizing forces. The Tutorial on Rotordynamics, APITechnical Publication 684 (2005), presents a discussion of themodified Alford’s equation. The more involved Level II analysisrequires that the dynamic properties of labyrinth seals be includedimplicitly through the use of an appropriate labyrinth seal code.Initial labyrinth seal computer codes based on the Iwatsubo, et al.(1982), solution were developed by Childs and Scharrer (1986b).Kirk (1988a, 1988b, 1990) further extended the work of Childs andScharrer.

First published experimental results for gas labyrinth stiffnesscoefficients are presented in Benckert and Wachter (1979, 1980).Damping coefficients were not obtained since only static pressuremeasurements of the individual chambers were made. Thieleke

and Stetter (1990) as well as Kwanka, et al. (1993), have alsocarried out similar efforts.

The research summarized in Childs (1993) provided the firstmeasurements of labyrinth seal damping coefficients. While theresults gave the first comprehensive basis for comparison againstpredictions, Childs and Ramsey (1991) revealed the importance oftesting at or near the application conditions. The seal test rig wasfurther extended toward this goal by Childs and Scharrer (1986a)and then again by Elrod, et al. (1995).

Wagner and Steff (1996) further expanded the existing experi-mental knowledge database to geometries and gas conditionsmatching industrial applications, namely, pressure differential,size, and speed. Pressures of 70 bar (1015 psi) were possible atsurface speeds of up to 157 m/s (515 f/s).

The main objective of this paper is to examine the stabilityresults for several industrial representative centrifugal compres-sors. The API Level I modified Alford’s cross-coupling forcecalculation is examined to determine if it is indeed a conservativeestimation of the compressors destabilizing forces by comparing itto the API Level II labyrinth seal calculated forces based on Kirk(1988a, 1988b, 1990). Also, specifics of the labyrinth analysis areexamined to determine what parameters are key in determiningcentrifugal compressor stability. Some of the parameters examinedinclude bearing clearance tolerance range, labyrinth seal clearance,and labyrinth seal inlet swirl effects.

THE KAYBOB INSTABILITY

As a historical perspective, a brief summary of the Kaybobinstability will be presented (Smith, 1975; Fowlie and Miles,1975). This nine-stage low-pressure natural gas injection compres-sor was commissioned in 1971 in Alberta, Canada. Key operatingparameters are summarized in Table 1. The maximum continuousspeed (MCS) is 11,400 rpm with 18 MW gas at 1150 psi inlet and3175 psi discharge. The bearing span, Lb, to midshaft diameter,Dms, ratio is 13.2, indicating a very flexible shaft. The compres-sor’s cross section is shown in Figure 1.

Table 1. Example Centrifugal Compressor Design Data.

Figure 1. Kaybob Injection Compressor. (Courtesy, Smith, 1975)


Parameter Kaybob Example #1 Example #2 Example #3 Example #4 Example #5

Type Injection HydrogenPropane

Refrigeration Injection #1 Injection #2Mixed

Refrigeration# Stages 9 10 3 4 4 5

Configuration Back-Back Straight Straight Straight Straight StraightSpeed (rpm) 11,400 10,750 3,000 12,700 9,900 3,000

Rotor Weight (lbm) - 1,340 540 422 44,730Horse Power (hp) - 12,000 65,000 30,000 11,000 93,000

Pin (psi) 1,150 800 20 1,200 5,600 60Pdisch (psi) 3,175 1,725 100 3,300 9,000 320

Mole Weight 18 6 44 17 25 25Bearing Span (in) 59.7 65 230 59.3 47.8 222

Dms (in) 4.5 5.75 22.4 6.69 6.59 21.8Lb/Dms 13.2 11.3 10.3 8.86 7.26 10.2

Bearings 5 Pad Tilt 5 Pad Tilt 5 Pad Tilt 4 Pad Tilt 5 Pad Tilt 5 Pad TiltSeals Oil Dry Gas Dry Gas Dry Gas Dry Gas Dry Gas

Balance Piston Tooth Laby Honeycomb Tooth Laby Tooth Laby Tooth Laby Tooth LabyAlford’s Qa (lbf/in) - 57,250 23,046 68,015 72,854 41,907

36,689

The severity of the instability may be seen in the orbit of Figure2. Note that the outline of the five pad tilting pad bearing is clearlyevident in the 6.0 by 9.0 mils peak-to-peak orbit. From Figure 3,the 6.3 mil instability is obviously subsynchronous, reexciting thecompressors first critical speed.

Figure 2. Kaybob Instability Orbit. (Courtesy, Smith, 1975)

Figure 3. Kaybob Instability. (Courtesy, Fowlie and Miles, 1975)

Attempts to eliminate the instability included bearing redesigns,oil seal modifications, labyrinth seal modifications, balance pistonmodifications, a vaneless diffuser retrofit, a squeeze film damperretrofit, and, finally, at least two rotor redesigns (Figure 4). Thesecond rotor redesign included increasing the midshaft diameter.Initially, existing impeller forgings were used, cutting and weldingthe impeller hub to increase the impeller inside diameter to accom-modate the increase in shaft diameter (Figure 5).

Clearly, this effort was extremely costly and time consuming.However, it was instrumental, along with the Ekofisk instability, inproviding motivation for improved analytical capabilities, ulti-mately resulting in the existing stability and labyrinth seal codes aswell as the new API stability specification.

LOGARITHMIC DECREMENT

The key parameter in stability analyses and the API stabilityacceptance criteria is the logarithmic decrement or log dec. The logdec is a measure of the rate of decay of free oscillation and is a con-venient way to determine the amount of damping present in thesystem. Greater damping values produce faster decay rates andmore stable systems.

Figure 4. Kaybob Rotor Modifications. (Courtesy, Smith, 1975)

Figure 5. Kaybob Impeller Modification. (Courtesy, Fowlie andMiles, 1975)

The log dec is defined as the natural logarithm of the ratio of anytwo successive amplitudes. Referring to Figure 6, the log dec isdefined as:

(1)

For stable systems, with a positive rate of decay, the log dec ispositive. For unstable systems with a negative rate of decay, the logdec is negative. Stable systems with positive log dec values containsufficient damping to overcome an initial excitation. The resultingdisplacements will dissipate over time. Conversely, unstablesystems with negative log dec values do not contain sufficientdamping to overcome the excitation, resulting in increasing dis-placements over time.

The log dec can also be related to the real, s, and imaginary, ωd,parts of the eigenvalue as:

(2a)

(2b)

where ωd and Nd are the damped natural frequency in rad/sec andrpm, respectively.


27

Original Rotor Lb = 59.7"

1st Modification Lb = 53.4"

2nd Modification Lb = 53.4" w/ Increased Shaft Diameter

δ =⎛⎝⎜

⎞⎠⎟ln

X

X1

2

δ πω

= −⋅2 s

d

δ = −⋅60 s

Nd

Figure 6. Stable Vibration Wave Form.

EXAMPLE #1—12,000 HP, 10 STAGE HYDROGEN COMPRESSOR

The first example is a 10 stage 12,000 hp hydrogen centrifugalcompressor with a 65 inch bearing span and a Lb/Dms = 11.3. Therotor weighs 1340 lbm and operates at an MCS of 10,750 rpm withdry gas seals and five pad tilting pad journal bearings (Table 1).The inlet pressure is 800 psi with a 1725 psi discharge pressure anda gas mole weight of 6.0. Base log dec values with zero aerody-namic cross-coupling, Q, are 0.26 and 0.40 for minimum andmaximum bearing clearances, respectively.

From API Specification 617, Seventh Edition (2002), themodified Alford’s equation is:

(3)

where:βc = 3.0hp = 12,000 hp (total all stages)Hc = Varies stage-to-stageDc = Varies stage-to-stageN = 10,750 rpmρratio = 1.5 (total across compressor)

Q is calculated for each stage using the above values. Since thestage density ratio and horsepower were not available, the stagehorsepower was assumed to be one-tenth of the total value shownabove. The stage density ratio was assumed to be the total densityratio shown above, raised to the power of 1/10. The impellerdischarge width and impeller diameter are stage-to-stage variables.The summation of all 10 Q values is the API Alford calculated oranticipated cross-coupling value of Qa = 57,250 lbf/in. With Qalumped at the rotor midspan, the resulting log dec values are �0.40and �0.23 for minimum and maximum bearing clearances, respec-tively. The API stability acceptance criterion is a log dec greaterthan 0.1. Thus, a Level II analysis is required.

Prior to any analysis, it had already been decided to use ahoneycomb seal on the balance piston. Thus, the stability resultsreported above include the honeycomb seal dynamic properties(Scharrer and Pelletti, 1994).

With the labyrinth seal geometry, stage gas properties, and stagepressures used as input, a labyrinth seal analysis (Kirk, 1990), isconducted for each of the 10 impeller eye seals. A gas swirl valueat the seal inlet is assumed to be 0.6 (60 percent of rotationalspeed). To be conservative, minimum eye seal clearances are usedconsidering the machining tolerance range for the seal and the sealsleeve. The resulting total labyrinth calculated Q for all 10 eyeseals is 15,700 lbf/in.

The shaft seals, with a much lower pressure drop, are neglected.Additionally, the seal flow enters the seal from a stationary part.Thus, the inlet swirl value is low compared to the eye seal swirl(Figure 7). High inlet swirl results in high Q values.

Figure 7. Centrifugal Compressor Labyrinth Seals.

With the total eye seal Q lumped at the rotor center to be con-servative, the resulting log dec values are 0.08 and 0.28 forminimum and maximum bearing clearances, respectively.

Instead of estimating the labyrinth eye seal inlet swirl, it can becalculated by modeling the gap between the impeller face and thestator as in Figure 8 (Kirk, 1990). With the inclusion of impellergap modeling, the inlet swirl is calculated at 0.52 with a resultingQ value for all eye seals of 7600 lbf/in. With this total eye seal Qlumped at the rotor center, the resulting log dec values are 0.17 and0.35 for minimum and maximum bearing clearances, respectively.As with the Level I analysis, the API stability acceptance criterionfor Level II is a log dec greater than 0.1. Thus, this compressorpasses API over the bearing clearance tolerance range.

Figure 8. Labyrinth Seal Modeling for Inlet Swirl Calculation.

These results are summarized in Table 2 and shown in Figures 9and 10 for minimum and maximum bearing clearances, respec-tively. Also of note from Figures 9 and 10 are the Q values for zerolog dec, Q0. These are 22,500 and 40,000 lbf/in for minimum andmaximum bearing clearances, respectively. Clearly for this appli-cation, the predicted stability levels using the modified Alford’sforce is conservative.


X

12

t

High Swirl, High Δ P

Low Swirl, Low Δ P

( )( ) ( )Q

hp

H D Nc

c cratio=

β ρ63 000,

Table 2. Hydrogen Compressor Stability Summary.

Figure 9. Hydrogen Compressor Stability—Minimum BearingClearance.

Figure 10. Hydrogen Compressor Stability—Maximum BearingClearance.

Eye Seal Inlet Swirl Effects

The effect of inlet swirl on total eye seal Q is shown in Figure11. The corresponding effect on stability is presented in Figure 12.Note that the inclusion of swirl brakes (Childs and Ramsey, 1991;Moore and Hill, 2000), on the eye seals with an estimated inletswirl value of 0.3 increases the log dec to 0.28 for minimumbearing and eye seal clearances.

Eye Seal Clearance Effects

The effect of the seal clearance on total eye seal Q is shown inFigure 13. The machining tolerance range is 16 to 20 milsdiametral. Assuming that the minimum eye seal clearance decreasesby 4 mils diametral due to centrifugal expansion of the impeller, theminimum operating clearance of 12 mils diametral is also shown onFigure 13. The corresponding effect on stability for an inlet swirl of0.6 and minimum bearing clearance is presented in Figure 14.

Figure 11. Hydrogen Compressor Q Versus Laby Eye Seal InletSwirl.

Figure 12. Hydrogen Compressor Stability Versus Laby Eye SealInlet Swirl.

Figure 13. Hydrogen Compressor Q Versus Laby Eye SealClearance.

The effects on stability of converging and diverging eye sealclearances are illustrated in Figure 15. Assuming a constantclearance distribution for all eye seal teeth of 12 mils diametral, theresulting log dec is 0.6. If the clearance converges from seal inlet(16.0 mils diametral) to seal discharge (8.0 mils diametral), the logdec decreases to �0.11. Conversely, a divergent clearance frominlet (8.0 mils diametral) to discharge (16.0 mils diametral)produces a log dec of 0.23. These values assume minimum bearingclearance and an inlet swirl value of 0.6.

Clearly, a divergent clearance distribution is preferable forimproved stability. While the eye seal principal stiffness is negative


29

Bearing Q Q c Eye Seal Nd LogClearance (lbf/in) Description Inlet Swirl (cpm) Dec, δMinimum 0 Base - - 3,545 0.26Minimum 57,250 Qa, Alford 3.0 - 3,590 -0.40Minimum 15,700 Laby Analysis - 0.60 Estimated 3,551 0.08Minimum 7,600 Laby Analysis - 0.52 Calculated 3,547 0.17Minimum 22,500 Q0 (Q for Log Dec = 0) - - 3,558 0.00Maximum 0 Base - - 3,361 0.40Maximum 57,250 Qa, Alford 3.0 - 3,321 -0.23Maximum 15,700 Laby Analysis - 0.60 Estimated 3,333 0.28Maximum 7,600 Laby Analysis - 0.52 Calculated 3,348 0.35Maximum 40,000 Q0 (Q for Log Dec = 0) - - 3,318 0.00

β

CalculatedSwirl = 0.52

Min Laby Clearance

EstimatedSwirl = 0.6

δ

CalculatedSwirl = .52

Balance Piston Honeycomb Seal Included

Min Bearing Clearance

EstimatedSwirl = .6

Min Laby Clearancew/ Eye Seal

Swirl Brakes

Laby Inlet Swirl = 0.6

Max

Nom

Min

Min Operating

δ

Laby QSwirl = 0.6

Laby QSwirl = 0.52

Min Laby Clearance


Q0

Alford’s Qa

δa = - 0.40

Level II Analysis Required

δa < 0.1

δ

Laby QSwirl = 0.6

Laby QSwirl = 0.52

Min Laby Clearance


Q0

Alford’s Qa

δ a = - 0.23

Level II Analysis Requiredδ a < 0.1

Figure 14. Hydrogen Compressor Stability Versus Laby Eye SealClearance.

Figure 15. Hydrogen Compressor Stability—Effect of ClearanceProfile.

for a diverging clearance, this effect is minimal. Also, the leakageactually decreases for the diverging clearance case compared to aconstant clearance. For this compressor, the impeller growth withspeed increases from the eye seal discharge location (impeller eye)to the eye seal inlet location resulting in a divergent seal clearance.

Balance Piston Honeycomb Seal—Cell Clogging and Inlet Swirl Effects

One problem with honeycomb seals is that the honeycomb cellsmay clog or fill with debris. Other nonlabyrinth type seals are alsoknown to clog, such as hole pattern seals. Hole pattern seal testresults with plugged holes may be found in Moore and Soulas(2003).

This clogged cell effect for honeycomb seals is illustrated inFigure 16 for minimum bearing clearance and three differenthoneycomb seal inlet swirl values: 0.8, a pessimistic value; 0.6, amore realistic value; and 0.3, simulating the inclusion of a swirlbrake. The plot shows that as the honeycomb cells fill or clog,stability decreases. For the 0.6 inlet swirl case, as long as there areless than 48 percent of the cells filled, the log dec is greater than0.1, the API acceptance value. For inclusion of a swirl brake (0.3inlet swirl), the log dec is greater than 0.1 as long as there are lessthan 60 percent filled cells.

EXAMPLE #2—65,000 HP, 3 STAGE PROPANEREFRIGERATION COMPRESSOR

The second example is a three stage, three section 65,000 hppropane compressor in a refrigeration service with a 230 inch

Figure 16. Hydrogen Compressor Stability Versus Percent FilledBalance Piston Honeycomb Cells.

bearing span and a Lb/Dms = 10.3. The rotor weighs in excess of36,000 lbm and operates at 3000 rpm with dry gas seals and fivepad tilting pad journal bearings (Table 1). The inlet pressure is 20psi with a 100 psi discharge pressure and a gas mole weight of 44.Low base log dec and relatively small destabilizing forces charac-terize large refrigeration compressors. This machine is noexception. The base log dec (rotor and bearings only, Q = 0.0) ofthis compressor ranges from 0.18 to 0.13 for the range of bearingtolerances.

In the past, questions have arisen concerning the applicability of“Wachel” type equations (Wachel and von Nimitz, 1981) for com-pressors with multiple process sections. The concerns have beenaddressed, for the most part, by applying the modified Alford’sequation on a stage by stage basis. In this application, each stagerepresents a process section with side streams added to the mainflow prior to the second and third impellers. Figure 17 presents therotordynamic model of the compressor.

Figure 17. Propane Compressor Cross Section.

Applying Equation (3) to this service, the anticipated destabiliz-ing force, Qa, can be calculated. On a per wheel basis, the totalanticipated destabilizing force is found to be Qa = 23,046 lbf/in.Applying this to the rotor center yields a log dec of 0.15 and 0.09for minimum and maximum stiffness bearings, respectively. Thestability sensitivity plot is shown on Figure 18. (The tolerancerange for the bearing clearances and oil inlet temperatures definesthe range of bearing stiffness.) As required by API since the worstcase δa<0.1, a Level II analysis was performed using the methoddeveloped by Kirk (1990) to predict the behavior of the impellerlabyrinth seals and balance piston. These destabilizing forces areapplied at the physical location of the seal. For the same range ofbearing stiffness, the log dec is calculated for the following condi-tions:

• Rotor and bearing only

• Rotor, bearing, and impeller labyrinth seals

• Rotor, bearing, impeller labyrinth seals, and balance piston

Table 3 contains the results of the Level II analysis and the esti-mation of rotor stability using the modified Alford’s force. As


δBalance Piston Honeycomb Seal Included

Laby Inlet Swirl = 0.6

Max

Min

Nom

Min Bearing Clearance

Min Operating

δ


Laby Inlet Swirl = 0.6Min Bearing Clearance

Min Operating Eye Seal Cs = 6.0 mils dia.

Diverging Cs

Converging Cs

Constant Cs

δ

Figure 18. Propane Compressor Stability.

expected, the excitation force due to the labyrinth seals (includingthe balance piston) is small and has a minor impact on the rotorstability. For this application, the predicted stability levels usingthe modified Alford’s force are conservative.

Table 3. Propane Refrigeration, Injection #1, and Injection #2Compressor Stability Summary.

EXAMPLE #3—30,000 HP, 4 STAGE INJECTION COMPRESSOR

The third example is a four stage 30,000 hp centrifugal compres-sor in an injection service with a 59 inch bearing span and a Lb/Dms= 8.86. The rotor weighs 540 lbm and operates at 12,700 rpm withdry gas seals and four pad tilting pad journal bearings (Table 1). Theinlet pressure is 1200 psi with a 3300 psi discharge pressure onnatural gas. In terms of injection service, this compressor would beconsidered near the lower end of the discharge pressure range.However, the high horsepower per rotor weight would place it nearthe top of that range. Recognizing this fact, the manufacturer conser-vatively designed the compressor with a larger central shaft section(Figure 19). The stiffer shaft produces bending modes with higherrelative bearing motion as compared to the shaft center for the firstmode. This permits the bearing damping to be more effective in con-trolling the shaft center and results in higher log dec values.

Figure 19. Injection #1 Compressor Cross Section.

As in the prior examples, the Level I analysis is comparedagainst the Level II analysis to determine conservatism. (Note: ALevel II analysis is required since Q0<2*Qa.) The modifiedAlford’s force is calculated to be 68,015 lbf/in reflecting the highhorsepower of the application (Figure 20). Table 3 contains theresults of the stability analysis for the same conditions as inexample 2. For applications in the midrange of pressure, thebalance piston effect on rotor stability is typically equivalent to theimpeller eye seals if both are labyrinth type with no antiswirlfeatures. This can be seen for this compressor as the decrease in logdec produced by the impeller labyrinth seals is nearly equal to thatproduced by including the balance piston. As before, the stabilitylevel predicted using the modified Alford’s force is conservative inrelation to the Level II analysis performed.

Figure 20. Injection #1 Compressor Stability.

EXAMPLE #4—11,000 HP, 4 STAGE INJECTION COMPRESSOR

The fourth example is a four stage 11,000 hp centrifugal com-pressor in an injection service with a 48 inch bearing span and aLb/Dms = 7.26. The rotor weighs 422 lbm and operates at 9900 rpmwith dry gas seals and five pad tilting pad journal bearings (Table1). The inlet pressure is 5600 psi with a 9000 psi dischargepressure on natural gas. In terms of injection service, this com-pressor would be considered near the higher end of the dischargepressure range and in the midrange of horsepower per weight ratio.In this application, destabilizing forces are expected to be higherdue to the elevated gas densities in the compressor. In fact, a LevelII analysis is required due to the average gas density of 115 kg/m3.As before, the manufacturer conservatively designed the compres-sor with a larger central shaft section to counter the expectedhigher destabilizing forces (Figure 21).

Figure 21. Injection #2 Compressor Cross Section.

For this service, the modified Alford’s force is calculated to be72,854 lbf/in, roughly equal to the compressor in example 3(Figure 22). Higher gas densities offset the higher horsepower ofthe previous example. Thus, the anticipated level of destabilizingforce is roughly the same for the two examples. Table 3 containsthe results of the stability analysis for Level I and Level II


31

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

0 20 40 60 80 100

δ

Minimum Bearing Stiffness

Maximum Bearing Stiffness

Alford’s Qa = 23,046 lbf/in

Q0 = 70,000 lbf/in

Q0 = 103,000 lbf/in

Level II Analysis Requiredδa < 0.1

δa = 0.09

-0.1

0.1

0.3

0.5

0.7

0.9

1.1

1.3

1.5

1.7

0 20 40 60 80 100

δ

Minimum BearingStiffness



Q0 = 95,300 lbf/in

Level II Analysis RequiredQ0 < 2*Qa

δa = 0.24

Example #2 Example #3 Example #4

ConfigurationBearingStiffness

PropaneRefrigeration Injection #1 Injection #2

Rotor / Bearing(Base log dec) Minimum 0.18 1.55 0.84

Maximum 0.13 0.86 0.39+ Alford’s Qa

Minimum 0.15 0.82 0.52Maximum 0.09 0.24 0.10

+ Laby SealsMinimum 0.17 1.25 0.72Maximum 0.12 0.65 0.30

+Laby Seals &Balance Piston Minimum 0.17 1.00 0.38

Maximum 0.12 0.48 0.12

Predicted Log Dec, δ

analyses. As with the other injection compressor, a Level IIanalysis was required since Q0<2*Qa, indicating that an insuffi-cient safety margin exists between the anticipated destabilizingforce and the amount needed to drive the system unstable.

Figure 22. Injection #2 Compressor Stability.

As noted, the Alford’s force approximation of the destabilizingforce was nearly equal for the two injection compressors. Table 3confirms this fact when comparing both Level II analyses. Thechange in log dec from the base value to the condition including alldestabilizing forces is roughly �0.55 and �0.40 for the firstinjection compressor and �0.45 and �0.30 for the secondinjection compressor for the range of bearing coefficients. This isconsidered close given the difference in rotor and bearinggeometry. Unlike the previous example, the balance piston isproducing the majority of the destabilizing force. This was recog-nized in the early design stages and a shunted balance piston(Kanki, et al., 1988) was employed in the final configuration.

Finally, the Alford’s force is shown to be slightly conservativefor one bearing condition and somewhat optimistic at the other.However, the conservatism is still seen in the worst case predic-tion, which would invoke a Level II analysis. Additionally, a LevelII analysis was required due to the average gas density in the com-pressor. This compressor successfully passed a full load testwithout stability problems.

DISTRIBUTED VERSUS LUMPED ANALYSISAND EXAMPLE #5—MIXED REFRIGERATION

In the previous three examples, the conservatism of the modifiedAlford’s equation was determined by comparing the resulting logdec values from the Level I and Level II analyses. In this section,a more direct method of comparing an equivalent lumped destabi-lizing force is presented similar to Memmott (2000). The example3 injection compressor is used along with a larger refrigerationcompressor. The 93,000 hp mixed refrigeration compressor(example 5) has a 222 inch bearing span and a Lb/Dms = 10.2. Thetwo compressors are compared on Figure 23. The rotor weighs44,730 lbm and operates at 3000 rpm with dry gas seals and fivepad tilting pad journal bearings (Table 1). The inlet pressure is 60psi with a 320 psi discharge pressure using a 25 MW gas.

Figure 23. Mixed Refrigerant Compressor Cross Section.

To produce an equivalent destabilizing force as calculated by themodified Alford’s equation, an equivalent cross-coupled stiffnessis calculated for each seal using the following relation:

(4)

The k and C values are determined from a labyrinth seal analysis(Kirk, 1990).

The equivalent modal cross-coupling at the rotor center isdefined as:

(5)

The modal influence factor, Mf, is determined from the normalizedmode shape of the first damped natural frequency and representsthe displacement at the seal location. The reduced cross-couplingforces are summed for all seal locations. This modal cross-coupling, Qm, along with the log dec value calculated in the LevelII analysis is plotted on the stability sensitivity plots for the twocompressor examples at the maximum bearing stiffness only.

Figure 24 presents the results for the first injection compressor,example 3. The minimum modal factor for this compressor wasonly 0.91 reflecting the stiffer shaft operation of the compressor.Given the labyrinth seal and balance piston forces, the modalcross-coupling was calculated to be 44,933 lbf/in. From Table 3,the Level II log dec including all destabilizing forces was 0.48 atthe maximum bearing stiffness. Plotting this point on the sensitiv-ity chart, one finds that the point lies almost directly on the linederived by placing a varying amount of cross-coupling at the rotorcenter. It needs to be emphasized that the log dec plotted, δm, wascalculated from the Level II analysis with the seal effects locatedat the physical location of the seal. From this one can conclude thefollowing:

• The modal reduction produces a reduced cross-coupling forcedirectly comparable to the modified Alford’s force.

• The modal cross-coupling also provides an indication of howmuch margin the rotor has based on the actual seal coefficients. (Inthis case, the 44,933 lbf/in is compared to the Q0 amount of 95,300lbf/in. A safety margin of roughly two exists or more simply thedestabilizing effect of the labyrinth seals could be two times largerthan calculated before an unstable condition is predicted.)

Figure 24. Injection #1 Compressor Stability with Modal Q.

The calculation is repeated for the mixed refrigeration compres-sor (example 5). For this compressor, the minimum modal factorwas 0.56 representing the more flexible bending mode of the shaft.Two configurations of seals are included in the Level II analysis,one with a shunted balance piston and one without a shunt. Bothmodal cross-couplings and the final Level II log dec values areplotted (Figure 25). As with the injection compressor, both points


-0.1

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 20 40 60 80 100

δ




Q0 = 100,200 lbf/in


δa = 0.103

-0.1

0.1

0.3

0.5

0.7

0.9

1.1

1.3

1.5

1.7

0 20 40 60 80 100

δ




Q0 = 95,300 lbf/in

Modal δm = 0.48

Modal Qm = 44,933 lbf/in

Injection & MR drawn to scale.

Q k Ceq cs= − ω

Q M Qm f eq= ∑

lie closely to the sensitivity line from the Level I analysis. This istrue even in the case of the shunted balance piston where the netcross-coupling term is negative (or stabilizing).

Figure 25. Mixed Refrigerant Compressor Stability with Modal Q.

CONCLUSIONS

The API Level I modified Alford’s cross-coupling force calcula-tion was examined and determined to be indeed a conservativeestimation of the compressors’ destabilizing forces. Several indus-trial applications were examined including a hydrogen compressor,two high-pressure injection compressors, one with a high horse-power to weight ratio, and two large refrigeration compressors,including a multisection configuration. For all cases, the modifiedAlford’s force was determined to produce the worst case stabilitylevel.

Also, specifics of the Level II analysis were examined todetermine what parameters are key to determining centrifugalcompressor stability. Some well-known influences includingbearing clearance tolerance and labyrinth seal inlet swirl wereshown to have a major impact on the stability level of the 10 stagehydrogen compressor of example 1. Not so well known was theimpact of the labyrinth eye clearance profile. Varying the toothclearance slope from 8 mils diametral convergent to 8 milsdiametral divergent at the impeller eye seals only, changed thepredicted log dec from �0.11 to 0.23 for a constant inlet swirl of0.6. This provides another simple tool to increase the stability ofmarginal centrifugal compressors.

Finally, a modal approach was presented to permit direct com-parison of the Level I and Level II destabilizing forces. Beyondconfirming the conservatism of the modified Alford’s force, themodal approach also permits use of the stability sensitivity plot toapproximate the safety margin of the labyrinth seal coefficientsagainst a zero log dec threshold.

NOMENCLATURE

Cs = Seal diametral clearance (mils)C = Principle damping (lbf-sec/in)Dc = Impeller diameter (inch)Dms = Midshaft diameter (inch)Hc = Minimum width of the impeller or discharge volute (inch)hp = Horsepower (hp)k = Cross-coupled stiffness (lbf/in)Lb = Bearing span, (inch)Mf = Modal influence factorN = Speed (rpm)Nd = Damped natural frequency (cpm)Pdisch = Discharge pressure (psi)Pin = Inlet pressure (psi)Q = Aerodynamic cross-coupling (lbf/in)Qa = Anticipated aerodynamic cross-coupling (lbf/in)

Qeq = Equivalent cross-coupling (lbf/in)Qm = Modal cross-coupling (lbf/in)Q0 = Aerodynamic cross-coupling for zero log dec (lbf/in)s = Real part of eigenvalueX1,2 = Amplitude (mils)β = Efficiency factorδ = Log decδa = Log dec for Qaδm = Final log dec from the Level II analysisρratio = Density ratioωd = Damped natural frequency (rad/sec)ωcs = Damped first natural frequency (rad/sec)

REFERENCES


API Technical Publication 684, 2005, “Tutorial on Rotordynamics:Lateral Critical Speeds, Unbalance Response, Stability, TrainTorsional and Rotor Balancing,” Second Edition, AmericanPetroleum Institute, Washington, D.C.

Alford, J. S., 1965, “Protecting Turbomachinery from Self-ExcitedRotor Whirl,” ASME Journal of Engineering for Power, 87,(4), pp. 333-344.

Bansal, P. N. and Kirk, R. G., 1975, “Stability and Damped CriticalSpeeds of Rotor-Bearing Systems,” ASME Journal ofEngineering for Industry, Series B, 97, (4), pp. 1325-1332.

Barrett, L. E., Gunter, E. J., and Allaire, P. E., 1976, “The Stabilityof Rotor-Bearing Systems Using Linear Transfer Functions—A Manual for Computer Program ROTSTB,” Report No.UVA/643092/MAE81/124, School of Engineering and AppliedScience, University of Virginia, Charlottesville, Virginia.

Benckert, H. and Wachter, J., 1979, “Investigations on the MassFlow and the Flow Induced Forces in Contactless Seals ofTurbomachines,” Proceedings of the Sixth Conference onFluid Machinery, Scientific Society of Mechanical Engineers,Akadémiaki Kikado, Budapest, pp. 57-66.

Benckert, H. and Wachter, J., 1980, “Flow Induced SpringCoefficients of Labyrinth Seals for Application toRotordynamics,” NASA CP-2133, pp. 189-212.

Chen, W. J., 1996, “Instability Threshold and Stability Boundariesof Rotor-Bearing Systems,” ASME Journal of Engineering forGas Turbines and Power, 118, pp. 115-121.

Childs, D. W. and Scharrer, J. K., 1986a, “ExperimentalRotordynamic Coefficient Results for Teeth-on-Rotor andTeeth-on-Stator Labyrinth Gas Seals, ASME 86-GT-12.

Childs, D. W. and Scharrer, J. K., 1986b, “An Iwatsubo-BasedSolution for Labyrinth Seals: Comparison to ExperimentalResults,” ASME Journal of Engineering for Gas Turbines andPower, 108, (2), pp. 325-331.

Childs, D. W., and Ramsey, C., 1991, “Seal Rotordynamic-Coefficient Test Results for a Model SSME ATD-HPFTPTurbine Interstage Seal With and Without a Swirl Brake,”ASME Journal of Tribology, 113, pp. 113-203.

Childs, D. W., 1993, Turbomachinery Rotordynamics: Phenomena,Modeling, and Analysis, New York, New York: John Wiley &Sons, Inc.

Edney, S. L., Fox, C. H. J., and Williams, E. J., 1990, “TaperedTimoshenko Finite Elements for Rotor Dynamics Analysis,”Journal of Sound and Vibration, 137, (3).

Elrod, D. A., Pelletti, J. M., and Childs, D. W., 1995, “TheoryVersus Experiment for the Rotordynamic Coefficients of an


33

-0.05

0.00

0.05

0.10

0.15

0.20

-50 -25 0 25 50 75 100

δ



Q0 = 69,100 lbf/in

δm = 0.106

Qm = -12,403 lbf/in Qm = 23,300 lbf/in

δm = 0.08

Shunted BP

Non-Shunted BP


Interlocking Labyrinth Gas Seal,” ASME Paper 95-GT-432,Presented at the International Gas Turbine and AeroengineCongress and Exposition, Houston, Texas.

Fowlie, D. W. and Miles, D. D., 1975, “Vibration Problems withHigh Pressure Centrifugal Compressors,” ASME 75-Pet-28.

Geary, C. H., Damratowski, L. P., and Seyer, C., 1976, “Design andOperation of the World’s Highest Pressure Gas InjectionCentrifugal Compressor,” Paper Number OTC 2485, Presentedat the Eighth Annual Offshore Technology Conference,Houston, Texas.

Gunter, E. J., 1966, “Dynamic Stability of Rotor-BearingSystems,” NASA SP-113.

Gunter, E. J., 1967, “The Influence of Internal Friction on theStability of High Speed Rotors,” ASME Journal ofEngineering for Industry, Series B, 89, (4), pp. 683-688.

Iwatsubo, T., Matooka, N., and Kawai, R., 1982, “Spring andDamping Coefficients of the Labyrinth Seal,” NASA CP-2250,pp. 205-222.

Kanki, H., Katayama, K., Morii, S., Mouri, Y., Umemura, S.,Ozawa, U., and Oda, T., 1988, “High Stability Design for NewCentrifugal Compressor,” Rotordynamic Instability Problemsin High-Performance Turbomachinery, NASA CP-3026, pp.445-459.

Kirk, R. G., 1980, “Stability and Damped Critical Speeds: How toCalculate and Interpret the Results,” CAGI Technical Digest,12, (2).

Kirk, R. G., 1988a, “Evaluation of Aerodynamic InstabilityMechanisms for Centrifugal Compressors—Part I: CurrentTheory,” ASME Journal of Vibration, Acoustics, Stress, andReliability in Design, 110, (2), pp. 201-206.

Kirk, R. G., 1988b, “Evaluation of Aerodynamic InstabilityMechanisms for Centrifugal Compressors—Part II: AdvancedAnalysis,” ASME Journal of Vibration, Acoustics, Stress, andReliability in Design, 110, (2), pp. 207-212.

Kirk, R. G., 1990, “A Method for Calculating Labyrinth Seal InletSwirl Velocity,” ASME Journal of Vibration and Acoustics,112, (3), pp. 380-383.

Kwanka, K., Ortinger, W., and Steckel, J., 1993, “Calculation andMeasurement of the Influence of Flow Parameters onRotordynamic Coefficients in Labyrinth Seals,” RotordynamicInstability Problems in High-Performance Turbomachinery,NASA CP-3239, pp. 209-218.

Lund, J. W., 1965, “Rotor Bearing Dynamics Design Technology,Part V,” AFAPL-TR-65-45, Aero Propulsion Laboratory,Wright-Patterson Air Force Base, Dayton, Ohio.

Lund, J. W., 1974, “Stability and Damped Critical Speeds of aFlexible Rotor in Fluid Film Bearings,” ASME Journal ofEngineering for Industry, 96, (2), pp. 509-517.

Memmott, E. A., 2000, “Empirical Estimation of a Load RelatedCross-Coupled Stiffness and the Lateral Stability ofCentrifugal Compressors,” Presented at the 18th MachineryDynamics Seminar, Canadian Machinery VibrationAssociation, Halifax, Nova Scotia, Canada.

Moore, J. J. and Hill, D. L., 2000, “Design of Swirl Brakes forHigh Pressure Centrifugal Compressors Using CFDTechniques,” Proceedings of the Eighth InternationalSymposium of Transport Phenomena and Dynamics ofRotating Machinery, (ISROMAC-8), Honolulu, Hawaii, pp.1124-1132.

Moore, J. J. and Soulas, T., 2003, “Damper Seal Comparison in aHigh-Pressure Re-Injection Centrifugal Compressor DuringFull Load, Full-Pressure Factory Testing Using DirectRotordynamic Stability Measurement,” Proceedings of DETC‘03, ASME Design Engineering Technical Conferences andComputers and Information in Engineering Conference,Chicago, Illinois.

Murphy, B. T. and Vance, J. M., 1983, “An Improved Method forCalculating Critical Speeds and Rotordynamic Stability ofTurbomachinery,” ASME Journal of Engineering for Power,105, (3), pp. 591-595.

Nelson, H. D. and McVaugh, J. M., 1975, “The Dynamics ofRotor-Bearing Systems Using Finite Elements,” ASMEJournal of Engineering for Industry, 98, (2), pp. 593-600.

Ramesh, K. and Kirk, R. G., 1993, “Stability and Response ofRotors Supported on Active Magnetic Bearings,” Proceedingsof the 14th ASME Vibrations and Noise Conference, DE- 60,Vibration of Rotating Systems, pp. 289-296.

Rouch, K. E. and Kao, J. S., 1979, “A Tapered Beam FiniteElement for Rotor Dynamics Analysis,” ASME Journal ofSound and Vibration, 66, pp. 119-140.

Ruhl, R. L. and Booker, J. F., 1972, “A Finite Element Model forDistributed Parameter Turborotor Systems,” ASME Journal ofEngineering for Industry, Series B, 94, (1), pp. 126-132.

Scharrer, J. K. and Pelletti J. M., 1994, “Commercial Applicationsof Space Propulsion Turbomachinery ComponentTechnology,” SAE Paper Number 941197, Presented at theSAE International Aerospace Atlantic Conference andExposition, Dayton, Ohio.

Smith, K. J., 1975, “An Operational History of FractionalFrequency Whirl,” Proceedings of the Fourth TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas, pp. 115-125.

Thieleke, G. and Stetter, H., 1990, “Experimental Investigations ofExciting Forces Caused by Flow in Labyrinth Seals,”Rotordynamic Instability Problems in High-PerformanceTurbomachinery, NASA CP-3122, pp. 109-134.

Wagner, N. G. and Steff, K., 1996, “Dynamic LabyrinthCoefficients from a High-Pressure Full-Scale Test Rig UsingMagnetic Bearings,” Rotordynamic Instability Problems inHigh-Performance Turbomachinery, NASA CP-3344, pp. 95-111.

Wachel, J. C. and von Nimitz, W. W., 1981, “Ensuring theReliability of Offshore Gas Compression Systems,” Journal ofPetroleum Technology, pp. 2252-2260.


Phillip Dowson is General Manager,Materials Engineering, with ElliottCompany, in Jeannette, Pennsylvania, andhas 34 years of experience in the turboma-chinery industry. He is responsible for themetallurgical and welding engineering forthe various Elliott product lines within thecompany. He is the author/coauthor of anumber of technical articles related totopics such as abradable seals, high tem-perature corrosion, fracture mechanics,

and welding/brazing of impellers.Mr. Dowson graduated from Newcastle Polytechnic in

Metallurgy and did his postgraduate work (M.S. degree) inWelding Engineering. He is a member of ASM, ASTM, and TWI.

Wenchao Wang is a Senior MaterialsEngineer with Elliott Company, inJeannette, Pennsylvania. He joined thecompany in 1997, and has been involved inmaterials’ related R&D projects, produc-tion supports, and aftermarket services. Heis experienced in turbomachinery failureanalysis, protective/functional coatings,and remaining life assessment.

Dr. Wang received a B.S. degree(Metallurgical Engineering, 1984) from the

University of Science & Technology Beijing, and M.S. (1994) andPh.D. (1997) degrees (both in Materials Science & Engineering)from the University of Cincinnati. He is a member of ASM.

Agustin Alija is a Maintenance Engineerof the Hydrocarbon and Energy Business ofPBBPolisur (a Dow Chemical ownedcompany), in Bahia Blanca, Argentina. Hejoined the company in June 2003.

Mr. Alija has a B.S. degree (Mechanicaland Aeronautical Engineering, 2000) fromthe National University of Cordoba inArgentina.

ABSTRACT

In today’s marketplace, a large percentage of oil refinery, petro-chemical, and power generation plants throughout the world havebeen trying to reduce their operation cost by extending the servicelife of their critical machines, such as steam turbines and hot gasexpanders, beyond the design life criteria. The key ingredient inplant life extension is remaining life assessment technology. Thispaper will outline the remaining life assessment procedures, andreview the various damage mechanisms such as creep, fatigue,creep-fatigue, and various embrittlement mechanisms that canoccur in these machines. Also highlighted will be the varioustesting methods for determining remaining life or life extension ofcomponents such as high precision stress relaxation (STR) test,which determines creep strength, and constant displacement rate(CDR) test, which evaluates fracture resistance. Other tests such asreplication/microstructure analysis and toughness tests will also bereviewed for calculating the remaining life or life extension of thecomponents. Use of the latest computer software will also be high-lighted showing how creep-life, fatigue-life, and creep/fatigue-lifecalculations can be performed. Also shown will be actual lifeextension examples of steam turbines and hot gas expander com-ponents performed in the field.

INTRODUCTION

In recent years, from oil refinery to petrochemical and powergeneration industries, more and more plants throughout the worldare facing a common issue—aging turbines, usually over 30 yearsold. Questions bearing in managers’ minds are what is the machinecondition and whether they can be continually operated (if yes,how long). The answer is significant not only for safety concernsbut also for cost reduction, especially with today’s limited budgets.Therefore, there is an increasingly strong desire for the engineer-ing aftermarket service to perform “remaining life assessment” ofsteam turbines and hot gas expanders.

Remaining life assessment is to use metallurgical and fracturemechanics methodologies to predict the remaining life of struc-tures and components that have been in service for an extendedperiod of time, usually close to or beyond the designed life.Traditionally, if parts are found with material degradations ordamages during an overhaul, they might be scrapped and replacedfor risk-free consideration; even though they might have someuseful life. Remaining life assessment offers a possible tool toestimate the useful remaining lifetime and avoid prematurescrapping of the parts. So remaining life assessment is considered

77

REMAINING LIFE ASSESSMENT OFSTEAM TURBINE AND HOT GAS EXPANDER COMPONENTS

byPhillip Dowson

General Manager, Materials Engineering

Wenchao WangSenior Materials Engineer

Elliott Company


andAgustin Alija

Maintenance Engineer

Dow Chemical PBBPolisur

Bahia Blanca, Argentina

to be an attractive method/process for cost reduction and reductiondowntime.

Remaining life assessment has often been improperly referred toas “life extension.” Actually this analysis cannot extend thelifetime of the components. It can only assess the useful remaininglifetime, based on the metallurgical examinations and theoretical(fracture mechanics) calculations. If such assessments indicate theneed for extensive replacements and refurbishments, life extensionmay not prove to be a viable option. Above and beyond thisobjective, remaining life assessment technology serves many otherpurposes. It helps in setting up proper inspection schedules, main-tenance procedures, and operating procedures. It should, therefore,be recognized at the outset that development of techniques forremaining life assessment is more enduring in value and broader inpurpose than simply the extension of plant life. For instance, it hasbeen possible to extend the inspections from six to 10 years formodern rotors, on the basis of assessments based on fracturemechanics, resulting in considerable savings.

In implementing remaining life assessment procedures, theappropriate failure definition applicable to a given situation mustbe determined at the outset, and the purpose for which the assess-ment is being carried out must be kept in mind. While determiningthe feasibility of extended plant life may be one objective, a morecommon objective is the setting of appropriate intervals for inspec-tion, repair, and maintenance. In this context, remaining lifeassessment procedures are used only to ascertain that failures willnot occur between such intervals. It should never be assumed thathaving performed a remaining life assessment study for a 20-yearlife extension, one could then wait for 20 years without interimmonitoring. Periodic checks to ensure the validity of the initialapproach are essential. In this sense, remaining life assessmentshould be viewed as an ongoing task, rather than a one-timeactivity.

A phased approach, in which the initial level includes nonincur-sive techniques followed by other levels of actual plantmonitoring, then followed by nondestructive inspections anddestructive tests would be the most logical and cost-effectiveapproach. In Level I, assessments are performed using plantrecords, design stresses and temperatures, and minimum values ofmaterial properties from the original equipment manufacturer(OEM). Level II involves actual measurements of dimensions,temperatures, simplified stress calculations, and inspectionscoupled with the use of minimum material properties from theOEM. Level III involves indepth inspection, stress levels, plantmonitoring, and generation of actual material data from samplesremoved from the component (destructive testing). The degree ofthe detail and accuracy of the results increases from Level I toLevel III, but at the same time, the cost of the assessment alsoincreases. Depending on the extent of the information availableand the results obtained, the analysis may stop at any level orproceed to the next level as necessary.

In evaluating the failure criteria or remaining life, one needs tounderstand the various failure mechanisms that can occur. In tur-bomachinery components, the failure criteria can be governed byone or a combination of the following failure mechanisms:

• Fatigue—high cycle or low cycle

• Corrosion/corrosion fatigue

• Stress corrosion cracking (SCC)

• Erosion—solid particle or liquid impingement

• Erosion corrosion

• Embrittlement

• Creep rupture/creep fatigue

• High temperature corrosion/embrittlement

• Mechanical (foreign objective) damage

However, in remaining life assessment, usually only thosemechanisms depending on temperature and time are taken intoaccount. For example, for turbine casing, engineers usually focuson thermal stress-induced low cycle fatigue, creep rupture, andtempering embrittlement cracking. These failures usually are slowprocesses, therefore, they can be assessed and forecasted byexamining the warning evidences in the material.

Countless works have been done to study the behaviors offatigue crack initiation/propagation and creep or embrittlementrupture in steels and alloys. Scientists and engineers have reachedsuch a level that, by knowing the flaw size or microstructure dete-rioration/damage, one can theoretically calculate and predict theremaining lifetime of the parts, based on the knowledge of thematerial properties and understanding of the stress distributions.

FATIGUE

Failures that occur under cyclic loading are termed fatiguefailures. These can be vibration stresses on blades, alternatingbending loads on shafts, fluctuating thermal stresses during start-stop cycles, etc. There are two types of fatigue: low cycle fatigue(LCF), high cycle fatigue (HCF). Traditionally, low cycle fatiguefailure is classified occurring below 104 cycles, and high cyclefatigue is above that number. An important distinction betweenHCF and LCF is that in HCF most of the fatigue life is spent incrack initiation, whereas in LCF most of the life is spent in crackpropagation because cracks are found to initiate within three to 10percent of the fatigue life. HCF is usually associated with lowerstress, while LCF usually occurs under higher stress.

Remaining life of casings or rotating components is generallybased upon crack growth consideration. Fracture mechanics is themathematical tool that is employed. It provides the concepts andequations used to determine how cracks grow and their effect onthe strength of the structure. At the authors’ company fracturemechanics is utilized in analyzing the structural integrity of com-ponents that have been in operation to determine whether thecomponent is suitable for further operation. Based upon crackgrowth analysis one considers a number of scenarios.

From an initial defect size ao one must determine critical flawsize ac for fast fracture.

(1)

where:KI = Applied tensile mode I stress intensity factorKIC = Plane strain fracture toughness of material

For LCF, determine how many cycles for ao to grow to ac. ForHCF, one must prevent crack growth. Consequently for HCF, ΔKI< ΔKth, where:

ΔKI = Stress intensity factor range ksi inch

ΔKth = Threshold stress intensity factor range below which fatigue crack growth (or corrosion fatigue crack growth)does not occur

Further discussion of fracture mechanic concepts can be foundelsewhere (Dowson, 1995, 1994).

CREEP RUPTURE AND STRESS RUPTURE

Evidence of creep damage in the high temperature regions ofblade attachment areas of rotors has been observed in someinstances (Bush, 1982). The rim stresses and metal temperature atthese locations are assessed against the creep rupture data for thatparticular grade of steel/or material. Traditionally one has used aLarson-Miller (LM) plot of the type shown in Figure 1.

The degree of safety margin depends on the user and what lowerbound design curve is applied. Since these curves are based uponthe chemistry, variation in chemistry for a particular grade canhave an effect on the Larson-Miller curve. Also, Larson-Miller


a a

K Ko c

I IC

→→

Figure 1. Larson-Miller Curve of Cr-Mo-V Alloy Steel (ASTMA470 Class 8).

curves are generally based upon creep rupture tests done for 104 to3 � 104 hours and very few data at 105 hours. Consequently, thedata for longer hours are generally extrapolated. Since most of thecreep rupture data is done with smooth bar specimens, the effect ofnotch ductility at long-term service has not been done. Short-termnotched bar tests may fail to predict the onset of notch sensitivity.Notch sensitively is not an inherent property but depends on thetemperature, stress, stress state, and strain rate.

In assessing remaining life of the components due to creep, suchas blade attachments, crack initiation is used as the criterion.However, with the emergence of cleaner steel and fracturemechanics and an increasing need to extend the life of acomponent, application of crack growth techniques have becomecommon in the past decade.

For crack initiation as the fracture criterion, history-based cal-culation methods are often used to estimate life.

Methods for Crack Initiation Due to Creep

For the analytical method, one must have accurate operatinghistory of the components, which may consist of temperature,applied loads, changes in operation, such as shut downs orvariation in speed or pressure. A simplistic estimation of the creeplife expended can be made by assessing the relaxed long-term borestresses and rim stresses against the standard rupture data using thelife fraction rule. The life fraction rule (LFR) states that at failure:

(2)

where ti is the time spent at a given stress and temperature and triis the rupture life for the same test conditions.

• Example:The purpose of this example is to illustrate the use of the life-

fraction rule. A steam turbine piping system, made of1.25Cr-0.5Mo steel designed for a hoop stress of 7 ksi, wasoperated at 1000°F (538°C) for 42,500 hours and at 1025°F(552°C) for the next 42,500 hours. Calculate the life fractionexpended using the life fraction rule. From the Larson-Millerparameter curve of the steel, it is found that, at σ = 7 ksi,

tri at 1000°F = 220,000 hourstri at 1025°F = 82,380 hours

Life fraction expended, ti/tri, at 1000°F42,500

= = 0.19220,000

Life fraction expended, ti/tri, at 1025°F42,500

= = 0.51682,380

The total life fraction expended is 0.71.

This rule was found to work well for small changes in stress andtemperature especially for CrMoV rotor steel. However, for stressvariations, the actual rupture lives were lower than the predictedvalues. Consequently, the LFR is generally valid for variable-tem-perature conditions as long as changing creep mechanisms andenvironmental interaction do not interfere with test results.However, the possible effect of material ductility on the applica-bility of the LFR needs to be investigated.

Nondestructive Techniques

Conventional nondestructive evaluation (NDE) techniques failto detect incipient damage that can be a precursor to crack initia-tion and subsequent rapid failure. However, there are other NDEtechniques that have been developed for estimating the life con-sumption. These include microstructural techniques and hardnessbased techniques.

Metallographic Examination

Metallographic techniques have been developed that cancorrelate changes in the microstructure and the onset of incipientcreep damage, such as triple point cavitation at the grain bound-aries. For this technique, measurements by replication techniqueare taken on crack sensitive areas that are subjected to the highertemperatures and stresses. These areas are generally indicated byexperience and analysis of previous damages.

The creep damage measured by replication is classified into fourdamage stages:

• Isolated cavities (A)

• Oriented cavities (B)

• Macrocracks (linking of cavities) (C)

• Formation of macrocracks (D)

Figure 2 shows the location of the four stages on the creepstrain/exposure time curve (Neubauer and Wadel, 1983).

Figure 2. Replicas for Remaining Life Assessment.

In applying this approach Neubauer and Wadel (1983) classifiedthe stages into five stages, which are Undamaged, Stage A, StageB, Stage C, and Stage D. These stages corresponded roughly toexpended life fracture (t/tr) values of 0.27, 0.46, 0.65, 0.84, and 1,respectively, using the conservative lower bound curve.

REMAINING LIFE ASSESSMENT OF STEAM TURBINE AND HOT GAS EXPANDER COMPONENTS 79

1

10

100

25 27 29 31 33 35 37 39 41 43 45

(T+460)(logt+20)x10^-3

Str

ess

[ksi

]

t

ti

ri=∑ 1

Consequently, the remaining life can be calculated using the rela-tionship as shown in Equation (3):

(3)

where t is the service life expended and tr is the rupture life.

For undamaged material and damaged stages A, B, and C, theremaining life was found to be approximately 2.7t, 1.17t, 0.54t,and 0.19t, respectively. Then by applying a safety factor of 3 to thecalculations, the safe reinspection intervals will become 0.9t, 0.4t,0.18t, and 0.06t, respectively. This approach has been developedand implemented in the power generation industry (Viswanathanand Gehl, 1991). It was found to give increased inspectionintervals as compared to the Neubauer and Wadel (1983) approach,as shown in Table 1.

Table 1. Suggested Reinspection Intervals for a Plant with 30 Yearsof Prior Service.

This approach has been applied by several utilities and realizedsignificant savings in inspection costs. Other investigationsindicate that there are wide variations in behavior due to differ-ences in grain size, ductility, and impurity control (Carlton, et al.,1967). For conservatism, the authors’ company adapted theNeubauer and Wadel (1983) approach and classified the five stagesas follows:

1. Undamaged material—Equipment can run and be reinspected atnext shutdown.

2. Class A—Reinspection would be three to five years.

3. Class B—Reinspection would be one and one-half to three years.

4. Class C—Replacement or repair would be needed within sixmonths.

5. Class D—Immediate replacement or repair would be required.

Hardness Measurement

The first attempt to develop hardness as a technique todetermine creep damage was by Goldhoff and Woodford (1972). Intheir study a good correlation was observed between room temper-ature hardness measured on exposed creep specimens and the postexposure rupture life (Figure 3).

If similar calibration could be established between prior creeplife expended or the remaining life fraction in the post exposure testand the hardness values for a range of CrMoV steels, this methodcould be applied to estimation of remaining life. However, data ofthis nature are not available in sufficient quantity. Other work doneby Viswanathan and Gehl (1992) showed a lot of promise wherethey attempted to use the hardness technique as a stress indication.They observed that the application of stress accelerated thesoftening process and shifted the hardness to lower parametervalues compared with the case of simple thermal softening on a plotof hardness versus a modified Larson-Miller parameter (Figure 4).

Destructive Techniques

Newer tests to ascertain the useful life of used and/or repairedcomponents have been utilized by the authors’ company. Design-for-performance is a recently developed methodology for evaluatingthe creep strength and fracture resistance of high temperaturematerials. Whereas the traditional approach to creep design involves

Figure 3. Correlation Between Post-Exposure Rupture Time in theStandard Test at 538°C (1000°F) and 240 MPa and RoomTemperature Hardness for Cr-Mo-V Rotor Steel.

Figure 4. Plot of Hardness Ratio Versus G Parameter for Long-Term Heating and Creep of Cr-Mo-V Rotor Steel.

long-term testing and attempts to incorporate microstructuralevolution in the test measurements, the new approach aims toexclude these changes in a short time high-precision test. The testmay also be used to evaluate consequences of such changes inservice-exposed samples. The new methodology recognizes thatseparate tests are necessary to measure creep strength and fractureresistance. For creep strength, a stress versus creep rate response isdetermined from a stress relaxation test (SRT), and for fractureresistance a constant displacement rate (CDR) test of a notchedtemple specimen is performed at a temperature where the part ismost vulnerable to fracture (Woodford, 1993).

Constant Displacement Rate Test

A description of the standardized CDR test is found elsewhere(Pope and Genyen, 1989). The data from the CDR test aretabulated in a curve similar to the load displacement curve for an


t tt

tremr= −⎛

⎝⎜⎞⎠⎟1

Inspection Interval (Years) Damage Classification Wedel-Neubauer EPRI-APTECH

Undamaged 5 27 A. Isolated Cavities 3 12 B. Oriented Cavities 1.5 5.4 C. Linked Cavities (Microcracks) 0.5 1.8 D. Macrocracks Repair Immediately Based on fracture mechanics

ordinary elevated tensile test. For a typical tensile test fracturebecomes unstable after the peak load is reached. On the other hand,in the CDR test, since the deformation is controlled at a constantrate and the notch is midway between the controlling extensome-ter, fracture rarely becomes unstable.

For a valid CDR test, the criterion for failure was considered tobe the value of “displacement at fracture,” defined as the point ofintersection of the 100 pound load line and the descending loaddisplacement curve. The “displacement at failure” is measuredfrom the start of the test to the point where the load displacementcurve decreases below 100 pounds (Figure 5).

Figure 5. Example of Load Displacement Curve from CDR Tests at1200°F and 2 mils/in/hr.

An example of how the environment can affect the notch sensi-tivity of the material is indicated by Figure 6. This exampleillustrates the effect of air exposure on IN738.

Figure 6. Constant Displacement Rate Tests Comparing CrackGrowth Resistance in Heat Treated and Oxygen EmbrittledSpecimens.

Stress Relaxation Test

Specially designed samples were tested on an electromechanicaltest system fitted with self-aligning grips, a 1500°C (2732°F) shortfurnace, and a capacitive extensometer. Details of the specimengeometry and extensometer sensitivity are provided elsewhere(Woodford, et al., 1992).

The standard test procedure involved loading the specimen at afast rate of 10 MPa/sec (1450 psi/sec) to a prescribed stress andthen switching to strain control on the specimen and monitoringthe relaxation stress. The inelastic (principally creep) strain-rate iscalculated from the following equations, Equations (4) and (5).

(4)

(5)

Where �e is the elastic strain, �.

e is the elastic strain rate, �I is theinelastic strain (principally creep strain), �

.I is the inelastic strain

rate, �t is the total strain, σ is the stress, and E is the elasticmodulus measured during loading. Using this procedure, stressversus strain-rate curves were generated covering up to five ordersof magnitude in strain-rate in a test lasting less than 5 hours.

An example of the data generated in such tests is provided inFigure 7 for Waspaloy® material. This shows stress versus predictedtime to 1 percent creep for Waspaloy®. By utilizing these data one canplot a stress versus Larson-Miller parameter for 1 percent predictedcreep of Waspaloy® compared to rupture data (Figure 8). From thedata shown on the curve, the stress relaxation test can generate creep-stress rupture data in less than a few weeks as compared to thetraditional approach that incorporates long time testing.

Figure 7. Stress Versus Predicted Times to 1 Percent Creep forStandard Waspaloy®.

Figure 8. Stress Versus Larson-Miller Parameter for 1 Percent Pre-dicted Creep of Standard Waspaloy® Compared with Rupture Data.

One major objective to this framework has been that effects ofvery long time exposures that could influence stress rupture lifewill not be accounted for. However, Woodford believes that sucheffects, i.e., precipitation of embrittling phases and grain boundarysegregation of harmful elements, are expected to influence thefracture resistance rather than creep resistance. The authors’company has utilized this methodology to generate data for hightemperature materials and weldments. Current methods are beingdeveloped for miniature specimens taken from serviced blades.From these data, it is envisioned that establishment of a set ofminimum performance criteria will enable repair/rejuvenation/replacement decisions to be made.

CREEP/FATIGUE INTERACTION

For components that operate at higher temperature where creepgrowth can occur, one must take into account the creep crackgrowth at intervals during the fatigue life of the component. Thefollowing is an example of a high temperature steam turbine rotor


ε ε εe I t Cons t+ = = tan

� �ε ε σI e

d

dt= − = −

1

Ε

that failed catastrophically at a power plant in Tennessee (Saxena,1998). The authors utilized the latest computer software to demon-strate how creep-life, fatigue life, and creep/fatigue-lifecalculations can be performed, and how inaccurate the calculationwould be without accounting for the creep life.

The power plant rotor was operated for 106,000 hours and hadincurred 105 cold starts and 183 hot starts. The material was 1CR-1Mo-0.25V forging and had been operating at a temperature of800°F. The cracks originated from several majority node set (MnS)clusters with the original flaw size of 0.254 inch � 5.51 inch and0.7 inch from the bore of the rotor (Figure 9).

Figure 9. Schematic of the Intermediate Pressure (IP) Section ofthe Rotary Showing the Size and Location of the Primary andSecondary Flaws Beneath the Seventh Row (7-R) of Blades.

Step 1—Assessment of Low Cycle Fatigue Life

The principle is that fatigue crack growth follows equations suchas the Paris law:

(6)

Cf and nf are constants that depend on the material and environ-

ment.The stress intensity factor range ΔK depends on the stress level

at the crack tip. The life assessment criterion is that critical cracksize ac is not to be exceeded. In other words,

(7)

Figure 10 and Figure 11 show the stress intensity factor calculationtogether with the crack model that was used.

Figure 10. Computer Software Module. Stress Intensity FactorCalculation.

Figure 11. Computer Software Model—Elliptical SubsurfaceCracked Plate under Membrane and Bending Stresses.

By computing the information a plot of stress intensity factorversus crack depth was done. Based on the plane strain fracturetoughness of the material, the critical crack size 0.42 inch for coldstart and 0.48 inch for hot start were determined (Figure 12).

Figure 12. Critical Crack Size Calculation—The Critical CrackSize ac is 0.42 Inch for Cold Start and 0.48 Inch for Hot Start.

Fatigue crack growth using the Paris law was computerized(Figure 13, Figure 14) and the low cycle fatigue crack growth wasdetermined for both cold and hot starts. Figure 15 shows the lowcycle fatigue crack growth, which does not compare very well withthat of the real life.

Figure 13. Fatigue Crack Growth.

The reason for the calculated life being much longer than the reallife is that the hold time effect (or creep cracking effect) is not


da

dNC Kf

n f= Δ

a ac≤

Figure 14. Fatigue Crack Growth Calculation Model.

Figure 15. Shows Low Cycle Fatigue Crack Growth for Cold andHot Starts Compared to Real Life.

taken into account. Consequently, one must run a creep-fatigueremaining life assessment. The principle is that high temperaturecrack propagation is the summation of high temperature fatigueplus primary creep plus secondary creep.

Creep-fatigue crack growth:

(8)

where h is the hold time in each cycle (which is 368 hours in thiscase). The creep crack driving force consists of two parts:

(9)

The data that are input into the code are shown in Figure 16 andFigure 17. The calculation shows that, accounting for creep effect,the creep-fatigue growth is much closer to the real life (Figure 18).

This example demonstrates that if the material/component isoperating in the creep mode, one must perform a creep-fatigueanalysis instead of fatigue only. Generally a rule of thumb is that ifonly low cycle fatigue crack growth is counted and creep is not,then the calculated lifetime is about 10 times longer than the reallife. By utilizing this software program, more accurate remaininglife assessment can be achieved for materials operating in the creepregime under cyclic loading. Also, the lesson being learned is thatthis rotor should have been examined by ultrasonic inspectionevery five years.

Figure 16. Creep/Fatigue Model.

Figure 17. Creep/Fatigue Material Data.

Figure 18. Compares Low Cycle Fatigue and Creep-Fatigue CrackGrowth to the Real Crack Growth.

EMBRITTLEMENT

Trends toward increasing size and operating stresses in compo-nents, such as large turbine-generator rotors, require higherhardenability steels with increased strength and fracture toughness.However, higher hardenability steels especially those containing


0.12

0.14

0.16

0.18

0.2

0.22

0 500 1000 1500 2000 2500

Cycles

Cra

ck D

epth

[in

ch]

Cold StartHot StartReal Life

( )[ ] ( )da

dtC C t

C

hKc t

q

Creep

f n

Fatigue

f= +� ��

Δ

( ) [ ] ( )( )[ ] ( )( )

( )( )[ ]

( )( )( )

�C t CK

n t

n p

n pC

tCt

p mp m

h

p p

imaryCreep

SecondaryCreep

=−

+

⎡

⎣

⎢⎢⎢

⎤

⎦

⎥⎥⎥

++ +

+ +⎛⎝⎜

⎞⎠⎟

++ −

− + −+

* //

*/

Pr

*2 1 1

2 21 2 1 1

11

1

1

1 1

1ν

Ε� ��

0.12

0.14

0.16

0.18

0.2

0.22

0 500 1000 1500 2000 2500Cycles

Cra

ck D

epth

[in

ch]

Low Cycle Fatigue

Real Life

Creep-Fatigue

nickel and chromium are usually much more susceptible to a phe-nomenon called temper embrittlement. The term temperembrittlement refers to a shift in the brittle-to-ductile transitiontemperature when steels/rotor shafts are heated or cooled slowlythrough the temperature range of 660°F to 1060°F. This shift in thebrittle-to-ductile transition temperature can be reversed by heatingat a temperature of 1100°F or higher then fast cooled.Consequently, when examining rotor or casings that have been inservice and operated within this temper embrittlement temperaturerange, the property toughness becomes an important criterion.

Evaluation of Toughness

Due to the advancement of fracture mechanics, it has nowbecome possible to characterize toughness in terms of critical flawsize ac. The definition of ac depends on the conditions under whichfinal rapid fracture occurs following the initial phase of subcriticalcrack growth. At rotor grooves and rotor bores where final fractureis likely to occur at low temperatures during start-stop transients,ac is dictated by linear elastic fracture mechanics. ac is given by anexpression of the form:

(10)

where KIC is the fracture toughness of the material or critical stressintensity for fracture, M is a constant to a given flaw size andgeometry, and σ is the nominal applied stress.

A typical loading sequence for turbine rotors shows variation intemperature and stress and their effect on critical flaw size ac forcold start sequence (Viswanathan and Jaffee, 1983). Figure 19shows the cold start sequence and associated variations in stress(σ), temperature (T), and critical flaw size (ac) as a function of timefor the power plant rotor, which failed catastrophically.

Figure 19. Illustration of Cold-Start Sequence and AssociatedVariations in Stress (σ), Temperature (T), and Critical Flaw Size(ac) as Functions of Time from Start.

Region A consists of a warmup period after which the rotor wasgradually brought up to speed (Region B). Once continuousoperating speed was reached (Region C) approximately 3 hoursafter the warm up period, maximum loading was applied. Analysisof the failure location (seventh row) showed that stresses reacheda peak value of 74 ksi (520 MPa), 11/2 hours after the maximumcontinuous speed had been attained. The critical crack size reachedits lowest value of 0.27 inch (6.9 mm) at a temperature of 270°F(132°C) and 74 ksi (510 MPa). Since variation in temperature,stress, and material fracture toughness at the defect location candictate the ac value for the rotor, one must calculate for the worstcombination of these variables to prevent failure.

This can be done by using lower scatter band values of KIC.However, to determine KIC for rotors, large specimens need to betaken to satisfy the plane strain conditions required for a valid test.A more common practice is to determine the ductile-brittle fractureat transition temperature (FATT) using samples extracted from the

periphery of a rotor and converting the FATT into KIC using a cor-relation such as the one shown in Figure 20 (Schwant and Timo,1985).

Figure 20. Turbine Rotor and Wheel Toughness Data.

Viswanathan and Gehl (1991) evaluated a number of data fromnumerous exposed CrMoV rotors and defined a lower thresholdband for low alloy steels. The lower limit line for CrMoV steel isdefined by the equation (Viswanathan and Wells, 1995) below:

(11)

where KIC is expressed in ksi-(in) 1/2 and TE is the excess temper-ature (T-FATT) expressed in degrees Fahrenheit. Once the FATT isknown, a KIC versus temperature T curve can be established andused to determine ac versus T.

However, such procedures tended to be conservative and therehave been various other nondestructive and relatively destructivetests involving removal of very small samples to determinetoughness of rotors. These techniques investigated include eddy-current examination, analytical electron microscope, secondary ionmass spectroscopy (SIMS), compositional correlations, Augerelectron spectroscopy, chemical etching, use of single Charpyspecimen, and small punch tests. The techniques that show themost promise and have currently been applied in service applica-tion are:

• Correlation based on composition.

• Small punch testing.

• Chemical etching.

Correlation Based on Composition

An American Society for Testing and Materials (ASTM) specialtask force on large turbine generator rotors of Subcommittee VI ofASTM Committee A-1 on steel has conducted a systematic studyof the isothermal embrittlement at 750°F of vacuum carbon deox-idized (VCD) NiCrMoV rotor steels. Elements, such as P, Sn, As,Sb, and Mo were varied in a controlled fashion and the shifts inFATT, (Δ) FATT were measured after 10,000 hours of exposure.From the results, the following correlations were observed inEquation (12):

(12)

where ΔFATT is expressed in degrees Fahrenheit and the correla-tion of all the elements are expressed in weight percent. According


K M aIC c= σ π

K T TIC = + + +95042 05872 0 00168 0 000001632. . . .Ε Ε

( )ΔFATT P Sn As Mo P Sn= + + − − ×13544 12950 2100 93 810 000,

to this correlation, the elements P, Sn, and As increase temperembrittlement of steels, while Mo, P, and Sn interaction decreasethe temper embrittlement susceptibility.

All available 10,000 hour embrittlement data are plotted inFigure 21 as a function of calculated ΔFATT using Equation (11),(Newhouse, et al., 1972). A good correlation is observed betweencalculated and experimental ΔFATT. The scatter for these data isapproximately ±30°F for 750°F exposure and ±15°F for the 650°Fexposure.

Figure 21. Correlation Between Compositional Parameter “N”and the Shift in FATT of NiCrMoV Steels Following Exposure at650°F and 750°F for 8800 Hours.

Other correlations for determining the temper embrittlementsusceptibility of steel, such as the J factor proposed by Watanabeand Murakami (1981) and X

_factor proposed by Bruscato (1970),

are widely used.These factors are given by:

(13)

(14)

The Figure 22 and 23 show the relationship between increase ofFATT and J factor and X

_factor at 399°C (750.2°F) for a 3.5 percent

NiCrMoV steel.

Small Punch Testing

Small punch testing of small disk-like specimens subjected tobending loads have found good correlation for determining theductile-brittle transition temperature (Baik, et al., 1983). Theprocedure consists of thin plate 0.4 � 0.4 � 0.02 inch subjected toa punch deformation with a 0.09 inch diameter steel ball in aspecially designed specimen holder. The test is performed atvarious temperatures and from the load deflector curves obtainedat various temperatures, the fracture energy is calculated. Thefracture energy is plotted as a function of test temperature todetermine the ductile-to-brittle transition temperature. The areaunder the deflector curve denotes the energy absorbed during thetest. This test procedure has been used successfully on a number ofretired rotor samples to determine the TSP (ductile-to-brittle transi-tion) and found to correlate well with the Charpy FATT values(Foulds, et al., 1991) (Figure 24).

Figure 22. Correlation Between Compositional Factor “J” and theShift in FATT of NiCrMoV Steels Following Exposure at 650°F and750°F for 8800 Hours.

Figure 23. Relationship Between Increase of FATT and X_.

Figure 24. Correlation Between Charpy FATT and Small PunchTransition Temperature.


( )( )J Si Mn P Sn= + + 104

( )X P Sb Sn As= + + +10 5 4 102

Chemical Etching

The grain boundaries of embrittled steel are attacked preferen-tially by picric acid solutions consisting of saturated picric acidsolution with an addition of 1 gram of tridecyl/benzene sulfonate(per 100 ml of aqueous picric acid). Correlations have been madebetween the grain boundary groove depth as measured metallo-graphically even from plastic replicas and the ΔFATT of the sampledue to prior temper embrittlement. By using plastic replicas thistechnique becomes very attractive for field use (Kadoya, et al.,1991).

Also Kadoya, et al., (1991) were able to correlate the width ofthe grain boundary groove measured from plastic replicas using ascanning electronic microscopes (SEM), with the FATT of thesamples and actual rotors. They also correlated the followingequation based on regression analysis of a number of materialvariables. This equation was found to predict the FATT within ascatter of ±20°C.

(15)

where:J is defined as (Si + Mn) (P + Sn) 104

W = Width of grain boundary grooveHV = Vicker hardness of sample/rotor

By utilizing this equation, the calculated FATT values werecompared with critical FATT values and the results were veryattractive (Figure 25).

Figure 25. Comparison Between Measured FATT and PredictedFATT Using the Method of Etching.

HIGH TEMPERATURE CORROSION

Fluid catalytic cracking (FCC) hot gas expanders operate inenvironments that can be both corrosive and erosive. Although it iswell documented that the source of erosion comes from the regen-erated catalyst that is carried with the hot flue gas from the FCC,its effect on high temperature corrosion has only begun to beunderstood by the authors’ company. Papers published by theauthor outline the relationship of stress and temperature on thehigh temperature corrosion/fracture mechanics of Waspaloy® invarious catalyst environments (Dowson, et al., 1995; Dowson andStinner, 2000).

The nature of the corrosion attack is primarily influenced by thetype of crude oil stock, which in time has a bearing on the resulting

flue gas composition, regenerated catalyst, and the nature andquality of additions injected into the FCC process.

When evaluating remaining life assessment of hot gas expandersespecially the rotating components, one must consider the effect ofthe environment such as high temperature corrosion. The authors’company has developed a fracture mechanics model that incorpo-rated both the effect of oxide wedge formation and the apparentreduction in fracture toughness of Waspaloy® in contact with thecatalyst residue. By utilizing this model one can predict whetherfracture will occur under various environmental/operating condi-tions of the hot gas expander. The authors’ company periodicallytests catalysts from end-users’ hot gas expanders to determine ifoxide wedge can occur and what life span to reach the critical sizefor failure. Generally, if the catalyst is active, then high tempera-ture corrosion will occur. Consequently, a blade will be removedfrom the unit and examined metallographically to determine theoxide wedge depth. Based upon the depth and time of operation,the remaining life can be estimated (Figure 26).

Figure 26. Stress Intensity Profile Versus Oxide Wedge Depth forUnit A. Critical Oxide Wedge Depth for Failure Was Defined as ac.In the Failed Blade, a2 Was Found to Exceed ac. In an IntactBlade, a1 Was Less Than ac.

CASE STUDY 1—REMAINING LIFE ASSESSMENTOF STEAM TURBINE CASING

Background Information of the Turbines

The subject ethylene plant has three steam turbines that drivethree trains of compressors on the main deck. The units are listedin the following. The three turbines were commissioned in August1981. They have been using the same steam source and areoperated in similar conditions (as summarized in Table 2). Thiscase study only discusses the remaining life assessment of one ofthem, a seven-stage 14 MW steam turbine.

According to the plant logbook, in the past 23 years, the turbinewas started/stopped 24 times, which are plotted in Figure 27.During the last 23 years of operation, only two failure incidentswere recorded. The first one was a damage on the rotor, due to con-densate inlet during the startup of the turbine in the late 1980s. Theother was found in the latest turnaround in 2004, in which a bladefailure and multiple cracks at steam balance holes were found onthe sixth stage disk that was subject to repair. The casing neverexperienced any problem in the past. During the turnaround in1995, a remaining life assessment was performed by Wiegand(1995), and the casing was found to be in a very good condition, asshown later in the top photo of Figure 30. The turbine componentmaterials are listed in Table 3.


FATT W H Si Mn

P Cr Sn JV= + + +

+ + + + +9912 1609 816 4 6520

3320 3104 3404 0 282 3256

. . .

. . .

Table 2. Turbine Operating Conditions.

Figure 27. Turbine Start/Stop Record.

Table 3. Turbine Component Materials.

Field Inspection

The turbine upper casing was visually examined. No cracks orreportable indications were found. The split-line bolt-holes wereexamined with dye-penetrant inspection, which did not reveal anycracks. The casing was polished at six different spots shown inFigure 28. The microstructure was observed with a portable opticalmicroscope, and replicated for documentation (Figure 29).Microstructure deteriorations, such as carbide precipitation at grainboundaries and bainite degradation, were evident. However, thecasing did not show any creep voids at any of the spots beingexamined. Hardness was measured at the same spots and theresults are listed in Table 4.

Microstructure Deterioration of the Casings Material

In order to study the microstructure deterioration, a fragment ofthe sample being cut off from the casing was reheat treated torestore the original microstructure for comparison. The reheattreatment was done in two steps: normalized at 1685°F (918°C) for1 hour and controlled air-cool, then tempered at 1250°F (677°C)for 2 hours and air-cool. The controlled air-cool from austenitetemperature was to obtain a similar percentage ratio of proeutec-toid ferrite and bainite as of the original heat treatment. Assumingthe casting was 23/4 inch (70 mm) thick, a cooling rate of 54°F perminute (30°C/m) was employed and the resultant ferrite/bainiteratio simulated the original heat treatment very well.

The reheat treated and service-exposed samples were mountedand polished and examined with an optical and microscope SEM.Their hardness was also measured. The findings are summarized inthe following.

• Comparing with the restored microstructure, the service-exposed samples showed profound carbide precipitation at grainboundaries, and the extent of the precipitation increased withincreasing the service exposure time (Figures 30, 31, 32).

Figure 28. The Turbine Casing and the Replication/CutoffLocations. F Represented the Hottest Spot, While E Representedthe Coolest Spot. D Was at the Transition Radius of Steam Chest toCasing Barrel, Which Was Invisible From this Photo.

Figure 29. Micrograph of a Replica Taken from Spot D. CarbidesPrecipitated and Partially Networked at Grain Boundaries.However, No Creep Voids Were Observed There. Five Percent NitalEtched.

Table 4. Casing Hardness Was Fairly Consistent from Spot to Spot.


123456789

101112

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

Year

Mon

th

Designed Conditions Actual Conditions

Temperature, F ( C)

Inlet 1st Stage Disk

Inlet Pressure,

psig (kg/cm2g)

Inlet Temperature,

F ( C)

Inlet Pressure,

psig (kg/cm2g)

806 (430) 692 (367) 696 (48.9) 797 (425) 710 (50)

Steam-End Casing Rotor Forging Blades Diaphragms

1.25Cr-0.5Mo alloy steel

casting, ASTM A217

Grade WC6

(limited to 950 F service)

Ni-Mo-V alloy

steel, ASTM A470

Class 4

AISI 403 stainless

steel for all stages

Alloy steel plate ASTM A517 Grade

F & A516 Grade 60 for 2nd to 5th

stages. Gray cast iron ASTM A278

Class 40 for 6th & 7th stages.

Brinell Hardness [HB] (converted from Equotip data)

Spot ID# 1st Reading 2nd Reading 3rd Reading Average

A 152 150 155 152

B 151 151 156 153

C 154 151 153 153

D 159 158 159 159

E 137 141 140 139

F 154 154 154 154

Figure 30. The Extent of Carbide Precipitations at GrainBoundaries Increased with Increasing the Service-Exposure Time.These Precipitations Were the Results of Long-Term Diffusion atElevated Temperature. A Reheat Treatment (Normalizing PlusTempering) Generally Restored the Microstructure by Dissolvingthe Carbides. Five Percent Nital Etched.

Figure 31. Microstructure of the 23 Years of Service-ExposedCasing Material. Note Deteriorated Bainite and CarbidesPrecipitated at Grain Boundaries and in Ferrite Phase.

Figure 32. After Being Reheat Treated, the Bainite Phase WasRestored, and the Carbides Were Generally Dissolved.

• Service-exposed samples also showed carbide precipitationinside the proeutectoid ferrite phase.

• Evidence of microstructure degradation of the bainite phase wasnoticed in the service-exposed sample, due to the transformation ofcementite (M3C) into different carbides (M2C and M7C3).

• No creep voids were found in either sample.

• Hardness was measured as HRB 91 (HB 160) for the service-exposed sample and HRB 93.5 (HB 169) for the reheat treatedsample, indicating a slight hardness decrease after a long-termexposure at elevated temperature.

Discussion and Conclusions

Turbine casings usually are normalized and tempered after beingcast. This heat treatment leads to a desirable combination ofstrength/hardness and toughness/ductility for the casing material,1.25Cr-0.5Mo alloy steel in this case. It also gives an excellentthermal stability for the casings. The typical treatment processcontains normalizing at 1685°F (918°C) followed by air-cool thentempering at 1250°F (693°C) followed by air-cool.


The microstructure after the heat treatment consists of proeutec-toid ferrite plus bainite. Bainite is a metastable aggregate of ferriteand cementite, which may degrade after long-term exposure atelevated temperature. The degradation was observed as a transfor-mation of cementite (M3C) into M2C and M7C3 carbides (Biss andWada, 1985). This transformation will lead to less carbides (involume percentage) and coarser carbide particles, which will resultin lower strength and hardness, which may reduce fatigue andrupture strengths of the material.

Carbon will also diffuse along the grain boundaries and into theproeutectoid ferrite phase to form carbides. Carbides at grainboundaries, especially in networked form, are considered detri-mental because they can result in embrittlement and reduce ruptureductility of the material. Another grain boundary damage that isalso related to long-term diffusion but was not seen in this case iscreep voids. Creep voids are considered as a higher-degree ofmicrostructure damage than the carbide precipitation, and awarning sign of component remaining life.

In the subject turbine casing, no creep voids were evident yet, sothe microstructure deterioration was considered moderate and notimmediately harmful to the casing life. It was concluded that thecasing was in class A condition and could be safely used for thefollowing service period (about five years). However, themicrostructure deterioration imposed a concern of the integrity of thecasing material and consequent potential risks on the creep strengthand rupture ductility of the casings. It was reported that 1.25Cr-0.5Mo low alloy steel could lose its rupture ductility after long-termexposure at elevated temperature (Demirkol, 1999). This behavior isso called rupture ductility trough, which is related to carbides pre-cipitation and may occur before creep voids are developed.

Therefore, in addition to continually monitoring the microstruc-ture with nondestructive replica techniques in the next turnaround,the authors’ company suggested that the ethylene plant conduct aLevel III remaining life assessment, which requires destructivematerial tests, in order to evaluate the extent of damage on creepstrength and rupture ductility of the casing material. The materialtests mainly consist of SRT and CDR test, which are described ina previous section. These two mechanical testings can supplyinformation that microstructure examination (replica) cannot.

CASE STUDY 2—REMAINING LIFE ASSESSMENTOF HOT GAS EXPANDER DISK

A life assessment was performed on a disk from a hot gasexpander that had been in operation for approximately 51,000hours of service. The present life assessment was necessary toensure that the properties of the disk had not degraded with time.This life assessment consisted of a microstructural evaluation aswell as mechanical testing.

Replicas of the microstructure of the disk were taken from fivelocations shown in Figure 33 and subsequently examined viaoptical microscope. A blade root from the disk that had seen thesame accumulation of time was also sectioned, metallographicallyprepared using the standard techniques, and examined.Specifically the microstructure was examined in the area of thefirst, third, and fourth landings. Also specimens were prepared toexamine corrosion products.

Representative micrographs of the microstructure of the root ofthe blade and of the disk may be seen in Figure 34. The structurewas found to be uniform and typical for A286 with an ASTM grainsize ranging from four to six. No creep voids were observed.

Several areas of high temperature corrosion attack were evidentin the area of the first and third root lines of the blade that wassectioned. A micrograph of this corrosion may be seen in Figure 35.

The maximum depth of penetration of corrosion product wasfound to be approximately 100 micrometers. No corrosion productwas observed in areas removed from this accelerated attack, whichsuggests that a thin Cr2O3 layer had formed over the surface of the

Figure 33. Diagram Indicating Location of Disk Replicas.

Figure 34. Optical Micrograph of Typical Microstructure of Rootof Blade #38. (15 ml HCI, 10 ml HN03, 10 ml Acetic Acid, 100�).

Figure 35. SEM Micrographs of Cross Section of Corrosion Seenin Radius of First Root Landing of Blade #38.

majority of the blade root and remained protective. This type ofcorrosive attack is expected for an iron chromium based alloyexposed to an atmosphere consisting of a mixture of sulfur andoxygen. Many high temperature alloys including A286 rely on theformation of the compact slow growing chrome oxide scale forprotection from the environment. Exposure to sulfur may lead tothe breakdown of this protective scale providing the necessary


conditions of temperature and pressure exist. The nature of thecorrosive products depends on both thermodynamic and kineticfactors, the details of which are beyond the scope of this paper. Amore detailed description of the mixed oxidants is given byDowson, et al. (1995), and Birks and Meier (1983).

It is not uncommon to see corrosion in the areas of blade rootdescribed in Figure 36. One explanation of this is that in theseareas there is a gap between the blade root and the disks that allowscatalyst into the blade root area. The presence of this catalyst maylead to high temperature corrosion as described by Dowson, et al.(1995), and Dowson and Stinner (2000). Some corrosion was alsoseen on the context surface of the land that is being proposed thatcorrosion in these areas is exacerbated by fretting. However, in thisinvestigation this type of corrosion occurred to a much lesserextent than seen in the areas mentioned previously.

Figure 36. Diagram Indicating Sectioning of Blade Root andLocation of Corrosion.

Because destructive testing could not be preformed on the actualdisk material, samples for stress rupture testing were taken fromthe root and lower airfoil of the blade as shown in Figure 37. It wasassumed that the properties of the sample are similar to those of thedisk roots. Modified stress rupture tests as developed by Dowson,et al. (1995), and Dowson and Stinner (2000) were also performedin order to determine if the corrosion products seen in the bladeroot area would cause cracking of the disk.

Figure 37. Photograph Illustrating Location from Which RuptureSpecimen Was Taken Relative to the Blade Root/Lower Airfoil.

The stress rupture specimen from the blade was found to havean elongation of 10 percent and a reduction of area of 14 percent.The fracture occurred in the smooth section of the bar after 195.5hours. This meets the authors’ company specification, whichrequires that the specimen last at least 30 hours with ruptureoccurring in the smooth section and possess an elongation of atleast 10 percent. The mill certification of the original bladematerial gives an elongation and reduction in area of 11.7 and 19percent, respectively, which is comparable to the present results.Thus, the rupture properties of the disk have not been significantlydegraded as a result of operation of the expander.

Modified Stress Rupture Testing

This test developed by Dowson, et al. (1995), is used todetermine the stress to cause fracture of the disk material in thepresence of spent catalyst. The stress may be converted to a stressintensity that may be compared to the stress intensity at the tip ofthe “wedge” created by the corrosive product. A determinationmay then be made as to whether cracking in the disk will occur. Itwas found that the presence of spent FCC catalyst may greaterreduce the rupture strength of Waspaloy®. For example, a sampletest of 1200°F and a stress of 95 ksi lasted over 900 hours whentested in air, while the same temperature and stress producedfailure in only 3 hours when the sample was tested in the presenceof catalyst. In the present investigation, an initial test of 1200°Fand a stress of 54 ksi confirmed that A286 is also susceptible to thistype of degradation. The sample failed in only 4.25 hours asopposed to 500 hours as predicted by the authors’ companymaterial property database for stress rupture in air. Therefore, inorder to get an accurate assessment of the rupture properties of thedisk in the area of the root in question, the modified stress rupturetest was used by Dowson and Stinner (2000). In the case at hand atemperature of 1065 F was chosen for the test based on the tem-perature of the disk rim at the point of maximum stress, which wasdetermined by finite element analysis to be in the radius of thefourth root landing. The stress used in the tests range from 62 to102 ksi. Each test consists of an exposure of stress and temperaturefor at least 20 hours. If the specimen did not break it was unloadedand the catalyst replaced and then retested at a stress that wasincreased by 10 ksi. This process, which was repeated untilfracture occurred at a stress of 102 ksi, corresponds to a stressintensity of 24.27 ksi inch. It should be noted that the catalyst thatwas used is kept at the authors’ company for testing purposes, i.e.,it was not from the actual end user’s unit. It is likely that this testcatalyst was much more contaminated than the end user’s catalyst,which will result in lower values of rupture stress than if the actualend user’s catalyst was used.

In order to determine the stress intensity of the tip of thecorrosion wedge it was first necessary to determine the stressprofile. The σ (X) of the area of the root in question, the stressintensity may be calculated from Equation (16):

(16)

where a is the wedge depth and the X is the coordinate dimension.The term M (x, a) is the weight function that is dependent on thegeometry of the cracked body. When considering the corrosion,wedge formed within the area of a disk rim, the geometry may beapproximated as an edge crack in a semi-infinite plate, Equation (17):

(17)

A linearized stress field of the fourth root landing of the disk rimof the end user’s expander was determined by a finite element


( ) ( )K x m x a dxI

a

= ∫σ ,

0

( )( )

m x aa x

x

a

x

a, . .=

−+ −⎛

⎝⎜⎞⎠⎟

+ −⎛⎝⎜

⎞⎠⎟

⎛

⎝⎜⎜

⎞

⎠⎟⎟

⎡

⎣⎢⎢

⎤

⎦⎥⎥

2

21 5693 1 279375 1

2

π

analysis. From these data an expression of σ (X) was derived andthe integral of Equation (16) evaluated to give the stress intensityat the tip of the corrosion wedge. The results of these calculationsare summarized in Figure 38, which is a plot of stress intensityversus corrosion wedge depth.

Figure 38. Stress Intensity Versus Corrosion Wedge Depth forFourth Root Landing of Disk.

Based on this plot, a corrosion wedge will have to penetrate toan approximate depth of .021 inches before the critical stressintensity is exceeded and catastrophic fracture occurs. Asmentioned, the maximum penetration seen was approximately .004inches. Because the growth of the corrosion product is parabolic,e.g., the growth slows with time, it is unlikely that the corrosionwould cause a failure of the disk providing the temperature andatmosphere of the expander are not changed significantly.

Conclusions

The microstructure in the ruptured product has not been signifi-cantly affected by the operation of the expanders. Some hightemperature corrosion was seen in the root area; however, it hasbeen determined that this corrosion will not result in the failure ofthe disk. Therefore, the disk will be used without significant risk offailure. It is recommended that this be reevaluated after an opera-tional period of four to five years.

REFERENCES

Baik, J. M., Kameda, J., and Buck, O., 1983, Scripta Met., 17, p.1143-1147.

Birks, N. and Meier, G. H., 1983, Introduction to High Temper-ature Oxidation of Metals, London, England: Edward Arnold.

Biss, V. A. and Wada, T., 1985, “Microstructural Changes in 1Cr-0.5Mo Steel after 20 Years of Service,” MetallurgicalTransactions A, 16A, January, pp. 100-115.

Bruscato, R. M., 1970, “Temper Embrittlement and CreepEmbrittlement of 2 °Cr-1 Mo Shielded Metal-Arc WeldDeposits,” Welding Journal, 35, p. 148s.

Bush, S. H., 1982, “Failures in Large Steam Turbine Rotors, inRotor Forgings for Turbines and Generators,” R.I. Jaffe,Editor, New York, New York: Pergamon Press, pp. I-1 to I-27.

Carlton, R. G., Gooch, D. J., and Hawkes, B. M., 1967, “TheCentral Electrical Generating Board Approach to ‘TheDetermination of Remanent Life of High Temperature TurbineRotors,’” I. Mech E. Paper C300/87.

Demirkol, M., 1999, “On the Creep Strength-Rupture DuctilityBehavior of 1.25Cr-0.5Mo Low Alloy Steel,” Journal ofEngineering and Environmental Science, 23, pp. 389-401.

Dowson, P., 1994, “Fitness for Service of RotatingTurbomachinery Equipment,” 10th Annual North AmericanWelding Research Conference, Columbus, Ohio.

Dowson, P., 1995, “Fracture Mechanics Methodology Applied toRotating Components of Steam Turbines and CentrifugalCompressors,” Third International Charles Parson TurbineConference, Materials Engineering in Turbine andCompressors, 2, pp. 363-375.

Dowson, P. and Rishel, D. M., and Bornstein, N. S., 1995, “Factorsand Preventive Measures Relative to the High TemperatureCorrosion of Blade/Disk Components in FCC Power RecoveryTurbines,” Proceedings of the Twenty-Fourth TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas, pp. 11-26.

Dowson, P. and Stinner, C., 2000, “The Relationship of Stress andTemperature on High-Temperature Corrosion FractureMechanics of Waspaloy in Various Catalyst Environments,”Proceedings of High Temperature Corrosion and Protection,Materials at High Temperatures, 18, (2), pp. 107-118.

Foulds, J. R., Jewett, C. W., and Viswanathan, R., 1991, “MiniatureSpecimen Test Technique for FATT,” ASME/1EEE JointPower Generator Conference, Paper No. 91-JPGC-PWR-38,ASME, New York, New York.

Goldhoff, R. M. and Woodford, D. A., 1972, “The Evaluation ofCreep Damage in CrMoV Steel in Testing for Prediction ofMaterial Performance in Structure and Components,” STP 515American Society for Testing and Materials, Philadelphia,Pennsylvania, pp. 89-106.

Kadoya, Y., et al., 1991, “Nondestructive Evaluation of TemperEmbrittlement in CrMoV Steel,” ASME/JEEE Joint PowerGenerator Conference, Paper PWR-Vol. 13, New York, NewYork.

Neubauer, E. and Wadel, V., 1983, “Rest Life Estimation of CreepComponents by Means of Replicas,” Advances in LifePrediction Methods, Editors, Woodford, D. A., and Whitehead,J. E., (New York, New York: ASME), pp. 307-314.

Newhouse D. L., et al., 1972, “Temper Embrittlement of AlloySteels,” ASTM STP 499, pp. 3-36.

Pope, J. J. and Genyen, D. D., 1989, International Conference onFossil Power Plant Rehabilitation, ASM International, pp. 39-45.

Saxena, A., 1998, “Nonlinear Fracture Mechanics for Engineers,”Boca Raton, Florida: CRC Press, p. 449.

Schwant, R. C. and Timo, D. P., 1985, “Life Assessment of GeneralElectric Large Steam Turbine Rotors in Life Assessment andImprovement of Turbogenerator Rotors for Fossil Plants,”Viswanathan, R., Editor, New York, New York: PergamonPress, pp. 3.25-3.40.

Viswanathan, R. and Gehl, S. M., 1991, “A Method for Estimationof the Fracture Toughness of CrMoV Rotor Steels Based onComposition,” ASME Transaction, Journal of EngineeringMat. and Tech., 113, p. 263.

Viswanathan, R. and Gehl, S. M., 1992, “Life-AssessmentTechnology for Power Plant Components,” JOM, February, pp.34-42.

Viswanathan, R. and Jaffee, R. J., 1983, “Toughness of Cr-Mo-VSteels for Steam Turbine Rotors,” ASME Journal ofEngineering Mat. and Tech., 105, pp. 286-294.

Viswanthan R. and Wells, C. H., 1995, “Life Prediction of TurbineGenerator Rotors,” Third International Charles ParsonsTurbine Conference, Materials Engineering in Turbines andCompressors, 1, pp. 229-264.


Watanabe, J. and Murakami, Y., 1981, “Prevention of TemperEmbrittlement of CrMo Steel Vessels by the Use of Low SiForged Steels,” American Petroleum Institute, Chicago,Illinois, p. 216.

Wiegand, R., 1995, “Petroquimica Bahia Blanca Turbine CasingAssessment,” An Internal Report.

Woodford, D. A., 1993, Materials and Design, 14, (4), p. 231.

Woodford, D. A., Von Steele, D. R., Amberge, K., and Stiles, D.,1992, “Creep Strength Evaluation for IN738 Based on StressRelaxation” from Superalloys, The Minerals, Metals andMaterial Society, Editors, Antolovich, S. D., Stusned, R. W.,Mackay, R. A., Anton, D. L., Khan, T., Kissinger, R. D., andKlarstrom, D. L., pp. 657-663.

ACKNOWLEDGEMENT

The authors are grateful for the support from the MaterialsEngineering Department at Elliott Company, and recognize ElliottCompany and Dow Chemical PBBPolisur for permission topublish this paper.


Thom Eldridge is Manager of theAero/Thermodynamics Engineering DesignDepartment at Dresser-Rand Company inOlean, New York. His previous responsibil-ities as supervisor of the RotordynamicsGroup included managing the analysis ofturbomachinery and guiding rotordynamicdevelopment activities. Before joiningDresser-Rand, he worked on bearing devel-opment, turbomachinery modeling, dataacquisition systems, and high-temperature

transducer design as a senior engineer and project team leader forBently Nevada.

Mr. Eldridge received a B.S. and an M.S. degree (MechanicalEngineering, 1991, 1994) from Washington State University, andan MBA degree (2002) from the University of Nevada. He is a reg-istered Professional Engineer in the State of California. Mr.Eldridge holds two patents and has written several papers ondamper seals and hydrostatic bearing applications.

Greg Elliott is a Senior Project Engineerin the Power Transmission Division ofLufkin Industries in Lufkin, Texas. Hisprimary responsibilities include productdevelopment and standardization, as wellas engineering systems development. Healso supports Lufkin engineering groupswith finite element analysis, fatigueanalysis, and other applications of engi-neering mechanics in machinery designand problem solving. Before joining Lufkin,

he worked in research and teaching at Texas A&M University.Mr. Elliott received a B.S. and an M.S. degree (Agricultural

Engineering, 1982, 1990) from Texas A&M University.

Ed Martin is a Project Engineer in thePower Transmission Division of LufkinIndustries, in Lufkin, Texas. He is currentlyresponsible for the engineering functions ofLufkin’s high-speed gear product line. Healso performs lateral rotordynamicanalyses and has been heavily involved introubleshooting gear vibration, noise, andtemperature problems during his 14-yearcareer with Lufkin.

Mr. Martin received a B.S. degree(Mechanical Engineering, 1990) from the University of Texas atAustin. He has given numerous presentations on gearing principlesand vibration issues, and is also an active member of the AGMA6011 committee on high-speed helical gearing.

ABSTRACT

Turbomachinery shafting and casings typically experience axialgrowth during thermal transients occurring at startup, shutdown,and changes of load. This growth does not usually present difficul-ties during operation. However, this paper discusses four recentturbomachinery trains that experienced radial subsynchronousvibration resulting from axial misalignment. In each example, thetrain included a gearbox with a double-helical gearset thatpresented a lateral vibration response at system torsional criticalfrequencies while operating at low load. Thermal growth resultedin axial misalignment that produced a “zero-backlash” conditionwhen one helix contacted on the back side of the teeth. Datacollected with an axial displacement probe verified the appearanceof lateral subsynchronous vibration was associated with excessiveaxial misalignment of the gearset. Data are presented from theseexperiences, including a discussion of other vibration phenomenathat could be mistaken for this behavior. Finally, design andanalysis recommendations are provided for preventing such occur-rences.

93

AXIAL ALIGNMENT AND THERMAL GROWTH EFFECTSON TURBOMACHINERY TRAINS WITH DOUBLE-HELICAL GEARING

byThom Eldridge

Manager of Aero/Thermodynamics Engineering Design


Olean, New York

Greg ElliottSenior Project Engineer

Ed MartinProject Engineer

Lufkin Industries

Lufkin, Texas

andM. Shukri Hitam

Turbomachinery Group Superintendent

Petronas Carigali

Terengganu, Malaysia

INTRODUCTION

The authors have encountered several instances of unacceptablegearbox lateral vibration arising from forces external to the gearboxcaused by axial misalignment. Most of the lessons learned fromthese experiences build upon previous findings (Zirkelback, 1979;Mancuso, 1986; Carter, et al., 1994). However, most emphasis hasbeen placed on parallel or angular alignment problems. Althoughaxial alignment problems may occur less frequently, the delays andexpense caused by such problems make it worthwhile to share thisinformation with others in the rotating equipment community. Thispaper presents necessary background, examples, and recommenda-tions for problem diagnosis and prevention.

Helical Gearing Background

Parallel-shaft, double-helical gearing is widely used in high-speed,high-power turbomachinery applications. In these applications,double-helical designs usually are the best choice because of theirhigh power density, smooth meshing, reduced thermal distortion,high-energy efficiency, and high reliability. The term “double-helical” refers to the use of two sets of helical teeth arranged withopposing helix directions. Figures 1 and 2 show typical double-helical gearsets used in turbomachinery trains. Figures 3 and 4illustrate basic gear terminology (Dudley, 1984; Townsend, 1991).

Figure 1. Double-Helical Gearset Used in a Motor—CompressorDrive.

Figure 2. Double-Helical Gearset Used in a Gas Turbine—Compressor Drive.

Axial thrust induced by gear mesh force is a characteristic ofhelical gearing that is particularly significant to the topic of axialalignment. The axial mesh force of a set of helical gear teeth is cal-culated (Dudley, 1984) as:

Figure 3. Gear Terminology.

Figure 4. Gear Mesh Centered.

(1)

where:FMESH = Mesh centering force (lbf)T = Transmitted torque (inch-lbf), per helixDP = Pitch diameter (inch)H = Helix angle (degrees)

The force can be quite strong. For example, in a typical double-helical gearset with a 12 inch (305 mm) pitch diameter and a28-degree helix angle transmitting 40,000 horsepower (30,000kW) from a gas turbine at 5000 rpm to a centrifugal compressor at10,000 rpm, each helix generates an axial force of about 11,000 lbf(49,000 N). When applying the formula, one must use the trans-mitted torque and pitch diameter together for the same element.When applying it to double-helical gearing, the actual transmittedtorque per helix should be used.

Under normal circumstances, the mesh forces act equally oneach element in opposite directions. This causes the axial meshforces to cancel internally within a double-helical gearset.Canceling the axial force eliminates the need for high-capacitythrust bearings that would otherwise be required in the gearbox toabsorb the axial mesh force. Typically, a gearbox with double-helical gearing will have one thrust bearing on the low-speed gear.If the connected equipment permits, a double-helical gearbox mayhave no thrust bearing. Additional information on helical gearingcan be found in numerous references (Dudley, 1984; Drago, 1988;Dudley’s Gear Handbook, 1991).

Double-Helical Mesh Axial Alignment

The axial alignment problem discussed in this paper pertains todouble-helical gearing. If the external axial forces on the gearelements are less than the mesh centering force, the mesh holds thepinion centered on the gear as shown in Figure 4. If the externalaxial force is zero, then the torque load on each helix is equal.


( ) ( )F T D HMESH P= ∗2 / tan

Backlash (clearance) between the teeth is provided in the gearsetto avoid contact of the nondriving (back) sides of the teeth. Indouble-helical gearing, backlash can be defined in the circumfer-ential and axial directions. Circumferential backlash is illustratedin Figure 3. The axial backlash is equal to the circumferentialbacklash divided by the tangent of the tooth helix angle. For typicalturbomachinery applications, a small double-helical gearset mighthave 0.015 inch (0.4 mm) circumferential backlash and 0.028 inch(0.7 mm) axial. Large gearsets may have three or four times thoseamounts. However, it is important to note that the axial backlash sodefined includes the total potential axial displacement of the pinionrelative to the gear in both directions. The pinion can move offcenter of the gear in each direction only by a distance equal to halfthe axial backlash. For a gearset with 0.028 inch (0.4 mm) of axialbacklash, the teeth will bind if one attempts to shift the pinion morethan 0.014 inch (0.4 mm) off center in either direction relative tothe gear. This is illustrated in Figures 4, 5, and 6.

Figure 5. Gear Mesh Offset Axially.

Figure 6. Gear Mesh Bound Axially.

In turbomachinery trains, often there will be axial forces causedby axial misalignment, thermal effects, thrust bearings, magneticcentering forces from electrical machines, gravity loads duringship pitch, and so on. As previously described, when an externalaxial force is applied to the double-helical gearset, it offsets part ofthe axial mesh force. This causes no harm (or axial displacement)if the external force is small compared to the axial mesh force. Thatis typically the condition while service torque is applied. If theexternal force is larger than the mesh axial force, the two meshinggear elements can be pushed axially out of proper engagement asshown in Figure 5. This is most likely to occur when torque load issmall, as may be encountered with a no-load string test, standbyoperation of a generator set, or during startup or shutdown tran-sients.

If the source of the external force reaches equilibrium with theaxial mesh force, the gearset may run satisfactorily on one helix (as

in Figure 5). This can occur, for example, when a flexible couplingis involved. If axial backlash is sufficient, the coupling axial forcedecreases as the pinion moves off center until the coupling axialforce equals the mesh axial force. In Figure 5, the backlash is suf-ficient to maintain clearance at the nondriving sides of the teethand permit the gearset to run smoothly. Note that when the gearsetis running on one helix, the total transmitted torque is carried onthat helix alone. Typically, when service torque load is applied, themesh force increases and the gearset centers itself axially. If theexternal axial force is significant compared to the axial mesh forceat service torque, it may cause unequal loading of the helices anddecrease load carrying capacity.

If the external force is greater than the axial mesh force, it willpush the gearset into a bound condition as shown in Figure 6. In thisstate there is still contact on the normally loaded (driving) toothflanks on one helix. On the other helix there is contact on the back(nondriving) flanks of the teeth (which should not normally be incontact). Running with contact on the driving side of one helix andthe nondriving side of the other helix is similar to operating withzero backlash. It is well known that high-speed, high-power gearingshould not be run without backlash (Dudley, 1984).

Quantifying External Axial Forces

As previously identified, external axial forces can originate froma number of sources. Frequently such a force is generated by theaxial displacement of a flexible element coupling (in compressionor tension). Many turbomachinery trains include either a disc ordiaphragm coupling between the gearbox and the connectedequipment shaft ends. Axial displacement of coupling flexibleelements can originate from prestretch of the coupling, axialalignment inaccuracies, and from thermal growth of equipment(both steady-state and transient conditions). The main purpose ofcoupling prestretch is to cancel out the steady-state thermal growthof the coupling and shaft ends. In this paper, transient thermalgrowth and axial alignment errors are shown to result in excessaxial forces that are capable of forcing the pinion out of center withthe gear of a minimally loaded gearset.

A simple estimate of the axial force generated by a flexibleelement coupling can be obtained through the axial stiffness of thecoupling:

(2)

where:FAXIAL = External axial force (lbf)DSE = Relative displacement of shaft ends attached to the

coupling (inch)K = Axial stiffness of the flexible element coupling (lbf/in)

The expected displacement of the shaft ends can be calculated.Commonly it will be a function of axial alignment error andthermal growth. Evaluation of the mesh centering force [Equation(1)] defines an upper bound to the allowable external axial force[Equation (2)]. This establishes a maximum allowable error inaxial alignment. It also provides a guide to the maximum allowabletransient thermal growth, before disengagement of the gearelements should be anticipated. In instances of large thermal tran-sients, it may be necessary to modify the steady-state axialalignment to accommodate the transient thermal growth condition.

To put this in perspective, consider the hypothetical 40,000 hp(30,000 kW) compressor drive mentioned previously. Assume anexternal axial force is applied to the pinion by axial displacementof a typical flexible-disc type coupling, K = 5000 lbf/in (875kN/m). At steady-state operating conditions with reasonably loosealignment accuracy, less than 0.050 inch (1.3 mm) error, the axialcoupling force would be 250 lbf (1100 N). For the gearset of thisexample, a 250 lbf axial mesh centering force would be generatedby a helix transmitting 450 hp (335 kW). The mesh will remaincentered if the transmitted power is greater than 1 percent of


95

F D KAXIAL SE= ∗

normal load. If a typical diaphragm-type coupling was used (K =20,000 lbf/in), the load necessary to keep the mesh centered couldbe closer to 3 or 4 percent of normal transmitted load.

INTRODUCTION TO EXAMPLES

Based on experiences such as those described in this paper, theauthors believe the resulting nonuniform transmission of motionfrom a bound mesh can supply broad-spectrum angular or lateralexcitation to the system. Experience has shown this can lead toexcessive vibration by directly stimulating lateral displacement orthrough interaction of lateral and torsional vibration in the gearset(Hudson, 1992). Additionally, low bearing stiffness (such asencountered in a gearbox with pressure-dam bearings operatingunder light load) has been reported to increase the interactionbetween torsional and lateral response (Pennacchi and Vania,2004). Finally, the combination of running on both the loaded andunloaded faces of the gears produces high relative variation in gearmesh stiffness. Variation in gear mesh stiffness has been discussedas a source of torsional excitation capable of producing noticeablelateral response (Iwatsubo, et al., 1984). If the axial force isextremely high, the wedging effect can cause tooth overload andfailure, and thrust bearing wear can be accelerated.

The following examples illustrate practical problems thatoccurred because of thermal transients and axial misalignment.Four examples are described with emphasis on their vibration sig-natures. They highlight some common causes of excessive axialforce in turbomachinery trains. Additionally, they provide lessonsfor designing trouble-free systems and for diagnosing problems.The first example will be discussed in considerable detail to illus-trate the diagnostic process and illuminate how similar phenomenacan be mistakenly blamed for the appearance of this subsynchro-nous vibration. The mechanics of the thermal transient growth andgear binding in Example 1 are presented in APPENDIX A. Thesecond through fourth examples are included to present other trainconfigurations that have experienced these phenomena and to illus-trate some points not covered in the first example.

EXAMPLE 1—GAS TURBINE, GEARBOX,CENTRIFUGAL COMPRESSOR

This problem arose during a no-load string test of a 15,000 hpgas turbine, speed-increasing gearbox, and centrifugal compressorshown schematically in Figure 7. Maximum continuous speeds(MCOS) were 9300 rpm (155 Hz) at the turbine and 11,840 rpm(197 Hz) at the compressor. The original string test setup was“open-air,” the compressor had inlet and outlet silencers andabsorbed only an estimated 250 hp. This train was very similar toan existing design that had performed well in field operation.

Figure 7. Layout of Gas Turbine, Gearbox, CentrifugalCompressor Train in Example 1.

As shown in Figure 8, string test results typically showed avariety of subsynchronous vibration (SSV) peaks in the gearboxlow-speed (input) shaft displacements with most notable compo-nents near 30 to 40 and 70 to 80 Hz. The highest SSV peaks wereapproximately 0.0005 inch (0.013 mm) near 40 Hz. Amplitude at1� was about 0.0005 inch (0.013 mm) and overall levelsapproached 0.002 inch (0.05 mm). Significant SSV appeared onlyafter operating at MCOS for 12 to 30 minutes, disappearing afterreturning to idle speed. Only the gearbox low-speed shaft exhibitedthe SSV. All gearbox journal and thrust bearing temperatures wereas expected. Review of relevant frequencies identified a small SSVpeak near 35 Hz in the gearbox shop test. It also was noted that thepredicted first torsional resonant frequency of the train was 41.8Hz and the second was 75.5 Hz (refer to Figure 9). Preliminarydiagnosis indicated axial alignment and journal bearing instabilityas possible contributors to the SSV problem.

Figure 8. Radial Vibration of Gearbox Input Shaft During TypicalString Test of Example 1.

Figure 9. Torsional Model of Train from Example 1.

Diagnostic Run without Compressor

A diagnostic test was run with the compressor uncoupled fromthe gearbox output shaft. If the issue was journal bearing instabil-ity, the problem would get worse as the minimal 250 hp


compressor load was removed. Decreasing bearing load bydecreasing torque should make the cylindrical bore, pressure-dambearings in the gearbox less stable. If the issue was an axialalignment error, eliminating the compressor (and its thrust bearing)would eliminate the problem.

Performance of the unit during the uncoupled test showed theSSV had been eliminated (refer to Figure 10). The SSV peak justbelow 40 Hz in the original vibration spectrum was gone. A smallerpeak appeared at 60 Hz, very close to the calculated 59.5 Hz firsttorsional resonance with the compressor removed from the train.This was a strong indication that there was some level oftorsional/lateral interaction occurring in the train and that interac-tion was contributing to the SSV. This validated the theory that thesource of the vibration was alignment related, most likely axialalignment.

Figure 10. Radial Vibration During Diagnostic Test Run of Turbineand Gearbox Alone in Example 1.

Diagnostic Run with DecreasedHigh-Speed Coupling Prestretch

An additional 0.020 inch (0.51 mm) shim was added to the high-speed coupling to decrease the prestretch to 0.018 inch. This didnot solve the problem. Results at MCOS were similar to the initialruns, except that the SSV peaks were higher in amplitude and lessvibration hash appeared between the peaks. SSV appeared approx-imately 10 to 15 minutes into the run, and discreet componentsreached 0.0011 inch (0.028 mm) levels. Frequencies were muchmore distinct at 39, 78, 116, and 155 Hz, all of which are related tothe 155 Hz running speed. It appeared decreasing prestretch madevibration worse during shutdown. When the unit was manuallytripped, the lateral vibration spiked to higher levels. This resultfocused attention on nonalignment issues.

Diagnostic Run with Light Torque Load

At this point attention turned to load. There was consensus thatif sufficient torque load was placed on the train, the problems beingexperienced would be entirely eliminated, regardless of whetherthe problem was caused by axial misalignment or journal bearinginstability. However, the only way to increase load on the trainduring the string test was to install an orifice on the compressoroutput, which was expected to add 250 hp additional load over theunthrottled configuration. This was done, but the resulting 500 hpload was not enough to make a significant difference, except thatthe time to onset of SSV was increased slightly.

Gearbox Modifications

Focus shifted to the gearbox journal bearing design as a meansof decreasing vibration. Although rotordynamic analysis of thegearbox journal bearing design indicated there should be stableoperation (even at zero load), it was decided that a different

bearing design could eliminate bearing stability as a possible factorand reduce vibration levels (Nicholas, et al., 1980). The reducedvibration level would also avoid machine protection trips andshould provide a longer diagnostic run to help clarify the effect ofother potential causes. The gear supplier developed a modifiedjournal bearing design that was optimized for stability at no loadand could be implemented by modifying the bearings that were inthe machine. The low-speed (input) shaft journal bearings wereremoved from the gearbox, modified, and reinstalled.

At the same time the journal bearings were being modified, aprovision was added to the gearbox for installing an axialproximity probe at the high-speed pinion. The intent was not tomeasure axial vibration of the pinion, but to monitor the axialposition of the pinion and provide insight on axial alignment. Itwas expected the pinion axial probe would enable detection of gearmesh axial misalignment similar to Figure 5 or 6 if such acondition was present.

Diagnostic Run with Axial Probe onPinion and Modified Journal Bearings

This run revealed that the subsynchronous vibration was corre-lated with axial misalignment of the gear mesh. Simultaneouslymonitoring the low-speed shaft radial vibration and pinion axialmovement was the key to solving the problem. Verifying the pinionposition and subsynchronous vibration levels changed in concertduring the test illuminated the solution.

As shown in Figure 11, subsynchronous vibration appearedmuch the same as it had in previous runs. At about 8:42 the speedwas increased from 9000 rpm to MCOS of 11,844 rpm. The lateralvibration probes showed that until 8:57:36 a clean subsynchronousspectrum was present with low amplitudes and no significantdiscreet components (light data in Figures 11 to 13). After thattime, two significant SSV frequencies appear at 38 and 82 Hz(roughly corresponding to the first and second torsional reso-nances). This SSV remained consistent for 65 minutes (dark datain Figures 11 to 13) and then for a brief period the SSV presentedvarying frequencies. After that, a clean subsynchronous spectrumwas present (second light data in Figures 11 to 13). During thedeceleration at 11:00, the SSV reappeared momentarily, butsubsided once the new speed of 9000 rpm was maintained. Themodified (stiffer) bearings appeared to reduce subsynchronousdiscrete amplitudes and eliminated much of the nondiscreet SSV(noise or hash). However, they did not prevent the SSV fromoccurring.

Figure 11. Radial Vibration after Installing Pinion Axial Probe inExample 1.

Figures 12 and 13 show high-speed pinion and low-speed gearaxial positions. In these axial position plots 200 mV output voltagechange corresponds to 0.001 inch (0.025 mm) axial positionchange (200 mV/mil). For example, during the period from 8:42


97

1X

Figure 12. High-Speed Pinion Position Plot for Example 1.

Figure 13. Low-Speed Gear Position Plot for Example 1.

until 8:56 the pinion probe voltage changed 3.4 volts. That repre-sents a movement of 0.017 inch (0.43 mm). Not only was thepinion undergoing substantial axial displacement, comparingpinion and gear positions revealed that the pinion moved consider-ably further axially than the gear. This measurement verified thepinion and gear were not remaining axially centered on oneanother (similar to the condition depicted in Figures 5 and 6).Further evaluation of the data showed that the SSV began when themesh became tightly bound (transition from light to dark data).Details of the evaluation are presented in APPENDIX A.

Final Qualification Run with IncreasedPrestretch in High-Speed Coupling

Based on results of the previous run, the coupling prestretch wasincreased an additional 0.040 inch (1 mm) to 0.058 inch (1.5 mm)and another run was completed. The modified bearings from theprevious run were used. The gearbox was able to run through thethermal transients and be subjected to numerous speed cycleswithout the SSV appearing (refer to Figures 14 and 15). The axialprobe indicated the pinion axial position leveled off smoothly atthe maximum displacement, with no indication of mesh binding.Based on these results it was concluded that the problem wassolved. The unit has subsequently entered service. No SSV hasbeen detected during startup, steady-state, transient/load change, orshutdown.

EXAMPLE 2—STEAM TURBINE, GEARBOX, GENERATOR

Similar vibration behavior was experienced on a steam turbine-driven generator with a speed-reducing gearbox. In this train, the

Figure 14. Vibration with Increased High-Speed Coupling Pre-stretch in Example 1.

Figure 15. Pinion Axial Position Leveling Off Gradually inExample 1.

steam turbine is the driver. It is coupled to the high-speed pinionthrough a flexible element coupling. The low-speed shaft of thegearbox carried the thrust bearing and was connected to an electri-cal generator rotating at 1800 rpm. The train layout is shown inFigure 16.

Figure 16. Layout of Turbine, Gear, Generator Train of Example 2.

During commissioning of the train, one of the checks performedwas to unload the steam turbine at full speed by removing thegenerator load. This produces an unloaded condition in the gearbox.The sudden unloading of the steam turbine resulted in rapid cooling(and shortening) of the turbine rotor as the steam rate fell. Duringthe first unloading cycle a subsynchronous vibration was measured


on the gearbox pinion shaft as shown in Figure 17 with no otherexcessive vibration appearing elsewhere in the train. The vibrationappeared discreetly at 750 cpm and 2100 cpm, with magnitudegreater than the synchronous component. These SSV frequenciescorresponded to the predicted first and second torsional natural fre-quencies of the train, 755 cpm and 2230 cpm, respectively.

Figure 17. Radial Vibration of Pinion in Example 2.

Diagnostic Run with Increased Coupling Prestretch

The axial alignment of the train was modified after these charac-terization runs by increasing the prestretch of the high-speed couplingby 0.023 inch (0.6 mm). With this modification, the SSV continued.

It was hypothesized that the source of the vibration could be athermal transient, causing axial displacement of the shaft end andshifting of the pinion to gear alignment. The sudden cooling of thesteam turbine rotor had the potential to produce 0.10 inch (2.5 mm)of shaft-end movement. Unlike the situation described in Example1 (when the growth of the compressor shaft tended to push thepinion shaft end toward the gear), in this example the motion of theturbine shaft end would be to pull the pinion shaft end away fromthe gearbox. In Examples 1 and 2, the axial force applied to thepinion is sufficient to overcome the mesh centering force of thedouble-helical gear set and drive the pinion out of axial alignmentwith the gear. With sufficient axial motion, the axial backlash of thegears would be consumed and the gearset would become “bound.”

Diagnostic Run with Axial Probe on Pinion

To verify the hypothesized behavior, an axial probe wasinstalled in the gearbox to measure motion of the pinion. Figures18 and 19 present similar behavior to that observed in Example 1.A sudden discontinuity in the pinion axial position [approximately0.020 inch (0.5 mm) motion] trace corresponded to the rise of sub-synchronous vibration. This 0.020 inch (0.5 mm) displacementduring operation agreed with the 0.018 inch (0.45 mm) of axial endplay measured during installation checks.

Diagnostic Run with Modified Thrust Bearing Clearance

In order to remove the constraint on pinion motion imposed bythe low-speed gear, the axial clearance of the thrust bearing wasincreased by 0.040 inch (1 mm) with the modification of the shimpack. The shims were intentionally changed to move the axialrunning position of the gear closer to the steam turbine. In effect,this decreased the prestretch of the high-speed coupling, decreasedthe pulling force on the pinion during the cooling transient (andincreased the prestretch of the low-speed coupling). The behaviorof the gearbox improved significantly. Limited vibration was seenduring subsequent unloading and deceleration runs. The unitentered service to the satisfaction of the customer. Figure 20 showsthe much cleaner waterfall spectrum. Prior to the changes, the 750cpm component was equal to the 1� vibration level, while the 2100cpm had been up to three times the 1� vibration level.

Figure 18. Radial Vibration after Installing Pinion Axial Probe inExample 2.

Figure 19. High-Speed Pinion Axial Position in Example 2.

Figure 20. Radial Vibration after Increasing Thrust BearingClearance in Example 2.

EXAMPLE 3—STEAM TURBINE, GEARBOX, GENERATOR

Before the experiences described in Examples 1 and 2, a similarvibration behavior was encountered on another steam turbine-driven generator with a speed-reducing gearbox (layout shown inFigure 21). Although it was not understood at the time of theincident, it later became apparent that this was another instance ofaxial misalignment. As in Example 2, the steam turbine is the driverand is coupled to the high-speed pinion through a flexible elementcoupling. In this example, there were two separate but identicaltrains, and only one of the trains experienced the high vibration.


99

1X

1X

Figure 21. Layout of Turbine, Gear, Generator Train of Example 3.

During commissioning of one of the trains, the gear unit repeat-edly tripped on high radial vibration of the pinion during startup.An examination of the pinion shaft vibration cascade plot (Figure22) during startup revealed the main component of radial vibrationto be subsynchronous, occurring at approximately 10 Hz once thetrain reached a generator speed of 950 rpm (full speed = 1800rpm). It was later realized that this SSV frequency corresponds tothe train’s calculated first torsional natural frequency of 10 Hz.

Figure 22. Radial Vibration of Pinion of Example 3.

At the time of the incident, the focus of the investigationcentered on a journal bearing oil whirl. Accordingly, the pinionbearings were redesigned in an effort to suppress whirl. Before theinstallation of the modified bearings, an error was discovered onthe installed prestretch of the high-speed coupling. The installedprestretch was 0.030 inch (0.8 mm) less than the calculatedrequired value. This mistake was not made on the second train,which operated well and showed no signs of SSV. The decisionwas made to correct the prestretch by removing 0.030 inch (0.8mm) of shim in the high-speed coupling at the same time themodified pinion bearings were being installed.

After the modifications, the pinion was able to reach full speedwhile showing no signs of SSV, and the train was commissionedsuccessfully. An interesting point about this example is that therewere two identical trains, but only one experienced a problem. Thetrain with the correctly installed prestretch and original bearingdesign did not exhibit SSV.

EXAMPLE 4—GAS TURBINE, GEARBOX, GENERATOR

This train contained a gas turbine driving a generator through aspeed-reducing gearbox. The turbine was connected to the gearboxpinion by a diaphragm-type coupling. The gearbox low-speed shaftflange was bolted directly to the generator shaft flange without acoupling. As shown in Figure 23, there were thrust bearings at thenondrive end of the single-shaft gas turbine and on each side of thelow-speed gear. The subsynchronous vibration occurred duringcommissioning while running at full speed and no load.

Figure 23. Layout of Gas Turbine, Gear, Generator Train ofExample 4.

The train was started and brought to full speed at no load.During the first hour radial vibration was low. After running atno load about an hour, the high-speed pinion radial vibrationbegan increasing and continued to increase during a period ofseveral minutes until the machinery protection system trippedon high pinion radial vibration. This trip occurred shortly beforethe turbine was expected to complete its startup thermaltransient. Limited vibration data were available from acondition monitoring system. The pinion radial vibration had asignificant spectral component in the general vicinity of low-speed gear design rotating speed of 1800 cpm. Initially it wasthought the pinion SSV was a forced response at the low-speedgear rotating frequency.

Following a test run to obtain better vibration data, it was deter-mined the dominant SSV peak did not track gear rotatingfrequency. Further, the vibration could be observed in casing accel-eration and pinion displacements. Figure 24 is a representativecasing velocity spectrum showing frequencies involved. Thissample was taken at rotating frequencies of 5109 cpm (85.1 Hz) atthe turbine and 1800 cpm at the generator. The vibration frequencypeak was fixed near 1875 cpm (31.2 Hz). Once again there was cir-cumstantial evidence that torsional effects could be a factorbecause the predicted first torsional natural frequency was 1894cpm, very close to the response frequency.

Figure 24. Original Vibration Spectrum for Example 4.


-

Axial alignment of the turbine to the gearbox pinion wasreviewed. It was found that coupling axial thermal growth was notincluded in the original high-speed coupling prestretch calculation.The new analysis indicated that before steady-state conditionswere reached there would be enough thermal growth in the turbine,coupling, and pinion to push the pinion away from the turbine intoa bound condition as shown in Figure 6. Further, the growth wassufficient to push the low speed shaft through its thrust bearingclearance. Once the mesh was bound and the gear was pushedagainst its thrust bearing the remaining thermal growth had to beabsorbed by axial deflection of the high-speed coupling. The rela-tively high coupling axial stiffness enabled even a small amount ofcoupling deflection to create sufficient force to bind the mesh.

Diagnostic Run with Increased Coupling Prestretch

A test run was made with high-speed coupling prestretchincreased from 0.110 to 0.130 inch (2.8 to 3.3 mm) and with anaxial position probe installed at the high-speed pinion. This changewas not sufficient to solve the problem. However, the axial probeconfirmed that the pinion was moving axially as predicted. Themachine appeared to run acceptably on one helix, but the SSVpicked up when the pinion reached its maximum axial displacementrelative to the gear. Further increase of coupling prestretch was con-sidered undesirable because of a turbine shaft axial excursion of0.16 inch (4.1 mm) away from the gearbox early in the run. Thusanother means of reducing axial misalignment was sought.

Diagnostic Run with Modified Thrust Bearing Clearance

Gearbox thrust bearing clearance was increased for the next run.The resulting combination of gear mesh and thrust-bearingclearance in the gearbox allowed ±0.06 inch (1.5 mm) of pinionaxial motion. Based on the axial alignment review, this wasbelieved to be sufficient to absorb the thermal transients.Following the modification, the machinery protection system didnot trip. Furthermore, the axial probes showed both the pinion andlow-speed gear were pushed axially away from the turbine.Vibration began to increase at roughly the same time after the startof the run as it had previously. However, before the vibrationreached an undesirable level, the pinion and gear were bothobserved to axially shift away from the turbine and the vibrationdecreased. This axial shift was possible only after the gearboxthrust bearing clearance was increased. The shift allowed thepinion to move away from the turbine without tightly binding themesh. This was additional evidence that transient axial misalign-ment had caused the SSV problem. The startup trip did notreappear in subsequent runs either with or without load. The mod-ification permitted the commissioning process to continue.

DIAGNOSIS OF AXIAL ALIGNMENT PROBLEMS

A few main points relevant to diagnosing axial alignmentproblems are summarized here. Although some may not bepractical in a particular situation, it would be prudent to considerthese checks if the usual diagnostic process does not identify theproblem. At a minimum, steps such as reviewing axial alignmentand checking predicted torsional frequencies should be taken. If anaxial alignment issue is found, the options described in theexamples and in the following section should be helpful.

• Review axial alignment calculations, installation results, andrelated factors. This includes thrust bearing clearances, gearsetbacklash, coupling stiffness, coupling prestretch, and distancebetween shafts. In particular, ensure coupling axial thermal growthis included in axial alignment calculations.

• Review predicted torsional natural frequencies for a match withobserved response frequencies. As seen in the examples, there isoften a correlation between measured gearbox response frequen-cies and predicted torsional frequencies. Spectral data with goodresolution taken during the vibration event are required.

• Evaluate the possibility of oil whirl or oil whip if fixed geometrybearings are used. These are complicated phenomena beyond thescope of this paper, but for most gearboxes a few basic observa-tions using a waterfall plot are sufficient for a first evaluation. Oilwhirl usually can be ruled out if the frequency of shaft motion doesnot remain at about 40 to 45 percent of running speed as speedincreases. On a waterfall plot oil whip looks like oil whirl initiallybut instead of continuing to track running speed it locks onto aresonant frequency and remains at that frequency as speedincreases.

• Monitor pinion and gear relative axial positions in conjunctionwith vibration during the test run. If there is no existing axial probeprovision, one can be added. In Examples 1, 2, and 4, an eddy-current proximity probe at the pinion allowed direct verification ofpinion axial movement. It is not necessary to have a high-qualityaxial probe target area since the purpose is to measure relativelylarge rotor movements using a direct current (DC) signal. Pinionand gear motion should be compared to the axial backlash in thegearset. Timing of these motions can be compared with changes invibration.

• Verify coupling (or enclosure) temperature rise and compare tothe expected value.

• Machinery protection system trips can hinder diagnosis. Theequipment suppliers involved can be consulted for ways to reduceresponse or for other options to prevent unnecessary trips thathinder diagnosis. For instance, in Example 1 a temporary gearboxbearing change reduced response during the diagnostic process.

• Perform test runs with varied loading. Lighter load reduces themesh centering force and may make the problem worse. Heavierload increases mesh centering force and makes the gearset moreresistant to external axial forces.

• Perform a test run with the driven equipment uncoupled. Inaddition to removing load, this can remove one source of axialforce on the gearset. This test is most productive if the gearboxoutput shaft does not have a thrust bearing.

DESIGN AND ANALYSIS TOPREVENT AXIAL ALIGNMENT PROBLEMS

The axial alignment of a turbomachinery train including adouble-helical gear unit should address all operating modes of thetrain. This requires consideration of both steady-state equilibriumthermal growth and transient thermal growth during startup,shutdown, or unloading cycles that may be encountered in fieldoperation. Test conditions must also be considered, particularly iftesting includes running at essentially no torque load. If properlyapplied at the design stage the following observations should helpprevent future problems. They may also be helpful in correctingproblems in existing trains.

• Coupling prestretch can help accommodate thermal growth inthe train. However, care must be taken to ensure the flexibleelements of the coupling will not be damaged. There may be axialposition excursions in both directions during startup thermal tran-sients. A compromise coupling prestretch value may be required toaccommodate both transient and steady-state axial alignment con-ditions.

• Coupling thermal growth should be included in prestretch cal-culations, as well as thermal movements of the shaft ends andcasings of connected machines.

• Coupling enclosures should be properly designed to preventexcessive coupling heating and thermal growth.

• Coupling flexible element axial stiffness can be lowered toaccommodate more axial growth without generating large axialforces.


101

• Thermal growth analysis of involved components must includeconsideration of maximum and minimum transient thermal states,such as when rotor components are hot and casings are cool. Thiswill allow a comprehensive axial alignment chart to be produced.

• A thrust bearing can be situated in the gearbox on the shaft towhich the largest external force is applied. This can prevent theaxial force from reaching the mesh. However, the thrust bearingmust be capable of absorbing the axial force. It should be kept inmind that purchasing specifications often dictate gearbox thrustbearing location. For example, API 613, Fifth Edition (2003),states that a double-helical gearbox shall have a thrust bearing onthe low-speed shaft unless specified otherwise by the purchaser.

• The gearbox thrust bearing clearance can be increased. Thegearset can even be allowed to float axially without a thrustbearing to absorb motion of connected equipment. However, thisoption may not be possible for some train layouts. Also, there canstill be a problem if the connected machines can exert axial forceon both gearbox shafts.

• Increased backlash can accommodate more mesh axial offset,but excessive backlash can reduce the load capacity of the gearingby excessively thinning the teeth. It can also increase the severityof transient torsional vibration in some trains.

• Casing fixation locations of connected machines can be alteredto reduce shaft end excursions.

• Changes of startup sequences or other thermal managementschemes can be considered.

SUMMARY

A double-helical gearbox relies on transmitted torque togenerate mesh axial forces that hold the pinion centered on thegear. If an external axial force is larger than the mesh force, it canpush the pinion and gear off center with one another. This is mostlikely to occur in unloaded trains such as a generator drive prior toapplication of electrical load or a compressor drive in a no-loadstring test. The external force is most commonly caused by axialmisalignment of the connected equipment. This misalignment maybe due to an error in design or installation, or it may be a temporaryeffect caused by transient thermal growth.

Axial alignment errors and thermal transients in turbomachinerytrains can produce axial displacements, which must be absorbed bythe coupling flexible elements. If the axial displacements are large,and the flexible element stiffness is high, the resulting forces andaxial motion of the pinion relative to the gear can be sufficient toconsume all the axial clearance in the gearset and bind the mesh,essentially producing a gearset running without backlash.

Four field examples have been presented where axially boundgearsets have excited subsynchronous torsional resonances. Themechanical coupling of torsional and lateral motion in a gearboxresults in the torsional vibration producing a lateral response.Depending on the stiffness of the bearings, varying levels of lateralresponse are measured, but as shown in the examples these vibra-tions can be sufficient to trip the machinery protection system.

Several diagnostic checks are available to help distinguish axialalignment problems from other sources of vibration. They rangefrom the relatively easy tasks of reviewing axial alignment andpredicted torsional frequencies to more difficult and expensivechecks such as special test runs. The diagnostic value of an axialprobe on the pinion has been well demonstrated. In Examples 1, 2,and 4, this probe allowed direct verification of the phenomena. Theverification allowed the involved parties to agree on a solution thatwould allow successful completion of the testing or installation ofthe equipment without compromising the operational performanceof the machinery.

Fortunately, options are available at the design stage to mitigatethe effects of axial thermal growth. Careful axial alignmentanalysis and judicious use of design parameters such as thrust

bearing location, thrust bearing clearance, gearset backlash, andcoupling prestretch can help prevent axial alignment problems.System designers, gear unit suppliers, and others such as couplingsuppliers should work together to ensure that the equipment willoperate satisfactorily.

Ultimately, concerns about transient thermal growth and subse-quent impact on operating alignment between gear elements are asystem-level issue. A comprehensive analysis should be performedto ensure the design respects the constraints of each systemcomponent. Good communication and cooperation at the designstage are crucial, particularly if there are large thermal effects orother special circumstances.

APPENDIX A—SUPPLEMENT TO EXAMPLE 1

Discussion of Physical Events DuringDiagnostic Run with Axial Probe on Pinion

The following paragraphs provide a physical interpretation ofthe problem deduced from position measurements, vibration data,and mechanics of the machine components. Refer to the sectiontitled “Diagnostic Run with Axial Probe on Pinion.” Figure A-1depicts the location of the axial probe on the pinion blind end,access panel.

Figure A-1. Photograph of the Axial Probe Installation in Example 1.

During axial alignment of the train, the bull gear was centered inits thrust bearing clearance and the pinion was centered relative tothe gear. At that point all components were at ambient temperatureand there was a small prestretch in the coupling flex elements[nominally 0.018 inch (0.5 mm)].

The train was started and idled as usual as the equipment beganto heat up. At 8:42 the compressor speed was increased from 9000to 11,844 rpm. This caused two things to happen. First, as speedincreased, the torque absorbed by the compressor increasedslightly and so did the mesh centering force. This caused the pinionto move axially, changing coupling displacement to maintain equi-librium between the mesh centering force and couplingcompression force. This is shown in Figure 12 as a downwardmovement of the position trace of about 1 V, indicating the pinionmoved about 0.005 inch (0.13 mm) toward the compressor (theprobes provide 200 mV/mil displacement). Because higher torque(and therefore an increased mesh centering force) caused thepinion to move closer to the compressor, it can be inferred thatthermal growth prior to 8:42 had pushed the pinion away from thecompressor and moved it off center relative to the gear. Figure 13shows that the low-speed gear simultaneously moved about 0.0015inch (0.038 mm) away from the compressor. Because the axialprobe on the low-speed gear was oriented in the opposite direction,downward motion of the gear position voltage indicates motionaway from the compressor.

The second occurrence at 8:42 was the temperature of the com-pressor rotor and case began rising substantially again. As the


compressor came to MCOS, the various parts of the high-speedshaft line began to quickly heat and grow axially. The thermallymassive case did not heat up as fast, and took longer to growaxially. The net result was that the compressor shaft end movedaxially toward the gearbox. The thermal differential growth of thehigh-speed rotor string from the mesh center to the compressorthrust bearing was 0.066 inch (1.7 mm) (0.066 inch = 0.042 inchcompressor shaft + 0.015 inch coupling + 0.009 pinion shaft).Furthermore, the low-speed gear was undergoing a similar thermaltransient that was pushing the gear away from the gas turbine andagainst the gear thrust bearing. This also served to compress thehigh-speed coupling and increase the axial load on the pinion.Normally (i.e., when under load), the torque transferred and thedouble-helix centering force would have kept the two gearelements in axial alignment, and the transient axial growth wouldhave been absorbed by the flexible elements of the coupling. Butthat did not happen in this test since torque load was near zero.

Recalling the discussion related to Figures 5 and 6, under light-load conditions, the maximum mesh axial centering force wasmuch less than at normal operating torque, approximately 100 lbf(445 N) in this instance. Based on the axial stiffness of thecoupling flex elements, K = 4000 lbf/in (700 kN/m), only 0.025inch (0.6 mm) of displacement was needed to produce in excess of100 lbf (445 N). The calculated 0.066 inch (1.7 mm) thermalgrowth and the 0.018 inch (0.5 mm) prestretch of the couplingresult in 0.048 inch (1.2 mm) of compression in the flex element,which should produce 192 lbf (855 N) axial load if the meshremained centered. This was sufficient to slide the pinion axiallyalong only one helix while the teeth of the opposite helix would notbe making contact (as shown in Figure 5).

From 8:42 to 8:57 the pinion position could be seen movingsmoothly away from the compressor as shown by the steady rise inthe pinion position voltage in Figure 12. During that time, speedand torque were constant and the pinion was moving axially alongwith the compressor shaft end. Because the pinion was movingaxially relative to the gear, it was clearly running on one helix asshown in Figure 5. Note that there was practically no SSV at thistime, indicating that the gearset can run smoothly on one helix.

At 8:57:36 (where the data transition from light to dark), themoment significant SSV first appeared as shown in Figure 11, thepinion axial motion abruptly stopped as shown by the sharp dis-continuity in Figure 12. This suggests that an obstacle hadpresented itself to restrict that motion. Based on distance traveledand the fact that there were no other obstructions to stop the pinion,it was clear that the pinion stopped moving axially because itreached a bound condition with the gear (as shown in Figure 6).

For about an hour after 8:57:36 the data show the mesh wasbound tightly. When the pinion stopped moving axially, the com-pressor thermal transient had not been completed and thecompressor shaft end position had not reached its full excursiontoward the gearbox. For the next hour the pinion axial position wasvery nearly constant even though the compressor shaft end wouldhave still been moving, initially growing toward the gearbox, thenpulling away as compressor case growth caught up with the rotor.Once the pinion reached the end of its travel in the mesh clearance,any additional compressor shaft growth had to be accommodatedby increasing axial compression of the high-speed coupling.Therefore during the hour that the pinion did not move, it musthave been forced tightly in the mesh. The SSV continued undi-minished throughout this hour period.

After one hour, corresponding to the time when the SSV beganto change and the data transition from dark to light in Figures 11 to13, the pinion began to return to its initial axial position. Thischange corresponds to the compressor casing warming and pullingthe compressor shaft end away from the gearbox. At thermal equi-librium, the compressor case provides 0.030 mil (0.8 mm) of axialgrowth in the opposite direction from the initial shaft growth. Eventhough the compressor shaft had begun moving back away fromthe gearbox before the end of the hour, the pinion did not move

axially relative to the gear until the coupling compressiondecreased sufficiently for the coupling axial force to reach equilib-rium with the axial mesh force. When this equilibrium was reachedthe pinion began moving with the compressor shaft end as itmoved back toward the compressor. At this point the pinion wasbecoming unbound and returning to the condition shown in Figure5. Shortly after the pinion began to move back toward the com-pressor, the discrete SSV components vanished, leaving thesubsynchronous spectrum clean.

When speed was decreased from 11,840 to 9000 rpm (at 11:00),the pinion position moved abruptly away from the compressor,momentarily returning back to the same axial position where itstopped moving axially during initial startup. This happenedbecause the speed decrease reduced the torque transmitted acrossthe gearset and the corresponding gear mesh centering force,allowing the remaining coupling axial force to push the pinionback to its bound position. During this period, SSV briefly reap-peared and then subsided.

It is important to note that numerous additional runs and stopcycles had produced many data sets for review. Careful observa-tion of the shutdown behavior indicated that the vibration spikes(at 40 and 80 Hz) were not tracking with the 1� speed as the traindecreased in speed. Instead, as train speed decreased, the 40 and 80Hz components remained at a fixed frequency as shown in severalof the preceding vibration plots. This behavior is not consistentwith an oil whirl. However, it does appear consistent with an exci-tation of the first and second torsional frequencies.

REFERENCES


Carter, D. R., Garvey, M., and Corcoran, J. P., 1994, “The Bafflingand Temperature Prediction of Coupling Enclosures,”Proceedings of the Twenty-Third Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 115-123.

Drago, R. J., 1988, Fundamentals of Gear Design, Boston, Mas-sachusetts: Butterworth-Heinemann.

Dudley, D. W., 1984, Handbook of Practical Gear Design, NewYork, New York: McGraw-Hill.

Dudley’s Gear Handbook, Second Edition, 1991, Townsend, D. P.,Editor, New York, New York: McGraw-Hill.

Hudson, J. H., 1992, “Lateral Vibration Created by TorsionalCoupling of a Centrifugal Compressor System Driven by aCurrent Source Drive for a Variable Speed Induction Motor,”Proceedings of the Twenty-First Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 113-123.

Iwatsubo, T., Arii, S., and Kawai, R., 1984, “The Coupled LateralTorsional Vibration of a Geared Rotor System,” Proceedings ofthe Third International Conference on Vibrations in RotatingMachinery, IMechE, C265, pp. 59-66.

Mancuso, J., July 24, 1986, “Disc vs Diaphragm Couplings,”Machine Design, pp. 95-98.

Nicholas, J. C., Barrett, L. E., and Leader, M. E., 1980,“Experimental-Theoretical Comparison of Instability OnsetSpeeds for a Three Mass Rotor Supported by Step JournalBearings,” Journal of Mechanical Design, ASME, 102, pp.344-351.

Pennacchi, P. and Vania, A., 2004, “Model-Based Analysis ofTorsional and Transverse Vibrations of Geared RotatingMachines,” Proceedings of the Eighth International Confer-ence on Vibrations in Rotating Machinery, ImechE, C623, pp.251-260.


103

Zirkelback, C., 1979, “Couplings—A User’s Point of View,”Proceedings of the Eighth Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 77-81.

BIBLIOGRAPHY

Eshleman, R. L., 1997, “Torsional Vibration of Machine Systems,”Proceedings of the Sixth Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 13-22.

ACKNOWLEDGEMENTS

The authors would like to thank the management of Dresser-Rand, Lufkin, and Petronas Carigali for their support in publishingthis paper. The authors also would like to acknowledge the supportof Mr. Ed Turner (Dresser-Rand) and Mr. Craig Kujawa (SolarTurbines) in collecting data, Mr. Rob Kunselman (Dresser-Rand)for creating graphics, and Mr. Richard Pilsbury (Dresser-Rand)and Mr. Mark Winthrop (Lufkin) for assisting in the modificationsrequired to perform the diagnostic testing.


Lars Brenne is currently a Staff Engineerat the R&D Department (RotatingEquipment) of Statoil ASA, in Trondheim,Norway. He has been involved in wet gascompression technology studies, tests, anddevelopment. Dr. Brenne began his careerwith Aker Kværner where he worked in theMechanical Engineering Department(Rotating Equipment). His primary respon-sibility was pump systems for the JotunFPSO. Upon completion of his graduate

studies in 2002, he joined Statoil’s Research Department.Dr. Brenne received his M.S. degree (Mechanical Engineering,

1997) from the Norwegian University of Science and Technology,and his Ph.D. degree (Thermal Energy, 2004) from the NorwegianUniversity of Science and Technology.

Tor Bjørge is currently a Staff Engineerat the R&D Department (RotatingEquipment) of Statoil ASA, in Trondheim,Norway. He has been involved in activitiescovering wet gas compression, compressortransient response, and NOx emissionsfrom gas turbines.

Dr. Bjørge received his M.S. degree(Mechanical Engineering, 1981) from theNorwegian University of Science andTechnology. Upon graduation, he joined

the Norwegian University of Science and Technology where heworked at the Engineering Thermodynamics Department. Hereceived his Ph.D. degree (Engineering Thermodynamics, 1988)from the Norwegian University of Science and Technology. Uponcompletion of his graduate studies, Dr. Bjørge worked as anAssociate Professor at the Engineering ThermodynamicsDepartment. His primary responsibility was as a lecturer withinThermodynamics, Heat and Mass transfer, and within research in

the same area. Dr. Bjørge then joined Statoil’s ResearchDepartment. He is a member of ASME.

José L. Gilarranz is currently a SeniorAero/Thermodynamics Engineer withDresser-Rand Company, in Olean, NewYork. He has been heavily involved in thedesign, specification, and use of advancedinstrumentation for development testing ofnew centrifugal compressor components.He began his career with Lagoven (nowPDVSA) and worked for three and a halfyears as a Rotating Equipment Engineer inPDVSA’s Western Division. His primary

responsibility was the evaluation and prediction of the aerothermalperformance of multistage centrifugal compressor packagesutilized by Lagoven in Lake Maracaibo. Upon completion of hisgraduate studies, he joined Dresser-Rand’s Development Engi-neering Group.

Dr. Gilarranz received his B.S. degree (Mechanical Engineer-ing, 1993) from the Universidad Simón Bolívar (Venezuela). Hereceived his M.S. degree (Aerospace Engineering, 1998) and hisPh.D. degree (Aerospace Engineering, 2001) from Texas A&MUniversity. He is a member of ASME and AIAA.

ABSTRACT

This paper presents the results of performance testing of asingle-stage centrifugal compressor operating under wet gas con-ditions. The test was performed at an oil and gas operator’s testfacility and was executed at full-load and full-pressure conditionsusing a mixture of hydrocarbon gas and hydrocarbon condensate.The effect of liquid was investigated by changing the gas-volumefraction between 1.0 and 0.97, which covers the range encounteredby the operator during regular gas/condensate field production inthe North Sea. Other parameters that were evaluated include the

111

PERFORMANCE EVALUATION OF ACENTRIFUGAL COMPRESSOR OPERATING UNDER WET GAS CONDITIONS

byLars BrenneStaff Engineer

Tor BjørgeStaff Engineer

Statoil ASA

Trondheim, Norway

José L. GilarranzSenior Aero/Thermodynamics Engineer

Jay M. KochStaff Engineer, Aero/Thermodynamics

andHarry Miller

Product Manager, Marketing


Olean, New York

compressor test speed, the suction pressure, and two differentliquid injection patterns. During the tests, the machine flowratewas varied from near surge to choke conditions; hence, the evalu-ation covered the entire operating range of the machine. Althoughthe test was primarily intended to evaluate the effects of the wetgas on the thermodynamic performance of the machine, themechanical performance was also investigated by measuring themachine vibration levels and noise signature during the baselinedry gas tests as well as during the tests with liquid injection.

INTRODUCTION

Centrifugal compressor packages utilized for upstream gas pro-cessing often must operate under wet gas conditions in which thefluid handled by the compression package contains a mixture ofliquid and gaseous phases. Typically, the liquid components of themixture are separated from the gas stream before they enter thecompressor by the use of scrubbers and separators locatedupstream of the compressor inlet. These devices are very large andheavy, requiring a large “footprint” (amount of floor space) ascompared to the gas compression package. A compressor with theability to directly handle wet gas without the need for separationequipment is very attractive from an economic standpoint, as itwould drastically reduce the size, weight, and cost of the gas com-pression package. For the case of future subsea compressionsystems, this capability is even more attractive because of the highcosts of deploying a compressor train and all of its associatedequipment under water.

Wet gas compression (WGC) technology represents new oppor-tunities for enhanced, cost-effective production from existing andfuture gas/condensate fields. Many oil and gas operators facefuture challenges in tail-end production, unmanned operation, andimproved recovery from topside and subsea wells. This empha-sizes the need to develop more robust compression systems, whichcan be designed for remote operation in unmanned topside instal-lations, or could be designed for subsea operation for reinjectionand/or transport boosting. The use of this technology for subseaboosting represents a new and exciting application for rotatingequipment, which will allow new gas/condensate field productionopportunities as well as enhanced recovery of existing gas/conden-sate fields and cost-effective production from marginal gas fields.

As mentioned above, these wet gas compression systems couldbe based on the use of a liquid tolerant dry gas compressor, whichcould boost a coarsely separated (via a scrubber) well-stream,however, an even more attractive solution would be the develop-ment of compression systems that can boost the well-streamdirectly. Many research projects and product qualificationprograms are currently underway to develop such a system eitherby modifying existing multiphase pump technology or by the adap-tation of currently available gas compression technologies (Scott,2004). Regardless of the choice of concept, the compressorsolution should be able to tolerate liquid ingestion for an extendedtime without failure. For the case of subsea applications, the highcost associated with the retrieval of the compressor from the seafloor accentuates the importance of a reliable design.

The work presented herein served as an initial test to verify themultiphase boosting capabilities of a centrifugal compressor aswell as to provide an oil and gas operator with data to compare theperformance of this technology with other available wet gas com-pression concepts. It is important to state that the test compressorused for this investigation was not originally designed for wet gasboosting, nonetheless it provided an economically viable test bedfor centrifugal compressor technology.

DESCRIPTION OF TEST VEHICLE

The test vehicle used for this work was a barrel-type, single-stage compressor, manufactured by the coauthors’ company. Saidcompressor was equipped with a high-head impeller, with adiameter of 0.384 m (1.26 ft), and a design flow coefficient of

0.02380. The compressor was originally designed to handle aninlet flow of 4332 Kg/min [2167 Am3/hr (76,526.88 ft3/hr)] of dryhydrocarbon gas (molecular weight of 18.49), with an inletpressure of 130.2 bar (1888.4 psi)and a discharge pressure of 161.8bar (2346.7 psi). Figure 1 shows a cross-section of the test com-pressor; the inlet and discharge nozzles are located at a 45 degreeangle with respect to the top dead center of the machine. Theoriginal design of this machine, which dates to 1986, was notintended for wet gas service, and hence the internal geometry wasnot optimal. Nevertheless, in order to increase the reliability of themachine, the original rotor design was modified to accommodatean electron-beam welded and vacuum furnace brazed impeller witha shrink fit to the shaft. The rest of the machine remained the same(i.e., casing and stationary components). This compressor wasequipped with a vaneless diffuser configuration.

Figure 1. Cross-Section of the Test Compressor.

The compressor was driven by a 2.8 MW synchronous electricmotor, through a speed increasing gearbox, with a gear ratio of6.607. A variable speed drive permitted the operation of the com-pressor within its speed range of 6000 to 13,000 rpm.

The test compressor is utilized in the coauthor’s closed loop testfacility, and was equipped to simulate the conditions expected fora centrifugal compressor operating under wet gas conditions.Figure 2 shows a schematic diagram of the test loop that was usedfor the evaluations. The major components of the test loopincluded a scrubber, the test compressor, a pump, a cooler, and aliquid injection module (mixer). The scrubber, here called guardseparator, was used to separate the dry gas (saturated hydrocarbonmixture) from the liquid (hydrocarbon condensate) in order topermit accurate measurement of the massflow of each stream(liquid and gas). The liquid stream was measured with a Coriolisflowmeter while the gas stream was measured with a calibratedorifice plate.

Figure 2. Schematic Diagram of the Wet Gas Test Loop.


Variable SpeedElectric Motor(MW)

Gas Flow

2 Phase Flow

Condensate

Gearbox

Compressor(rpm)

GuardSeparator

Filter

p, T, Pdyn, Q p, T, Q

p, T, Pdyn

Cooler/Heater

LiquidInjectionModule

..

A screw pump was used to handle the hydrocarbon condensateexiting the guard separator in order to increase its pressure toadequate values for injection into the gas stream. The liquidinjection module permitted the introduction of the condensate intothe gas upstream of the compressor inlet. The liquid could be intro-duced into the gas with two different patterns: as a uniformlydistributed droplet mist or as a liquid film uniformly coating thewetted surface of the inlet pipe. This was obtained by injecting theliquid through specially designed nozzles for the case of dropletflow or through a circumferential slit for the case of liquid filmflow. This required pressurization of the liquid phase and the useof a cooler/heater to regulate the temperature of the condensate toassure that it was injected into the gas stream at the same temper-ature as the gas. The liquid temperature was measured usingcalibrated PT-100 elements and static pressure measurements weremade using calibrated pressure transducers.

The temperature and static pressure of the gas stream enteringand exiting the compressor was measured using calibrated thermo-couples and pressure transducers installed in the pipeline inaccordance with the recommendations of ASME PTC 10 (1997).The measurements made on the gas exiting the compressor corre-sponded to that of the wet gas mixture, while the measurements ofthe liquid and dry gas streams at the inlet were made independentlyfor each phase. The compressor discharge piping had a 45 degreeslope upwards and also a diameter change from 0.203 to 0.305 m(.666 to 1.0 ft). Due to the risk of liquid accumulation in thispiping, liquid hold up was monitored using a gamma ray densito-meter. The same measurement was also performed upstream of theantisurge valve to detect any liquid accumulation. If this wasdetected, the test would have to be stopped due to the risk ofinjecting a liquid slug into the compressor.

The gas composition of the hydrocarbon mixtures utilized as testgas and liquid are shown in Table 1. The gas corresponds to an“export quality” lean hydrocarbon mixture (composed mostly ofmethane), which is typically commercialized for the Europeanmarket, while the liquid corresponds to the condensate receivedfrom the Sleipner field, which lies in the Norwegian North Sea.

Table 1. Gas and Liquid Compositions (Compositions Shown asMolar Percentages).

Based on the volumes of gas and liquid and the filling tempera-ture and pressure of the test loop, the composition of the gas andliquid streams and their associated thermodynamic states wereevaluated using a thermodynamic property package in combinationwith the measured pressure and temperature. The thermodynamicproperty package is a precursor of a commercially available gasproperty package, which allows the combination of reliable fluidcharacterization procedures with robust and efficient algorithms tomatch fluid descriptions to experimental pressure, volume, andtemperature (PVT) data. To increase the data accuracy, the gas andliquid densities were determined with a commercially availablethermodynamic calculation software. The thermodynamic data at

the inlet and discharge of the compressor were obtained anddisplayed online for each one of the measurement series and storedtogether with all of the measured and calculated parameters in onedata file.

In addition to the instrumentation described above, the test loopwas also equipped with dynamic pressure transducers, installed atthe inlet and discharge piping of the compressor. These transducerswere utilized to measure the fluctuating pressure components at theinlet and discharge of the machine. The signals from these instru-ments were displayed and recorded during the test in the form offrequency spectra, which enabled the test engineers to monitor thepressure signals in the process loop directly upstream and down-stream of the compressor. These measurements were correlated tothe noise level of the machine and permitted the comparison of thisparameter while the machine operated under dry and wet gas con-ditions. In addition, the probes could be used to assist in correlatingany possible subsynchronous rotor vibrations with pulsations in thegas stream (Marshall and Sorokes, 2000).

In order to minimize the complexity of the instrumentation forthe initial test, and since the primary mission was the study of thethermodynamic performance change with liquids introduced to thegas stream, it was decided to forego installation of additionalinstrumentation, which would have given more insight into themechanical reactions taking place in response to the various liquidloadings. As such, the installation of strain gauges on the impellerswith their attendant installation complexities, as well as convertingto active magnetic journal and thrust bearings, were held off forfuture test programs. The installation of a magnetic bearing shaftexciter (Moore, et al., 2002) onto the free-end of the compressorshaft would have provided a means to assess any variation of therotor natural frequencies, as well as to determine any change to therotordynamic stability of the compressor due to the addition ofliquids into the gas stream. The use of this device was also left toa future test program.

THEORETICAL FOUNDATION

Single-Phase (Dry Gas) Performance

For a centrifugal compressor the primary variables of interestare the amount of flow delivered, the pressure rise produced, andthe required power. The pressure rise and the efficiency of the gascompression are normally nondimensionalized to allow compari-son of different geometries and operating conditions (Stahley,2000; Colby, 2004). The polytropic compression process isselected for industrial compressors as it is better suited to handlethe wide range of gases used in industry (Schultz, 1962). Theequations for polytropic head coefficient, flow coefficient, effi-ciency, and power are shown below (ASME PTC-10, 1997).

(1)

(2)

(3)

(4)

(5)

PERFORMANCE EVALUATION OF A CENTRIFUGAL COMPRESSOR OPERATING UNDER WET GAS CONDITIONS 113

Composition in loop at 70 bar and 35 oC

Component Export gas Condensate Gas phase Liquid phase Total

N2 0.756 0.854 0.089 0.531 CO2 1.828 1.524 0.956 1.284 C1 90.373 90.933 25.920 63.474 C2 6.074 4.103 4.489 4.266 C3 0.844 0.024 0.341 0.955 0.600

I-C4 0.045 1.059 0.124 0.651 0.347 N-C4 0.064 7.690 0.654 4.632 2.334 I-C5 0.006 10.373 0.481 6.662 3.091 N-C5 0.006 12.015 0.454 7.856 3.581

C6 0.004 20.387 0.348 13.897 6.071 C7 18.616 0.137 12.932 5.541 C8 9.487 0.035 6.637 2.824 C9 4.392 0.008 3.084 1.307

C10+ 15.957 0.005 11.239 4.750 Mole weight 17.77 98.222 18.483 73.52 41.728

μppW

U= 2

UD N

=⋅ ⋅π60

( )Wn

np pp =

−⋅ ⋅ − ⋅

1 2 2 1 1ν ν

nn

n

pp

=⎛⎝⎜

⎞⎠⎟

⎛⎝⎜

⎞⎠⎟

1

1

2

1

1

2

νν

φπ

=⋅ ⋅ ⋅

�Q

DN1

6032

(6)

(7)

This formulation assumes a single-phase gas. If the compressorinlet stream contains both gas and liquid (i.e., wet gas), theseequations must be modified. The primary area of interest is the def-inition of polytropic work, which impacts the polytropic efficiencyand polytropic pressure coefficient.

Two-Phase (Wet Gas) Performance

The calculation procedure to estimate the performance of amachine operating under wet gas conditions is not described in anystandard, as is the case for dry gas. However, the thermodynamicapproach used for a single-phase gas, as stated above, can still beapplied to a two-phase fluid. For the case of the single fluid model,the required modifications are shown below:

(8)

(9)

where the two-phase specific volume is based on the homogeneousmodel:

(10)

The gas-volume fraction (GVF) is defined by:

(11)

The fluid power was derived from electric power readings, usingadequate calibration curves available from previous testing.

A different approach would be to consider a two fluid modelwhere each phase is treated individually. The polytropic head isthen calculated as:

(12)

where the fluid quality is defined as:

(13)

For the case presented in this work, the phase transitioncomponent was small due to a low pressure ratio through themachine and stable fluids. However, for higher pressure ratio mul-tistage compressors, the phase transition contribution cannot beneglected. For the case at hand however, the effect of phase transi-tion is only accounted for in the above expressions by changes inthe value of the polytropic exponent due to a lower discharge tem-perature.

For this work and both of the performance calculation modelspresented above, the two-phase head coefficient and two-phaseflow coefficient may be expressed as:

(14)

(15)

The efficiency for both cases is then expressed as:

(16)

where Pcs is the specific compressor shaft power, defined as thepower consumed by the compressor per unit mass of wet gas.

The compressor two-phase efficiency calculated with the use ofthe single fluid model was found to be virtually equal to the onecalculated via the two-phase fluid model (with a maximumdeviation of 0.8 percent), so the results described in this work willbe based on the single fluid model.

The performance of a wet gas compressor must be compared tothe alternative, which would involve the separation of the fluidstream into individual phases (dry gas and condensate), and thesubsequent boosting of the streams in separate compressor andpump units.

TEST PURPOSE AND VARIABLES

The wet gas testing presented in this work had several objec-tives. The first objective was to investigate the heat transfer ratebetween the gas and liquid condensate through the compressorand determine the state of thermal equilibrium. Anotherobjective was to evaluate the compressor performance (powerconsumption, pressure ratio, and temperature ratio) anddetermine the effects of directly handling a wet gas mixture asopposed to dry gas compression. The impact of liquid ingestionon the compressor mechanical behavior and the pressure pulsa-tions in the loop was also of interest, as was the liquid tolerancecapacity and robustness of the compressor. Finally, the testingwould create a foundation to evaluate the benefits and/ordrawbacks of centrifugal compression technology as opposed toother multiphase boosting concepts.

To achieve the test goals, the performance of the machine wasevaluated under several combinations of key parameters such assuction pressure, flowrate, rotational speed, gas-volume fraction,and liquid injection pattern following the data presented in Tables2 and 3. The test program was completed in a time frame of aboutfour weeks, during which the machine accumulated about 300hours of operation under wet gas conditions.

Table 2. Key Test Parameters with Range of Variation.


ηppW

h h=

−2 1

( )P m h h= ⋅ −� 2 1

nn

nTP

pp

TP

TP

=⎛⎝⎜

⎞⎠⎟

⎛⎝⎜

⎞⎠⎟

1

1

2

1

1

2

νν

( )Wn

np pp TP

TP

TPTP TP=

−⋅ ⋅ − ⋅

1 2 2 1 1ν ν

( )νρ ρTPgGVF GVF

=⋅ + − ⋅

1

1 1

GVFQ

Q Qg

g=

+

�

� �1

xm

m mg

g l=

+

�

� �

( ) ( )W xn

n

R

MWZ T

p

px p pp TP

gl

nn

= ⋅−

⋅ ⋅ ⋅ ⋅⎛⎝⎜

⎞⎠⎟ −

⎛

⎝

⎜⎜⎜

⎞

⎠

⎟⎟⎟

+ − ⋅ ⋅ −

−

10

11 1

2

11 1 2 11

1 1

1

ν

μννp TP

g

TP

p TPGVF

W

U= ⋅ ⋅1

1

12

φπTPTot

N

Q

GVF D=

⋅ ⋅ ⋅ ⋅

�1

1 6032

ηp TPp TP

cs

W

P=

Key Test Parameter Values Units

Suction Pressure 30 70 bar

Machine Speed 9651 10723 rpm

Gas Volumetric Flowrate

1600 1800 2000 2200 2400

Am3/hr

Gas-volume Fraction

1.0000 0.9994 0.9950 0.9900 0.9800 0.9700

N/A

Liquid Injection Pattern Uniform Droplet Fluid Film N/A

Table 3. Test Matrix Agenda.

The quality (x) of the wet gas mixture being injected into thecompressor was dependent on the predefined gas-volume fractionas well as the suction pressure at which the test was beingexecuted. Table 4 presents the values of quality for each GVF usedfor the tests for suction pressures of 30 and 70 bar (435.1 and1015.3 psi).

Table 4. Quality of the Wet Gas at the Compressor Inlet.

RESULTS AND DISCUSSION

Single-Phase (Dry Gas) Performance

Prior to the introduction of liquids into the test loop, the ther-modynamic performance of the machine was evaluated toestablish the baseline performance while it was operating undersingle-phase (dry gas) conditions (tests 1, 2, and 3 of Table 3).This baseline would be used for comparison with the resultsobtained during the operation of the machine under wet gas con-ditions. In addition, the baseline performance would be used toevaluate if the injection of liquids during the multiphase testinghad produced any noticeable effects (performance changes) afterthe tests were concluded. This would be done by running anotherseries of dry gas performance tests and comparing the results tothe initial baseline.

Figure 3 shows the results of the single-phase performance teststhat were executed before and after the evaluation with two-phase(wet gas) flow. As seen in the figure, the performance levels of themachine (polytropic head coefficient and efficiency) haveremained unchanged, that is, there is no evidence to suggest thatthe ingestion of liquid produced any significant variation in themachine’s performance levels. This implies that the compressorflowpath was not subjected to any significant damage during thewet gas tests. A boroscopic inspection of the inlet and impeller eyeareas executed after the tests confirmed that there was no evidenceof internal damage.

Two-Phase (Wet Gas) Performance

Figures 4, 5, and 6 show the performance of the compressorexposed to liquid with up to 3 percent of the inlet volume flow(GVF of 0.97). The wet gas tests are shown together with dry gasresults for comparison. Figure 4 presents a comparison of the com-pressor performance at two different speeds, while operating at asuction pressure of 70 bar, with the liquid being injected with auniform droplet pattern. Figure 5 shows the performance of thecompressor at two different suction pressures [p1 = 30 and 70 bar

Figure 3. Single-Phase Thermodynamic Performance Evaluation.Baseline Versus After-Test Conditions.

(435.1 and 1015.3 psi)], while operating at the same speed (9651rpm) and with the same liquid injection pattern (uniform droplet).Figure 6 shows a comparison of the machine performance for bothliquid injection patterns (droplet and fluid film), with the machineoperating at the same suction pressure [70 bar (1015.3 psi)] and thesame speed (9651 rpm).

Figure 4. Two-Phase Thermodynamic Performance Evaluation.Effects of Test Speed (p1 = 70 bar, Droplet Injection Pattern).

Figure 5. Two-Phase Thermodynamic Performance Evaluation.Effects of Suction Pressure (9651 rpm, Droplet Injection Pattern).


Test Number

Suction Pressure

Machine Speed Gas Flowrate

Gas-Volume Fraction

Liquid Injection Pattern

1 30 9651 All 1.0 N/A 2 70 9651 All 1.0 N/A 3 70 10723 All 1.0 N/A 4 70 9651 All All Droplet

5 70 9651 All 1.0, 0.995, 0.98 2200 Am3/hr :All Fluid Film

6 70 10723 All 1.0, 0.995, 0.98 2200 Am3/hr :All Droplet

7 30 9651 All All (*) Droplet 8 30 9651 All 1.0 N/A

(*) The test point at 2400 Am3/hr and GVF = 0.97 was not completed due to test loop limitations.

GVF Quality at 70 bar Quality at 30 bar 1.0000 1.0000 1.0000 0.9994 0.9931 0.9824 0.9950 0.9454 0.8699 0.9900 0.8959 0.7706 0.9800 0.8100 0.6254 0.9700 0.7377 0.5218

Head Coefficient - Predicted Head Coefficient - Baseline

Head Coefficient - After Test Efficiency - Predicted

Efficiency - Baseline Efficiency - After Test

10273 RPM

9651 RPM

No

rma

lize

d P

oly

tro

pic

He

ad C

oe

ffic

ien

t

Normalized Flow Coefficient

Efficiency

HeadCoefficient

No

rma

lize

d P

oly

tro

pic

Eff

icie

ncy

0.00

0.20

0.40

0.60

0.80

1.00

1.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

0.60 0.70 0.80 0.90 1.00 1.10 1.20 1.30 1.40 1.50 1.60

10723 RPM

9651 RPM

GVF_1.00 GVF_0.995 GVF_0.99 GVF_0.98 GVF_0.97


9651 RPM

10723 RPM

0.00

0.20

0.40

0.60

0.80

1.00

1.20

0.7 0.8 0.9 1.0 1.1 1.2 1.3

Normalized Two-Phase Flow Coefficient

0.00

0.20

0.40

0.60

0.80

1.00

1.20



p1 = 70 bar

p1 = 30 bar

0.00

0.20

0.40

0.60

0.80

1.00

1.20

0.7 0.8 0.9 1.0 1.1 1.2 1.3


0.00

0.20

0.40

0.60

0.80

1.00

1.20

Figure 6. Two-Phase Thermodynamic Performance Evaluation.Effects of Liquid Injection Pattern (9651 rpm, p1 = 70 bar).

As seen in Figures 4 through 6, the redefined polytropic headand flow coefficients [Equations (14) and (15)], valid for two-phase flow, are capable of merging the data from the various wetgas operating conditions with those corresponding to the dry gasoperation. Furthermore, these figures show that the efficiencydrops when the amount of liquid is increased and that this effect ismuch more pronounced at lower pressures. The more pronouncedeffect at lower pressures is due to the increasing density differencebetween the gas and the condensate when the suction pressure isreduced while maintaining a constant GVF. The increasing densitydifference leads to a considerable increase in the mass fraction ofliquid entering the compressor. As shown in Table 4, at a GVF of0.97, the mass fraction of liquid (1-x) increases from 0.2623 at 70bar (1015.3 psi) to 0.4782 at 30 bar (435.1 psi).

The reduction in the machine efficiency as the mass fraction ofliquid increased is due to larger internal losses in the compressor.The test vehicle was not instrumented internally, so the availabledata were insufficient to identify the source of the increased losses.The compressor manufacturer plans to evaluate this issue by per-forming two-phase computational fluid dynamics (CFD)simulations of the compressor at the test conditions. Another wayto provide insight into this phenomenon would be to run additionaltests with internal instrumentation.

Figure 6 shows very little difference between the performance ofthe machine when subjected to uniform droplet and fluid filminjection patterns. Consequently, it is thought that the compressorinlet serves as a mixing element and makes the flow pattern insidethe impeller relatively independent of the injection method. Thiseffect will also be evaluated via two-phase CFD calculations. Thedistance between the point of liquid injection and the center of theimpeller was limited to approximately three times the internaldiameter of the compressor suction nozzle. This was done in aneffort to ensure that the two-phase flow pattern was maintainedfrom the point of injection up to the impeller inlet.

In general, for the figures discussed above, there is a tendencyof a larger departure from a common head coefficient curve at 30bar (435.1 psi) when the deviation between the operating flowrateand the impeller design flowrate increases (GVF < 0.99).

The compressor specific power consumption is shown in Figure7. The data shown in the figure correspond to the tests with thecompressor operating at 9651 rpm, with a suction pressure of 70bar (1015.3 psi), and the liquid being injected in a uniform dropletpattern; nevertheless, the same behavior was observed for the otherwet gas test conditions. As shown in the figure, when liquid isinjected, the required specific power is reduced. To properlyevaluate the specific power associated to wet gas compression,

these data have to be compared with the data that would beobtained if the same amount of liquid and gas were to be trans-ported between the same two pressures (as independent streams).A separate gas and liquid boosting case is included in the figureassuming a GVF of 0.97. As can be seen the specific power con-sumption is lower than the values obtained for wet gascompression. The separate boosting data were based on the samepressure difference. However, when wet gas compression isutilized, the pressure drop in the scrubber may be avoided and therequired pressure boost is less than the one depicted in Figure 7.Furthermore, the possibility of simplifying the compressor systemby avoiding the scrubber and the appurtenant instrumentation mustalso be considered when the system is evaluated.

Figure 7. Two-Phase Thermodynamic Performance Evaluation.Compressor Specific Power Consumption as a Function of GVF(p1 = 70 bar, 9651 rpm, Droplet Injection Pattern).

Figures 8 and 9 show the effects of liquid injection on thepressure and the temperature ratios (discharge/inlet) through thecompressor. Once again, the data shown in the figures correspondto the tests with the compressor operating at 9651 rpm, with asuction pressure of 70 bar (1015.3 psi), and the liquid beinginjected in a uniform droplet pattern; nevertheless, the samebehavior was observed for the other wet gas test conditions. Thepressure ratio increased due to the increased density (andmolecular weight) of the fluid processed by the compressor. Thetemperature ratio slightly decreased for the case of liquid injection.This observation is explained by two mechanisms:

• Increased internal energy of the liquid phase, and

• A certain degree of liquid evaporation has occurred as the liquidpassed through the machine.

These results will also be evaluated by the compressor manufac-turer via two-phase analytical simulations, as they will be of greatimportance when designing multistage machines for wet gasoperation. The change of phase of the liquids inside the compres-sor may cause a mismatch between the subsequent stages of themachine, which may lead to performance shortfalls.

Dynamic Pressure Measurement

Figures 10 through 15 show the frequency spectrum of thedynamic pressure signals measured close to the inlet and dischargeflanges of the machine under three different test conditions (notethat the scales on all plots are the same). Figures 10 and 11 corre-spond to the machine operating at 9651 rpm with a suctionpressure of 70 bar (1015.3 psi) and show the effects of the liquidbeing injected in a uniform droplet pattern. Figures 12 and 13 cor-respond to the same compressor operating condition but with the




Droplet

Film

0.00

0.20

0.40

0.60

0.80

1.00

1.20

0.7 0.8 0.9 1.0 1.1 1.2 1.3


0.00

0.20

0.40

0.60

0.80

1.00

1.20

0.60

0.65

0.70

0.75

0.80

0.85

0.90

0.95

1.00

1.05

1.10

0.8 0.9 1.0 1.1 1.2 1.3



Separate Boosting, GVF = 0.97

Figure 8. Two-Phase Thermodynamic Performance Evaluation.Compressor Pressure Ratio as a Function of GVF (p1 = 70 bar,9651 rpm, Droplet Injection Pattern).

Figure 9. Two-Phase Thermodynamic Performance Evaluation.Compressor Temperature Ratio as a Function of GVF (p1 = 70 bar,9651 rpm, Droplet Injection Pattern).

liquid injection being performed as a fluid film in the periphery ofthe pipe. Finally, Figures 14 and 15 correspond to the same com-pressor speed, with a suction pressure of 30 bar (435.1 psi) and theliquid being injected in a uniform droplet pattern.

Figure 10. Effects of Two-Phase Flow over the Dynamic PressureMeasurements at the Machine Inlet (p1 = 70 bar, 9651 rpm,Droplet Injection Pattern).

Figure 11. Effects of Two-Phase Flow over the Dynamic PressureMeasurements at the Machine Discharge (p1 = 70 bar, 9651 rpm,Droplet Injection Pattern).

Figure 12. Effects of Two-Phase Flow over the Dynamic PressureMeasurements at the Machine Inlet (p1 = 70 bar, 9651 rpm, FilmInjection Pattern).

Figure 13. Effects of Two-Phase Flow over the Dynamic PressureMeasurements at the Machine Discharge (p1 = 70 bar, 9651 rpm,Film Injection Pattern).

It is important to state that the low frequency amplitudes (noise)that are evident on the spectrum plots are due to the fact that thedata acquisition and display system that was used to capture thedata, and to generate these figures, did not have the capability toaverage several fast Fourier transform (FFT) samples. Thishindered the ability to reduce the random noise components of thespectra. Furthermore, the pressure sensors were not installed flushto the pipe wall. They had a small recess [about 25 mm (.98


0.96

0.97

0.98

0.99

1.00

1.01

1.02

1.03

0.8 0.9 1.0 1.1 1.2 1.3



0.988

0.990

0.992

0.994

0.996

0.998

1.000

1.002

1.004

0.8 0.9 1.0 1.1 1.2 1.3



TP-70-D-9651-2200 GVF = 1 and 0.97; X = 1 and 0.738

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Frequency [Hz]

Sig

na

l A

mp

litu

de

Droplet Injection, p1 = 70 bar,

Flowrate = 2200 Am3/hr

1X Frequency = 161 Hz

Blade-Pass Frequency = 3056 Hz

GVF = 1.00

GVF = 0.97

Blade-pass frequency

0

0

TP-70-D-9651-2200 GVF = 1 and 0.97; X = 1 and 0.738

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Frequency [Hz]

Sig

na

l A

mp

litu

de





GVF = 1.00

GVF = 0.97


0

0

TP-70-F-9651-2200 GVF = 1 and 0.98, X = 1 and 0.810

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Frequency [Hz]

Sig

na

l A

mp

litu

de

Film Injection, p1 = 70 bar,




GVF = 1.00

GVF = 0.97


0

0

TP-70-F-9651-2200 GVF = 1 and 0.98, X = 1 and 0.810

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Frequency [Hz]


Sig

na

l A

mp

litu

de

Film Injection, p1 = 70 bar,




GVF = 1.00

GVF = 0.97

0

0

Figure 14. Effects of Two-Phase Flow over the Dynamic PressureMeasurements at the Machine Inlet (p1 = 30 bar, 9651 rpm,Droplet Injection Pattern).

Figure 15. Effects of Two-Phase Flow over the Dynamic PressureMeasurements at the Machine Discharge (p1 = 30 bar, 9651 rpm,Droplet Injection Pattern).

inches)] that may have contributed to the generation of some noisecomponents in the FFTs.

As may be seen in the figures, for the case in which the com-pressor is handling dry gas (GVF of 1.0), the frequency spectrumplots show a variety of peaks in the vicinity of the impeller bladepassing frequency. On the other hand, when liquids are injectedinto the flow (GVF = 0.98 or 0.97), the high-frequency compo-nents of the spectrum vanish. This behavior suggests that theliquids injected into the process gas dissipate the acoustic signals(evident by a reduction in the audible noise level) and dampen thepressure fluctuations. This phenomenon was observed for all of thetest conditions that were evaluated as shown in the test matrixabove (refer to Tables 2 and 3).

Rotordynamic Behavior

Although the main objective of the testing described in thispaper was to establish the effects of wet gas conditions over theaero/thermodynamic performance of a centrifugal compressor,another objective of similar importance was to evaluate the effectsof liquid ingestion over the rotordynamic behavior of the machine.For this, the shaft vibration was measured via eddy currentproximity probes installed at both ends of the machine (i.e., at thedriven and nondriven ends). The test compressor was originallysupplied with a pair of proximity probes at each journal bearing tomeasure the shaft vibration in the horizontal and vertical direc-tions. In addition, the axial position of the shaft was also measuredvia proximity probes at the free end of the machine. The signalsfrom these probes were displayed in the control room and were

linked to the process control system to provide appropriatemachinery protection. Said signals were also connected to a pro-prietary inhouse data acquisition and reduction system, whichpermitted the frequency spectrum of the shaft vibration signals tobe displayed during the tests. These frequency spectra wereutilized in the evaluation of the rotordynamic behavior of themachine by comparison of the spectra obtained under dry and wetgas conditions.

The test compressor had a bearing span of 0.727 m (2.39 ft),with an impeller bore of 0.132 m (5.2 inches), and a journaldiameter of 0.076 m (2.99 inches). The shaft was mounted on tilt-pad journal bearings, each of which had five pads, in the load-onpad configuration. The first natural frequency of the rotorbearingsystem is near 9800 cpm. This mode is well damped and is not acritical speed. The first critical speed of the compressor was deter-mined (analytically) to be between 20,300 to 21,130 cpm. Thisvalue is larger than the maximum speed at which the compressorwould be tested (10,723 rpm), so smooth operation was expectedwithin the operating envelope that would be used for the tests(9651 to 10,723 rpm).

Figure 16 presents the frequency spectra (FFT) of the shaft hor-izontal vibration component, measured at the driven end of themachine, while it was operating at 9651 rpm, with a suctionpressure of 70 bar (1015.3 psi) and handling a volumetric flow of2200 Am3/hr (77,692.27 ft3/hr). The bottom spectrum corre-sponds to operation with a GVF of 1.0 (dry gas); while the topspectrum shows the behavior of the machine while handling atwo-phase gas mixture, with a gas-volume fraction of 0.97, whichrepresents a gas quality (x) of 0.738. For this case, the liquid wasbeing injected into the gas with a uniform droplet pattern. As maybe seen in the figure, the vibration spectra for both the dry gas andthe wet gas compression are virtually the same. This suggests thatthe rotordynamics of the machine remain unaffected by the liquidinjection for the case of the condensate being injected with theuniform droplet pattern. This is due to the fact that if the liquidsare uniformly distributed throughout the gas, they do not produceany significant source of rotor excitation as they pass through theimpeller, nor do they affect the rotor unbalance levels (1�

vibration component). The above similarity was also encounteredwhen comparing the vibration spectra corresponding to GVFvalues of 0.98 and 0.99. The behavior of the machine while it wasoperating at 10,723 rpm, under the same suction pressure andliquid injection mechanism, presented similar characteristics andhence will not be shown.

Figure 16. Effects of Two-Phase Flow over the Vibration Responseof the Machine, Measured in the Horizontal Direction at theDriven End (p1 = 70 bar, 9651 rpm, Droplet Injection Pattern).

Figure 17 presents the frequency spectra of the shaft horizontalvibration component, measured at the driven end of the machine,while it was operating at 9651 rpm, with a suction pressure of 70


TP-70-D-9651-2200 GVF = 1 and 0.97; X = 1 and 0.738

0 100 200 300 400 500 600 700 800 900 1000

Frequency [Hz]

1X frequency (161 Hz)

Droplet Injection, p1 = 70 bar, Flowrate = 2200 Am3/hr

Sig

na

l A

mp

litu

de

GVF = 1.00

GVF = 0.97

0

0

TP-30-D-9651-2200 GVF = 1 and 0.97, X = 1 and 0.53

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Frequency [Hz]

Sig

na

l A

mp

litu

de





GVF = 1.00

GVF = 0.97


0

0

TP-30-D-9651-2200 GVF = 1 and 0.97, X = 1 and 0.53

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Frequency [Hz]

Sig

na

l A

mp

litu

de





GVF = 1.00

GVF = 0.97


0

0

bar (1015.3 psi) and handling a volumetric flow of 2200 Am3/hr(77,692.27 ft3/hr). The bottom spectrum corresponds to operationwith a single-phase (dry) gas, while the top spectrum represents thebehavior of the machine while handling a two-phase gas mixture,with a gas-volume fraction of 0.98 (gas quality of 0.810). For thiscase, the condensate was being injected into the gas with a liquidfilm pattern, which was uniformly distributed around the wettedsurface of the inlet pipe. As may be seen in the figure, the vibrationspectra for the dry and the wet gas cases are also very similar. Thissupports the belief that when the liquids are ingested in a uniformmanner by the compressor, they do not provide a sufficientlystrong source of excitation or unbalance to disturb the rotordy-namic behavior of the machine. The above similarity was alsoencountered when comparing the vibration spectra correspondingto GVF values of 0.97 and 0.99.

Figure 17. Effects of Two-Phase Flow over the Vibration Responseof the Machine, Measured in the Horizontal Direction at theDriven End (p1 = 70 bar, 9651 rpm, Film Injection Pattern).

Figure 18 presents the frequency spectra of the horizontalvibration component, measured at the driven end of the machine,while it was operating at 9651 rpm, with a suction pressure of 30bar (435.1 psi) and handling a volumetric flow of 2200 Am3/hr(77,692.27 ft3/hr). The bottom spectrum corresponds to dry gasoperation while the top spectrum represents the machine behaviorwhile handling a two-phase gas mixture, with liquids injected witha uniform droplet pattern. As may be seen in the figure, thevibration spectra for both the dry gas and the wet gas compressionshow similar trends at frequencies above the machine runningspeed. However, the spectrum corresponding to the wet gas com-pression shows some peaks in the subsynchronous range. Theappearance of a peak at one half the running speed suggests thepresence of some type of rotor instability. The compressor manu-facturer is currently conducting an investigation to determine thesource of this instability. Note, however, that this behavior wasobserved when the machine was operating with a gas-volumefraction of 0.97, which at the suction pressure of 30 bar (435.1 psi)represents a gas quality of 0.530. In addition, it is important to statethat the vibration amplitude at the running speed and its harmonicsdid not increase when the subsynchronous component appeared.Furthermore, this subsynchronous component disappeared whenthe gas-volume fraction was increased above 0.98 (quality wasincreased above 0.62).

The figures presented above provide a sample of the machinerotordynamic behavior that was observed during the tests. It isimportant to note that the behavior characteristics presented foreach combination of suction pressure, machine speed, and liquidinjection pattern were exhibited by the machine throughout thewhole range of volume flows that were evaluated during each testperiod. This information is not included in this paper as it would berepetitive and would produce an excessively long document.

Figure 18. Effects of Two-Phase Flow over the Vibration Responseof the Machine, Measured in the Horizontal Direction at theDriven End (p1 = 30 bar, 9651 rpm, Uniform Droplet InjectionPattern).

SUMMARY AND CONCLUSIONS

This paper presented the results of tests performed on a cen-trifugal compressor operating under wet gas conditions, showingthe effects of suction pressure, gas-volume fraction, machinespeed, and liquid injection pattern over the thermodynamic andmechanical performance of the machine.

The application of centrifugal compressor technology is a viableoption for single-stage, two-phase compression at gas-volumefractions at or above 0.97, corresponding to gas qualities as low as0.522 for suction pressures of 30 bar (435.1 psi) and 0.738 forsuction pressures of 70 bar (1015.3 psi). The exact level of gas-volume fraction will of course depend on the values of suctionpressure and pressure ratio, as well as the distribution of the liquidphase within the gas when it enters the machine. For future appli-cations, the relative densities and phase properties of the gas andliquids that are handled will need to be considered.

The thermodynamic evaluation of the machine showed thatrelative to a dry gas compressor, the compressor pressure ratioincreased when the gas-volume fraction was decreased within thevalues that were tested (GVF between 1.0 and 0.97). The increasein pressure ratio was attributed to the larger density of the fluid thatwas being handled by the compressor when liquids were injected.In addition, the compressor temperature ratio showed a slightdecrease when liquids were injected. This was probably caused bya transfer of energy from the gas to the liquid (heating of the liquid),and a limited condensate phase transition through the compressor.

The specific compressor power consumption was also reducedas the liquid fraction was increased. Nevertheless, when comparedto separating the liquid and vapor phases and boosting them asseparate streams, the specific power consumption for wet gas com-pression was larger.

For the data presented herein, the polytropic head for two-phasecompression can be represented by a nondimensional head coeffi-cient provided that the proper two-phase terms are included in thecalculations. The two-phase polytropic efficiency of the machinedecreased as the gas-volume fraction was reduced. This effect wasmore pronounced for the tests executed at the lower suctionpressure.

No evidence of liquid erosion was detected by visual inspectionof the machine internals after the test. It was noticed that theinternals of the machine were cleaned by the liquid that had beeningested.

A repeatable reduction in the noise level of the machine wasdetected when the compressor was handling the wet gas mixture.The dynamic pressure transducer data showed that the pressurefluctuations within the flow were attenuated by the presence ofliquid in the gas-stream.


TP-70-F-9651-2200 GVF = 1 and 0.98, X = 1 and 0.810

0 100 200 300 400 500 600 700 800 900 1000

Frequency [Hz]

Sig

na

l A

mp

litu

de

Film Injection, p1 = 70 bar, Flowrate = 2200 Am3/hr

GVF = 1.00

GVF = 0.98


0

0

TP-30-D-9651-2200 GVF = 1 and 0.97, X = 1 and 0.53

0 100 200 300 400 500 600 700 800 900 1000

Frequency [Hz]


Sig

na

l A

mp

litu

de

Droplet Injection, p1 = 30 bar, Flowrate = 2200 Am3/hr

GVF = 1.00

GVF = 0.97

0

0

In general, the test results showed that the vibration of themachine was not significantly affected by liquid hydrocarboningestion for both the uniform droplet as well as the fluid filminjection pattern. This may not be the case if the liquids are notuniformly distributed. Also, for the case in which the quality of thegas was below 0.62, the appearance of a subsynchronous vibrationsuggested that liquids could have been entrained in the seal areasat the impeller eye and balance piston, causing some type of rotorinstability.

The effects of liquid phase change that may occur inside themachine should be further investigated prior to embarking in thedesign of a multistage centrifugal compressor for wet gas applica-tions. Recall that the phase change inside the compressor maycause a mismatch between the stages, leading to performanceshortfalls. Furthermore, the effects of liquid ingestion over themachine internal loss mechanisms should be investigated.

DISCLAIMER

The information contained in this paper consists of factual dataand technical interpretations and opinions, which, while believedto be accurate, are offered solely for informational purposes. Norepresentation or warranty is made concerning the accuracy ofsuch data, interpretations, and opinions.

NOMENCLATURE

Parameters

D = Impeller exit diameterGVF = Gas-volume fractionh = Enthalpym.

= Mass flowMW = Molecular weightn = Polytropic volume exponentN = Machine rotational speedp = Pressure (absolute)P = PowerQ.

= Actual volumetric flowRo = Universal gas constantT = TemperatureU = Tangential velocityU2 = Impeller tip speedν = Specific volumeWp = Polytropic headx = Gas qualityZ = Compressibility factorμp = Polytropic head coefficientηp = Polytropic efficiencyφ = Flow coefficient

Subscripts

1 = Machine inlet2 = Machine dischargel = Liquid

g = Gasp = PolytropicTot = TotalTP = Two-phase

REFERENCES


Colby, G. M., 2004, “Performance Test Procedure—Revision 3,”D-R Internal Document # 003-085-001, Dresser-Rand Com-pany, Olean, New York.

Marshal, D. F. and Sorokes, J. M., 2000, “A Review ofAerodynamically Induced Forces Acting on CentrifugalCompressors, and Resulting Vibration Characteristics ofRotors,” Proceedings of the Twenty-Ninth TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas, pp. 263-280.

Moore, J. J., Walker, S. T., and Kuzdzal, M. J., 2002, “Rotordy-namic Stability Measurement During Full-Load, Full-PressureTesting of a 6000 PSI Reinjection Centrifugal Compressor,”Proceedings of the Thirty-First Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 29-38.

Schultz, J. M., 1962, “The Polytropic Analysis of CentrifugalCompressors,” ASME Journal of Engineering for Power, 84,pp. 69-82.

Scott, S. L., “Evolution of Subsea Multiphase Pumping and Wet-Gas Compression,” Presentation delivered during theWelcoming Address at the Thirty-Third TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas.

Stahley, J. J., 2000, “Field Performance Test Procedure—Revision1,” D-R Internal Document # 003-347-001, Dresser-RandCompany, Olean, New York.

ACKNOWLEDGEMENTS

The authors would like to thank Dr. D. Lee Hill and Dr. J. JeffreyMoore, formerly of Dresser-Rand, for their assistance during therealization of the wet gas tests. Also thanks to Mr. Mark Kuzdzal(Turbomachinery Symposium Advisory Committee Monitor) forreviewing this manuscript and providing his suggestions toimprove this paper.

Considering the fact that the tested compressor is a constituentpart of the K-lab test loop, the authors are grateful for the opportu-nity of using the K-lab compressor for wet gas testing, and wouldlike to thank Steinar Jørgensen and Evert Wahlberg of Statoil fortheir willingness and support during the test.

Finally, the authors would like to thank Statoil and Dresser-Randfor their funding of this work and permission to publish the results.


Gary M. Colby presently is a TestEngineering Supervisor with Dresser-RandCompany, in Olean, New York. He isresponsible for developing test methods tomeet objectives for production compressorsand analytical aerodynamic testing of cen-trifugal compressors. Mr. Colby has heldseveral engineering positions over his 32year career at Dresser-Rand Company. Themajority of his work experience has been inthe thermodynamic performance field of

centrifugal compressors. He has more than 14 years of experiencein testing of centrifugal compressors both in the shop and the field.

Mr. Colby studied Mechanical Technology for two years at theState University of New York at Alfred. He has authored severalpapers on hydrocarbon testing of compressors.

ABSTRACT

The purpose of this tutorial is to familiarize the rotatingequipment engineer with in-shop hydraulic performance testingmethods. The tutorial discusses the theory of similarity testing, theassumptions, and inherent errors. The presentation includes aselected case for demonstration. It is intended for engineers whodefine performance test requirements for assigned projects, reviewtest agendas, and witness vendor compressor performance tests.Allowable test departures described in the ASME PTC-10 (1997)are reviewed and discussed.

INTRODUCTION

Low pressure inert gas Type 2 (ASME PTC-10, 1997) hydraulicperformance testing of a centrifugal compressor is based upon thesimilitude between the specified condition volume reduction andthe test condition volume reduction. Comprehension of volumereduction and the variables that effect volume reduction of thecompressor are key to understanding the accuracy and limitationsof the test.

For a multistage compressor volume reduction is a criticalparameter. The volume reduction of the first stage determines theimpeller selection (capacity) of the second stage impeller that, inturn, determines the selection of the following stage and so forththrough the discharge of the compressor. Once the final selectionis determined the design volume reduction for the rotor is estab-lished. In conducting the in-shop performance test, the test volumereduction must match the design volume reduction to be represen-tative of the compressor performance on the specified gas, theobjective of the test.

VOLUME REDUCTION

The volume reduction of a compressor stage is dependentupon the polytropic head (work output) and efficiency of theimpeller, the gas density, and properties of the medium beingcompressed. Work applied to the gas through the impeller due tocentrifugal force and diffusion results in increased pressure at

the discharge of the stage reducing the gas volume. The ratio ofdischarge volume to inlet volume is the volume reduction of thestage.

To examine the effect of various parameters on the volumereduction of a stage the real gas polytropic head equation ispresented.

(1)

(2)

Using Equation (1) assume, for the purpose of demonstration,that all variables remain constant with the exception of head andpressure ratio. If the head is increased, the pressure ratio mustincrease to maintain equality in the equation. The higher pressureratio results in increased discharge pressure and lower dischargevolume. Increasing head increases the volume reduction; decreas-ing head decreases volume reduction. Now for example assumethat all the variables in Equation (1) are constant except formolecular weight and pressure ratio. If the molecular weight isreduced, the pressure ratio is also reduced to maintain equality. Thereduced pressure ratio results in less volume reduction for thestage. Decreasing the molecular weight decreases the volumereduction; increasing the molecular weight increases the volumereduction of the stage.

The same principle may be applied to the other variables inEquations (1) and (2) with reference to the pressure ratio and thefollowing Table 1 developed.

Table 1. Effect on Volume Reduction for Changes in OperatingParameters.

147

HYDRAULIC SHOP PERFORMANCE TESTING OF CENTRIFUGAL COMPRESSORS

byGary M. Colby

Test Engineering Supervisor


Olean, New York

HEADMW

T Zn

n

P

P

n

n= × × ×

−⎛⎝⎜

⎞⎠⎟

×⎛⎝⎜

⎞⎠⎟ −

⎡

⎣

⎢⎢⎢

⎤

⎦

⎥⎥⎥

−1545

111 1

2

1

1

n

n efficiencyfor an ideal gas

kk

−⎛⎝⎜

⎞⎠⎟

=⎛

⎝⎜⎜

⎞

⎠⎟⎟

−1 1

Note that the change in inlet pressure has no effect on thevolume reduction of the stage. The discharge pressure changes pro-portionately to maintain equality; the volume reduction is thesame. Pressure effect on the stage volume reduction is limited tothe pressure effect on the gas isentropic exponent (k) and the com-pressibility factor (z).

To visually represent volume reduction the author will use athree-stage compressor for an example. Figure 1 shows three stagecurves of flow coefficient (inlet capacity) versus head for the com-pressor. The design capacity of the first stage impeller is identifiedby the solid line. The volume reduction of this stage underspecified conditions yields the design capacity of the second stagerepresented by the solid line on the second stage curve. The thirdstage curve design capacity is established by the volume reductionof the second stage impeller.

Figure 1. Stage Curve Example for a Three-Stage Compressor.

If the compressor were operated on a molecular weight higherthan the specified molecular weight the volume reduction would begreater. For purpose of demonstration assume that only the moleweight is changed. Starting at the design inlet capacity to the firststage, the greater volume reduction due to the increase in molecularweight reduces the operating capacity into the second stage relativeto design. At the reduced capacity the second stage produces morehead and volume reduction, in addition to the higher mole weightresulting in further deviation from the design capacity at stagethree. The operating capacity at each stage at the higher volumereduction due to the higher mole weight is represented as a dashedline on the three stage compressor maps shown in Figure 1.

If this were a test using a higher than design mole weight gas,the volume reduction would be much greater than design. Thevolume reduction of the compressor would need to be reducedback to design in order to produce a relative performance to thedesign condition. This is achieved by lowering the head of thecompressor relative to design. Therefore, the shop test head will belower than the design head. The polytropic head of the compressoris compared, test to design, using the polytropic head coefficient.

The polytropic head coefficient (μ) and efficiency of a stage aredependent upon the geometry of the impeller, the capacity beingpassed, the speed, and the inlet Mach number. The polytropic headof a stage is the product of the head coefficient and the impeller tipspeed squared [Equation (3)].

(3)

The impeller geometry for the test is the same as design. The headcoefficient test to design, by definition, should be equal. Thereforeto reduce the head the speed of the compressor is lowered.

For a multistage compressor an overall μ may be calculatedusing the overall head and the summation of the impeller tip speedssquared [Equation (4)].

(4)

The capacity of the compressor is compared during the test using adimensionless flow coefficient, Q/ND3. Typically only Q/N is used

for the flow coefficient as the impeller diameter is the same test todesign.

The head coefficient and efficiency of an impeller vary with themachine Mach number. Machine Mach number is defined as thetip speed of the first stage impeller divided by the inlet sonicvelocity of the gas [Equation (5)].

(5)

Figure 2 shows a typical stage performance map of head coeffi-cient and efficiency versus inlet flow coefficient at three machineMach numbers. The variance between the test Mach number andthe design Mach number will affect the head coefficient (andvolume reduction) of the impeller. Therefore performance testing isrequired to be conducted at a Mach number relatively close to thedesign Mach number for each stage and is critical to the accuracyand relevance of the test. For this reason the ASME PTC-10 (1997)provides an allowable departure chart (Figure 3.3 of Code) for dif-ferences in the machine Mach number between test and design(Figure 3). The greater the machine Mach number the tighter theallowable tolerance. The Code only addresses the machine Machnumber at the first stage, assuming that all following stages havethe same percentage departure, test versus specified.

Figure 2. Typical Stage Curve at Three Mach Numbers.

Figure 3. Allowable Test Mach Number Departure from Design.(Courtesy ASME)


Capacity

Hea

d

Design VRD

High Mol. Wt. VRD

Capacity

Hea

d

Capacity

Hea

d

Stage 1 Stage 2 Stage 3

0.7 0.8 0.9 1 1.1 1.2 1.3normalized Flow Coefficient

0.6

0.8

1

1.2

1.4

norm

aliz

edH

ead

Co

effi

cien

t

0.7

0.8

0.9

1

1.1

no

rmal

ized

Ef fi

cie n

cy

Low Mach Number

Med Mach Number

High Mach Number

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6Mach No. Specified

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

Mac

hN

o.T

est

-M

ach

No

.S

pec

ifie

d

Fig. 3.3 from the ASME PTC 10-1997

HeadU

g=

×μ 2

μ =×

∑Head g

U2

Mach NumberU

A

U

kgRTZmachine = =

0

DETERMINATION OF TEST CONDITIONS

Of the variables that effect volume reduction the compressormanufacturer has control over the inlet temperature, test gasmedium, test speed, and inlet pressure.

The test inlet temperature is dependent upon the coolingcapacity of the manufacturers’ selected test stand. Typically the testinlet temperature is set at 100°F. The difference in volumereduction caused by a difference in inlet temperature betweendesign and test is corrected via the test speed.

Table 2 lists the typical inert test gases used for an ASME PTC-10 (1997) Type 2 in-shop performance test with the respectivemole weights and approximate k values.

Table 2. Typical Inert Gas Mediums for In-Shop PerformanceTesting.

Inlet pressure has no effect on volume reduction except for thegas property variance as discussed previously. Inlet pressure doeshave a significant effect on the test. The pressure directly relates tothe amount of mass compressed. The mass determines the powerrequirement as well as the level of effect the convection andradiation losses to the atmosphere have on the test results. The testReynolds number is directly proportional to inlet pressure. Theeffects of heat loss and Reynolds number on the performance testare discussed later. In most cases the test suction pressure may beapproximated at 10 percent of the specified value.

A sample compressor application is presented as an example todiscuss the selection of the test inlet conditions, gas medium, andspeed. The process compressor selected is a five-stage straight-through compressor design. The specified operating conditions areshown in Table 3.

If helium was selected as the test medium for the sample com-pressor the difference in molecular weight, 4.0 versus 26.3, wouldresult in the test volume reduction to be greatly reduced andtherefore the required test speed would exceed the maximum con-tinuous speed of the compressor. For this reason the test gasmedium selected is almost always higher in mole weight than thespecified gas. A second criterion for selection of the test gas is thek value of the test medium versus the k value of the specified gas.The k value not only affects the volume reduction but also the inletMach number. Generally the test gas with the closest k value to thespecified gas is selected.

Carbon dioxide was selected for the test gas in the sample com-pressor. The inlet temperature was set at 100 F. The inlet pressurewas set at 44 psia. The test speed to achieve the same overallvolume reduction may now be calculated based on thermodynamiclaw as follows.

Thermodynamic Law:

(6)

Therefore:

(7)

Table 3. Specified Operating Conditions.

Rearranging:

(8)

To have similitude between the specified gas conditions and the

test gas conditions, the volume reduction must be equal.

Therefore:

(9)

The polytropic volume exponent (n) is a function of the gasproperties and efficiency of the compressor. By definition, the effi-ciency of the test should equal the efficiency at design. Knowingthe test gas properties the test polytropic volume exponent may becalculated and the required test pressure ratio to meet the specifiedvolume reduction may be established.

The test required head is then determined to meet the specifiedvolume reduction using Equation (1).

Head for a compressor is proportional to the speed squared[refer to Equation (3)].

Therefore:

(10)

With the test gas selected and the test inlet conditions given thetest speed was calculated at 11,070 rpm. The overall test conditionsat the design point are shown in Table 4.

These conditions are now reviewed and compared to the ASMEPTC-10 (1997), Table 3.2, “Allowable Departures from SpecifiedConditions” for Type 1 and Type 2 tests. The comparison for thesample compressor is given in Table 5.

HYDRAULIC SHOP PERFORMANCE TESTING OF CENTRIFUGAL COMPRESSORS 149

� �V1V2

Inlet pressure 150 psia

Discharge pressure 510 psia

Inlet temperature 110 �F

Discharge temperature 274 �F

Molecular weight 26.3 Mols

Inlet compressibility 0.961

Inlet capacity 1999 Aft3/min

Mass flowrate 1342.8 lbs/min

Discharge capacity 747.0 Aft3/min

Polytropic head 44,401 ft lbs/lb

Power 2362 gas hp

Bearing loss 52 hp

Speed 13,100 rpm

Machine Mach number 0.658

Reynolds number 3.63E+06

Isentropic exponent (k) 1.187

PV cons tn = tan

PV P Vn n

1 1 2 2=

P

P

V

V

n

2

1

11

2

⎛⎝⎜

⎞⎠⎟ =

/

V

Vmust equal

V

V

P

Ptest specified test

n

1

2

1

2

2

1

1⎛⎝⎜

⎞⎠⎟

⎛⎝⎜

⎞⎠⎟ =

⎛⎝⎜

⎞⎠⎟

/

Speed SpeedHead

Headtest specified

test

specified

= ×

Table 4. Overall Test Conditions at Design Point.

Table 5. Allowable Departures for Sample Compressor at 11,070RPM.

There are inherent errors in Type 2 testing based on some of theassumptions taken when calculating the test conditions. The speedcalculation is based on the inlet and discharge endpoints assuming astraight line polytropic path on the pressure-enthalpy diagram. Thepath is truly a slight curve. There are differences in Mach numberbetween the test and specified condition. The allowable Machnumber departure between design and test is compared only at thefirst stage of a section. The departure is not equal at all stages in amultistage compressor. The compressibility factor (z) change frominlet to discharge typically is greater under specified conditions thanunder the low-pressure test condition. These all cause variations inthe individual stage volume reduction from specified. To evaluatethe overall effect of the variances, compressor performance at thetest condition is calculated on a stage by stage basis. The results arethen converted to the dimensionless head coefficient (μ), efficiency,and flow coefficient. This test performance curve is plotted on thepredicted curve at specified conditions for comparison. The resultsfrom the stage by stage calculation of the sample compressor arepresented in Figure 4. The head coefficient versus capacity curve forthe test condition is not in agreement with the specified gas curve.

Figure 4. Head Coefficient and Efficiency Versus Inlet FlowCoefficient at 11,070 RPM.

For the majority of applications the predicted stage by stagecurve will fall on the specified curve with little variation. Theauthor purposely selected an example where the curves do notmatch well, even though the test condition is within the allowabledepartures as given by the ASME PTC-10 (1997) for a Type 2 test.

To visualize what caused the variation at the design inlet flowcoefficient the test flow coefficients for each stage were normal-ized using their respective design values as the reference (Figure5). If the test is perfect, the test condition line will fall exactly uponthe design line in Figure 5. However at 11,070 rpm, the Codeestablished test speed, the first stage volume reduction is greaterunder the test condition than under specified conditions. Thisresults in the second stage operating at a flow coefficient less thandesign. As shown earlier if a stage operates at a lower capacity itwill generate more head. The additional head and the test variancesresult in stage two producing a greater volume reduction. Thiseffect continues through the stages until the fourth and fifth stagesoperate at almost 3 percent less relative capacity than under thespecified condition. Overall the volume reduction at the testcondition (11,070 rpm) is greater than the design and therefore apoor curve match.

Figure 5. Normalized Flow Coefficient—11,070 RPM.

To correct for the higher volume reduction the test speed of11,070 rpm was lowered reducing the volume reduction andimproving the stage by stage match. A corrected test speed of10,849 rpm was determined. A second stage by stage calculationwas conducted at the revised speed and the predicted curve plottedon the specified gas curve (Figure 6).

The predicted curve for the test is now found to agree closelywith the specified gas predicted curve. At the design inlet flowcoefficient the stage flow coefficients were normalized and plottedas before (Figure 7). At the lower test speed of 10,849 rpm no stageoperates greater than 1 percent from its design flow coefficient.


Inlet pressure 44.0 psia

Discharge pressure 170.2 psia

Inlet temperature 100.0 �F

Discharge temperature 348.9 �F

Molecular weight 44.01 mols

Inlet compressibility 0.987

Inlet capacity 1799 Aft3/min

Mass flowrate 587.9 lbs/min

Discharge capacity 672.0 Aft3/min

Polytropic head 31,707 ft lbs/lb

Power 743 gas hp

Bearing loss 37 hp

Speed 11,070 rpm

Machine Mach number 0.693

Reynolds number 1.03E+06

K value 1.261

0.7 0.8 0.9 1 1.1 1.2normalized Flow Coefficient

0.8

0.85

0.9

0.95

1

1.05

1.1

1.15

norm

aliz

edM

uan

dE

ffic

ienc

y

Design Condition Curve

Carbon Dioxide at 11070 rpm

0.96

0.97

0.98

0.99

1

1.01

1.02

no

rmal

ized

Flo

wC

oef

fici

ent

Design Condition

Test on CO2 at 11070 rpm

Inlet

Flange

Stage

One

Stage

Two

Stage

Three

Stage

FourStage

Five

DischargeFlange

Volume Reduction Review for Example Compressor

Figure 6. Head Coefficient and Efficiency Versus Inlet FlowCoefficient at 10,849 RPM.

Figure 7. Normalized Flow Coefficient—10,849 RPM.

The stage by stage test condition at the lower speed is againcompared to the allowable departure table to ensure that the Coderequirements are met. Table 6 yields the departure values from thespecified condition for the stage by stage calculations at the Codespeed of 11,070 rpm and the adjusted speed of 10,849.

Table 6. Allowable Departures for Sample Compressor at 10,849RPM.

The stage by stage calculation of the test condition is a check ofthe test condition for the combined effect of all variances betweentest and specified conditions. Requesting a plot of the test curve onthe specified curve from the manufacturer will validate that the testcondition meets the objective of the test.

DETERMINATION OF SUCTION PRESSURE WITHRESPECT TO AVAILABLE POWER, REYNOLDSNUMBER, AND HEAT LOSS TO ATMOSPHERE

In determining the test suction pressure consideration must begiven to the power available in the manufacturer’s shop, theReynolds number variance from specified, and the heat loss

through the compressor casing. Inlet pressure is directly related toinlet density and therefore the mass compressed and subsequentlythe power. The test must be run within the power limits of theavailable shop driver.

Reynolds number relates to the boundary layer and frictionallosses of the medium in the flowpath. Type 2 in-shop performancetests are almost always conducted at a lower Reynolds numberthan under specified conditions. The lower the Reynolds numberthe greater the frictional losses and lower the efficiency. Thismeans that the head and efficiency observed during the test will belower than what would be observed in the field at the same flowcoefficient. The ASME PTC-10 (1997) (Figure 3.5) provides achart showing the allowable departure in test Reynolds numberrelative to the specified value. The Code also provides a method tocalculate a correction to be applied to the test data for the variancein Reynolds number. Utilizing the ASME PTC-10 (1997) methodfor Reynolds number correction, a graph is presented for thesample compressor (Figure 8). For this sample the allowable rangefor test Reynolds number is 10 percent to 138 percent of thespecified value. If the test was conducted at the minimum 10percent limit the anticipated correction would be 1.2 percent. Notein Figure 8 that the correction is not linear. The lower the absolutevalue of Reynolds number the greater the correction exponentially.The minimum allowable Reynolds number absolute value given bythe Code is 90,000.

Figure 8. Reynolds Number Correction.

Most users do not allow the manufacturer to correct the test datafor the differences in Reynolds number, test to design. The manu-facturer should select a suction pressure to maintain the Reynoldsnumber correction at its lowest level within the power and pressurerestraints of the test stand. Even if correction is not allowed, it isrecommended that the user be aware of the test level.

There are two methods for determining the efficiency of thecompressor. A torquemeter may be used to measure the work inputinto the compressor. From this power the friction of the bearingsand seals may be subtracted to obtain the power input into the gas.The head, work output, is divided by the work input to achieve theefficiency. The second (most common) method, is the heat balancemethod where the work input is measured via the enthalpy risefrom inlet to discharge.

Using the heat balance method requires an understanding of theheat lost to atmosphere through the casing by way of conductionand radiation. The heat loss to atmosphere subtracts from the workinput measured and if not minimized, eliminated, or accounted forwill result in a falsely high efficiency. As the mass throughputincreases, the percentage of heat loss to atmosphere relative to the


0.7 0.8 0.9 1 1.1 1.2normalized Flow Coefficient

0.8

0.85

0.9

0.95

1

1.05

1.1

1.15no

rmal

ized

Mu

and

Eff

icie

ncy

Design Condition Curve

Carbon Dioxide at 10849 rpm

0.0E+000 1.0E+006 2.0E+006 3.0E+006 4.0E+006Test Reynolds Number

0.995

1.000

1.005

1.010

1.015

Hea

d an

d E

ffic

ienc

yco

rrec

tion

fact

or

0

50

100

150

Tes

t Suc

tion

Pres

sure

PSIA

0

40

80

120

Perc

ent o

f Sp

ecif

ied

Rey

nold

s N

umbe

r

0.96

0.97

0.98

0.99

1

1.01

1.02

no

rmal

ized

Flo

wC

oef

fici

ent

Design Condition

Test on CO2 at 10849 rpm

Inlet

Flange

Stage

One

Stage

Two

Stage

Three

Stage

FourStage

Five

DischargeFlange

Volume Reduction Review for Example Compressor

work input decreases. Here again the importance is seen of testsuction pressure. The higher the pressure the greater the mass. Forthe sample compressor the convection and radiation losses wereestimated at various test suction pressures and the results wereplotted in Figure 9. The 44 psia suction pressure selected for thesample compressor yields an approximate 0.44 percent heat loss toatmosphere, essentially countering the Reynolds number correc-tion. In the few cases where the heat losses cannot be minimized itis recommended that the casing be insulated for the performancetest. The user is advised to ask what the approximate heat lossesare for the in-shop performance test to ensure that the efficiencyobserved is representative.

Figure 9. Convection and Radiation Losses to Atmosphere.

MEASURING RECIRCULATION

At the specified operating point API 617, Seventh Edition(2002), states that the horsepower shall be guaranteed within 4percent of the quoted value. The percentage varies dependent onthe user. Note that efficiency is not guaranteed. The power requiredis the product of the total mass compressed and the work input intothe gas [Equation (11)]. It is recommended that where possible,any recirculation or leakage losses should be measured during theshop performance test and the results applied to the specifiedcondition in the conversion of results.

(11)

Shown in Figure 10 is a typical loop schematic for a straight-through single section compressor. The straight-through unit loopconsists of an inlet flowmeter, a discharge throttle valve, and a heatexchanger. The balance piston return line is piped through aflowmeter and into the test loop upstream of the main inletflowmeter. This enables measurement of the balance piston sealleakage. The predicted balance leakage is computed similar to anorifice equation. Manufacturers do not all use the same equationsfor calculating predicted leakages but the physics of the computa-tion is common. A simplified example is given in Equation (12).

(12)

where:

C = ConstantD = Seal diameterS = Seal clearanceP = Pressure

Subscripts:

h = High side of seall = Low side of seal

Figure 10. Typical Test Loop—Straight-Through CompressorDesign.

During the test the measured value of the balance leakage isknown via the flowmeter installed into the return line. Thedischarge pressure of the compressor may be assumed to be theupstream condition of the seal and the flowmeter upstreampressure assumed as the downstream condition. From these data apseudo clearance may be calculated and used to convert the resultsto specified conditions. This is considered a pseudo clearance asthe calculated value corrects for any error in the constant.

Back-to-back compressor designs have two major recirculationpaths that need to be accounted for in the gas power calculation:the division wall leakage and the end seal leakage. The divisionwall seal is between the first section discharge and the secondsection discharge. Leakage from the second section dischargepasses across the seal into the first section discharge through theinterstage and recompressed by the second section impellers. Theend seal leakage (also referred to as the seal balance leakage)passes across the end seal at the section two inlet back to the firstsection inlet. This leakage is compressed by the first sectionimpellers and then passes through the interstage piping and vesselsback into the second section suction. Essentially the division wallseal leakage is recycled around the second section and the end sealleakage is recycled around the first section.

Arrangement of the test loop will allow for the measurement ofthese leakages so that they may be compared to predicted valuesand used in the conversion of test results. Figure 11 shows a looparrangement for a back-to-back compressor that enables theleakages to be measured.

Each section is piped with its own test loop similar to thestraight-through example. An orifice is placed in the seal balanceline to measure the seal leakage directly. It should be noted herethat during the test each end seal is referenced separately for sealsupply (oil or gas) purposes. The two loops are connected by abalance line between the first section discharge and the secondsection inlet, which is also orificed. Figure 10 shows an additionalbalance line, discharge to discharge, the purpose of which will bediscussed later.

During the second section test the seal balance line valve is openas is the discharge to inlet loop balance line valve. The dischargeto discharge loop balance line valve is closed. This allows thesuction pressure of the second section to be very close to thedischarge pressure of the first section enabling leakage across bothof the seals in question. To maintain a mass balance in the system,


0 20 40 60 80 100 120 140 160Test Suction Pressure - PSIA

0.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

1.600

Hea

tLos

sto

Atm

oshe

reP

erce

ntof

Tot

al H

eatI

nput

GHPHead Mass

Efficiency=

××33000

( )mass C D S

P P

density

h l

h

= × × × ×−

Π

Figure 11. Typical Test Loop—Back-to-Back Compressor Design.

whatever leakage leaves the second section loop to the first sectionloop must return to the second section loop. It does so via thedischarge to inlet loop balance line. The division wall leakage isthe difference in mass flow between the loop balance lineflowmeter and the end seal flowmeter. The evaluation of theleakage rate is conducted in the same manner as the balance drumleakage. The upstream condition for the division wall is the secondsection discharge; the downstream condition is the first sectiondischarge. For the end seal, the upstream condition is the secondsection suction and the downstream condition the first sectionsuction. The flow through the second section impellers is thesecond section main orifice minus the seal balance orifice.

While testing the first section the valve arrangement must beadjusted to eliminate the recycles so that a proper enthalpy risemay be observed. If the valves were left open then the division wallleakage would mix with the first section discharge mass resultingin an error in the first section discharge temperature observed at theflange. During the section one performance test the discharge toinlet loop balance valve is closed and the discharge to dischargebalance line valve opened. The end seal valve is also closed. Thisarrangement causes the two opposing discharges to be at the samepressure eliminating the division wall seal leakage.

THE STANDARD TEST PROCEDURE

Typically a shop performance test is conducted at only therelative design point speed. The test typically consists of five testpoints at equally spaced capacities from overload to surge, withone point being at the design inlet flow coefficient. At each flowpoint the compressor is allowed to reach thermal equilibriumbefore the data point is recorded. Surge points are also observed attwo alternate speeds to define the slope of the surge line. The off-speed surge points are not normally thermally stable before thepoint is recorded as determining the head and flow coefficient ofthe point is the objective, not the efficiency. The off-speed surgepoints also assist the manufacturer in determining where, in whichstage, surge was initiated. If the three surge points are very close inflow coefficient, surge is initiated early in the compressor, the firstor second stage. A greater separation between the three surgepoints would indicate surge to be initiated in one of the later stages.Alternate speed points are also observed for constant speed com-pressors. In this case the slope of the surge line is established foroff-design mole weight conditions.

In the past determining surge was done by throttling the com-pressor until audible surge was detected. Many times this capacity

was not the actual surge point of the compressor as observed in thefield. Today with the advent of faster data acquisition equipmentand vibration monitoring, surge may be accurately determinedduring the shop test. There are five criteria that may be used todetermine the minimum stable flow point.

• Audible surge

• Onset of subsynchronous vibration

• Peak head

• Flow instability

• Pressure instability

Audible flow reversal in the compressor is readily evident as isflow and pressure instabilities. Observing the vibration frequencyspectrum is very important in identifying surge. Subsynchronousvibration will occur at the onset of surge providing there isadequate energy input into the unit. The frequency of the vibrationgenerally points to the source of the flow instability. Frequenciesin the 6 to 16 percent range may be indicative of a stall in the sta-tionary components of the flow path where frequencies in the 70 to90 percent range indicate surge is initiated in the impellers.Observing the head coefficient is another method of determiningsurge. In many cases the head drops off before any other indica-tions are noted. An example is presented in Figure 12. A unit wasthrottled toward surge slowly with a data acquisition system set tocalculate the performance every four seconds. As the unitapproached lower flow the head coefficient stopped rising andstarted to decrease as the flow was reduced further. The head thenstarted to rise again as the flow started to reduce further until anaudible surge was found. At the bottom of the head droop, subsyn-chronous vibration was observed at the journal bearing. Theminimum stable flow point was recorded at a capacity correspon-ding to the initial peak head point. This was at a 10 percent highercapacity than the audible surge capacity.

Figure 12. Head Versus Inlet Flow Coefficient—Transition Curvenear Surge.

ADDITIONAL SPEED LINE TEST

When should alternate test speed lines, in addition to thespecified design point speed line, be requested? It depends uponthe compressor, the process, and the user. Discussing the compres-sor operation with the process engineer will assist the rotatingequipment engineer with what are the critical parameters requiredfor the process to run efficiently. If there are alternate operatingconditions that are critical to the process the engineer may requesta second speed line corresponding to that condition to be tested. Itis a matter of risk assessment. The author recommends reviewingthe Mach number of the alternate condition with respect to thespecified operating point. If the variance is greater than would beallowed by the Code then an additional line may be warranted.


Section 1 Section 2

Inlet Flow Coefficient

Poly

tropic

Hea

dC

oef

fici

ent

Peak Head

Onset of

Sub-

synchronous

vibration

Audible

Surge

Methods used to extrapolate the overall test performance dataestablished at the specified operating point to alternate operatingpoints varies by manufacturer. A review of the manufacturer’smethod of extrapolation may also determine whether an additionaltest at an alternate operating point is warranted.

EVALUATION OF TEST RESULTS

Evaluating the pressure and temperature readings of the per-formance test to head and efficiency terms should always beconducted using real gas laws and an equation of state for gasproperties that is well referenced for the test medium. This tutorialis intended to address the method for establishing proper test con-ditions for in-shop performance testing. The author recommendsthe evaluation procedures outlined in ASME PTC-10 (1997) befollowed.

TYPE 1 PERFORMANCE TESTING

An ASME PTC-10 (1997) Type 1 test is conducted with thespecified gas at or very near the specified operating conditions.Deviations from the specified gas conditions are subject to limita-tions given in Tables 3.1 and 3.2 of the Code.

The same principles apply to the Type 1 test as to the Type 2 testdescribed earlier. It the test was conducted using the actualspecified gas and inlet conditions there would be no deviationsfrom the specified condition to the test. However, this is rarely thecase. In many instances the inlet temperature at the specifiedcondition cannot be achieved due to the cooling capacity of themanufacturers’ test stand and the specified gas is not available.

The test may be conducted on a gas different from the specifiedgas; however, the test gas must have a mixture k-value very closeto the specified gas. This is required to ensure that the Machnumber is the same as specified and that the thermodynamic con-version of the work input to the gas medium produces the samepressure ratio.

Typically the manufacturer has a local gas supply close to puremethane. With this source carbon dioxide, propane, or other gasesmay be blended to achieve the specified molecular weight and k-value. If the test inlet temperature is greater than the specified inlettemperature the test molecular weight will have to be higher thanspecified to offset the temperature change and any associatedchange in compressibility factor. The higher mole weight mixturemust still have a mixture k value close to specified.

Reviewing Equation 1 and Equation 5 it can be seen that theproduct of the gas constant, R (1545/mole weight), inlet tempera-ture, and inlet compressibility are common to volume reductionand machine Mach number. If a blended gas has the same k valueand the return-to-zero (RTZ) product is maintained at the inlet, theperformance will be the same as on the specified gas. The authorrecommends that the inlet RTZ product and inlet pressure be main-tained within ±2 percent of the specified value. The ASME PTC-10(1997) (Table 3.2) allows a much greater tolerance.

Comparable plots of compressor performance maps showingpolytropic head, efficiency, pressure ratio, and power may beproduced for the specified gas and the planned test gas medium.These plots will demonstrate how the compressor performanceoperating on the test gas blend and conditions conforms to thecompressor performance under the specified gas conditions. Thesecurves should be provided in the test agenda.

The Type 1 test point(s) should be established by agreementbetween the user and the manufacturer dependent upon theobjective of the test. It has been the author’s experience that thisobjective is not always well defined in the proposal stages of aproject. It is recommended that discussions between the user andthe manufacturer concerning the objectives of the test get started asearly in the project as possible. This will ensure that the test standloop design will accommodate all of the objectives.

SUMMARY

The information provided in this tutorial should provide theprocess and rotating equipment engineer with a better understand-ing of performance testing methods for centrifugal compressors.When reviewing the ASME PTC-10 (1997) the engineer will havea greater understanding of why the limitations outlined in the Codeare given.

NOMENCLATURE

MW = Molecular weight, molsT = Temperature, degrees FahrenheitZ = Compressibility factor, dimensionlessP = Pressure, pounds force per square inchn = Polytropic volume exponent, dimensionlessk = Isentropic exponentU = Tip speed, feet per secondμ = Polytropic head coefficient, dimensionlessA0 = Sonic velocity, feet per secondg = Gravitational constant, feet per second squaredV = Specific volume, cubic feet per pound massN = Speed, revolutions per minuteQ = Capacity, cubic feet per minuteGHP = Gas horsepowerMass = Mass flow rate, pounds per minuteD = Diameter, inchesS = Radial clearance, inchesDensity = Pounds mass per cubic footC = A given constant

Subscripts:

1 = Inlet2 = Dischargeh = High pressure side (upstream)l = Low pressure side (downstream)

REFERENCES




Minhui He is currently working as aMachinery Specialist at BRG MachineryConsulting LLC, in Charlottesville, Virginia.His responsibilities include vibration trou-bleshooting, rotordynamic analysis, as wellas bearing and seal analysis and design. Heis a member of STLE, and is also conductingresearch on hydrostatic journal bearingsand hydrodynamic thrust bearings for tur-bomachinery applications.

Dr. He received his B.S. degree (Chem-ical Machinery Engineering, 1994) from Sichuan University inChina. From 1996 to 2003, he conducted research on fluid filmjournal bearings in the ROMAC Laboratories at the University ofVirginia, receiving his Ph.D. degree (Mechanical and AerospaceEngineering, 2003).

C. Hunter Cloud is President of BRGMachinery Consulting, LLC, in Charlottes-ville, Virginia. He began his career withMobil Research and Development Corpo-ration in Princeton, New Jersey, as aTurbomachinery Specialist responsible forapplication engineering, commissioning,startup, and troubleshooting for produc-tion, refining, and chemical facilitiesworldwide. During his 11 years at Mobil,he worked on numerous projects, including

several offshore gas injection platforms in Nigeria, as well asserving as reliability manager at a large U.S. refinery.

Currently, Mr. Cloud also serves as Lab Engineer at theUniversity of Virginia’s ROMAC Laboratories, where he ispursuing a doctorate. His research focuses on the measurement ofturbomachinery stability characteristics. He is a member of ASME,the Vibration Institute, and the API 684 Rotordynamics Task Force.

James M. Byrne is President of RotatingMachinery Technology, Inc., in Wellsville,New York. He began his career designinginternally geared centrifugal compressorsfor Carrier, in Syracuse, New York. Mr.Byrne continued his career at Pratt &Whitney aircraft engines and became atechnical leader for rotordynamics. Laterhe became a program manager for Pratt &Whitney Power Systems, managing thedevelopment of new gas turbine products.

Mr. Byrne holds a BSME degree from Syracuse University, anMSME degree from the University of Virginia, and an MBA fromCarnegie Mellon University.

ABSTRACT

Widely used in turbomachinery, the fluid film journal bearingis critical to a machine’s overall reliability level. Their designcomplexity and application severity continue to increase makingit challenging for the plant machinery engineer to evaluate theirreliability. This tutorial provides practical knowledge on theirbasic operation and what physical effects should be included inmodeling a bearing to help ensure its reliable operation in thefield. All the important theoretical aspects of journal bearingmodeling, such as film pressure, film and pad temperatures,thermal and mechanical deformations, and turbulent flow arereviewed.

Through some examples, the tutorial explores how differenteffects influence key performance characteristics like minimumfilm thickness, Babbitt temperature as well as stiffness anddamping coefficients. Due to their increasing popularity, theoperation and analysis of advanced designs using directed lubrica-tion principles, such as inlet grooves and starvation, are alsoexamined with several examples including comparisons to manu-facturers’ test data.

155

FUNDAMENTALS OF FLUID FILM JOURNAL BEARING OPERATION AND MODELING

byMinhui He

Machinery Specialist

C. Hunter CloudPresident

BRG Machinery Consulting, LLC

Charlottesville, Virginia

andJames M. Byrne

President


Wellsville, New York

INTRODUCTION

The objectives of this tutorial are to provide the reader thefollowing with respect to fluid film journal bearings:

• A basic understanding of their physics and operational consider-ations

• A basic understanding of their modeling fundamentals

• The knowledge to better interpret more advanced papers andtopics

• A good reference source for the future

This tutorial is not:

• A design guideline. The authors do not intend to teach how todesign a bearing for any particular application. The literature isreplete with fine design guidelines including those by Nicholas andWygant (1995).

• A bearing primer. The authors expect the reader to have a basicunderstanding of fluid film bearings, their use, and basic operation.They do not describe all types of bearings, nor the evolution oftheir design.

• A thrust bearing tutorial. The authors focus solely on journalbearings, although many of the topics and much of the physics arealso applicable to thrust bearings.

The authors’ primary audience is plant machinery engineersevaluating new versus old designs to fix their problems as well ascentral engineering machinery specialists in charge of selectingand auditing bearing designs for new machinery. Their goal is toprepare these individuals to ask good questions of those perform-ing a bearing design or analysis. Plant engineers must understandthe limits of any analysis so that they can manage risk and assessalternatives.

Bearing designers will also find the material useful in supple-menting their expertise. These individuals must understand theunderlying physics behind the computer program they are running.They too must understand the limitations and risks associated withtheir analysis. They must understand all of the options and inputsto their bearing code, plus understand what the output is tellingthem.

Why are the fundamentals of journal bearing operation andmodeling important?

• Designs are more and more aggressive with less margin forerror.

• Loads and speeds continue to increase in new machinery.

• While the basic fluid dynamics of fluid film bearings are wellunderstood, secondary effects such as elastic deformations, heattransfer to the solids, and turbulence are less well established.

• Innovation breeds new designs and technologies that cause theold analysis methods to fall short.

• The desire for lower power loss and lower oil consumption

• The desire for improved reliability forces better understanding.

• The cost of redesign (trial and error) is enormous.

• The cost of a plant outage is greater.

• You cannot test everything!

How does a poor bearing design manifest itself?

• High bearing metal temperatures, eventually leading to bearingfailure

• High machinery vibrations

• Excessive power loss

• Excessive oil consumption

What are some common operational limits?

• Surface speeds: in the old days, less than 200 ft/s; today, up to450 ft/s

• Unit or specific load, WU: in the old days, less than 250 psi;today, up to 900 psi

• Babbitt lined bearings typically operate below 200°F, whilealternative materials and lubricants can run above 250°F.

• Film thickness values must typically be greater than 0.001 inch,with more aggressive applications above 0.0005 inch.

Figure 1 shows a number of severe applications in successfuloperation today. Twenty years ago, most of these designs wouldhave been ruled out as too aggressive. Today, they perform reliablyas a result of advanced design features and the tools necessary tomodel and predict their performance.

Figure 1. Severe Journal Bearing Applications.

Figure 2 shows an example of one such severe application. Inthis case, an old technology bearing design was replaced with amodern bearing design utilizing a number of advanced features.The results: a 30 degree temperature reduction with better dynamiccharacteristics. This bearing change allowed a multimillion dollarcompressor train to enter service.

Figure 2. Comparison of Bearings with Old and New Tech-nologies.

The first section of this tutorial begins with a discussion on theoperational aspects of fluid film bearings. Bearing geometricalaspects are discussed and the basic physics of fluid film bearingoperation are developed. The second section uses what was learnedin the operational section and describes the means by which onecan model or predict fluid film bearing behavior.


200

400

600

800

1000

150 200 250 300 350 400 450

Surface Velocity (f/s)

Limit CurveTmax < 200 oF

Limit CurveTmax < 220 oF

Older designs limited to250 psi and 200 ft/s

80

90

100

110

120

45 50 55 60 65 70

Inlet Temperature (degC)

Severe Application: N=10,580 rpm, 369 f/s, 298 psi

Results: 30 Degree F Temperature reduction!

New Technology bearing

Old Technology Bearing

200F

230F

122F 131F 140F 149F 158F

OPERATION

The operational characteristics of a journal bearing can be cate-gorized into static and dynamic aspects. Static characteristicsinclude a bearing’s load capacity, pad temperature, power loss, andthe amount of oil it requires during operation. A bearing’s loadcapacity is often measured using either eccentricity ratio thatrelates directly to its minimum film thickness, or the maximum padtemperature. A bearing’s dynamic performance is characterized byits stiffness and damping properties. How these properties interactwith the rotor system determines a machine’s overall vibrationalbehavior.

The main objective of this section is to provide a general under-standing of the basic physics that governs a bearing’s static anddynamic operation. By comparing several common bearingdesigns, the key performance issues of interest will be examined.At the end of this section, one should understand the following:

• Development of hydrodynamic pressure or load capacity

• Relationships between viscous shearing, temperature rise,power loss, and load capacity

• General influence of dynamic coefficients on rotordynamicsincluding stability

• Speed and load dependency of static and dynamic properties

• Different behaviors of fixed geometry and tilting pad bearings

Geometric Parameters

Before discussing the operational aspects of journal bearings,some basic geometric parameters need to be defined. Figure 3(a)shows an arbitrary bearing pad of axial length L and arc length θP.The pad supports the journal of radius RJ rotating at speed ω. Theradial clearance, c = Rb � RJ, allows the journal to operate at someeccentric position defined by distance e and attitude angle Φ. Theattitude angle is always measured with respect to the direction ofthe applied load W and the line of centers. For a fixed geometrybearing, the line of centers establishes the minimum film thicknesslocation. However, this is generally not true for a tilting padbearing. Instead, the trailing edge of a tilting pad often becomes theminimum film thickness point.

Figure 3. Bearing Geometry.

Since typically the journal position relative to the bearing is ofinterest, an eccentricity ratio is defined using the radial clearance:

(1)

At rest, normally the eccentricity ratio E would be expected to be1.0 with the journal sitting on the bearing pad. E can be greaterthan 1.0 if the shaft sits between two tilting pads.

Figure 3(b) defines some other key geometric parameters withrespect to a tilting pad bearing. The pad pivot offset is given by theratio:

(2)

Centrally pivoted pads (50 percent offset or α = 0.5) are the mostcommonly applied. However, 55 to 60 percent pivot offsets areoften seen in high load applications because of their relatively lowpad temperatures (Simmons and Lawrence, 1996).

While the bearing assembly radius Rb determines the largestpossible shaft size that can fit in the bearing, the individual padsmay be machined to a different radius indicated by Rp. These radiialong with RJ establish a very important bearing design parameter,preload or preset, which is defined by two clearances:

(3)

Preload or preset, m, is subsequently defined using their ratio:

(4)

Figure 4 shows how preload affects the relative film shapeswithin the bearing. For m = 0.0, the pad radius and assembly radiusare equal. Typically, preload values are positive, which, as shownin Figure 4, causes the shaft/bearing center (OJ = Ob) to sit lowerrelative to the pad center OP. Thus, one develops the connotationsof preloading the bearing. While many associate preload with onlytilting pad bearings, it can also be used in the design of fixedgeometry bearings. A lemon bore or elliptical bearing is the mostcommon example (Salamone, 1984).

Figure 4. Pad Preload.

All of these geometric design parameters can significantly affecta bearing’s static and dynamic characteristics. For example, tighterclearance and higher preload usually lead to greater load capacityand higher stiffness. Smaller clearance usually means higherBabbitt temperature, etc. In this tutorial, however, the authors willpredominately focus on the influences of the operating parameters,such as shaft speed and bearing load. Excellent discussions on theeffects of various geometric parameters can be found in Jones andMartin (1979) and Nicholas (1994).

Static Performance

To be classified as a bearing, a device must fundamentally carrya load between two components. A journal bearing must accom-plish this task while the shaft rotates and with minimum wear orfailure. Inadequate load capacity leads to either rubbing contactbetween the journal and bearing surfaces, or thermal failure of thelubricant or bearing materials. Therefore, the first step is to explainthe load carrying mechanism in a fluid film bearing.

FUNDAMENTALS OF FLUID FILM JOURNAL BEARING OPERATION AND MODELING 157

RJ

RJ

RB

RP

P

P

RB

(a) (b)

LINE OFCENTERS

Ee

cE E

Ee

cE

e

c

X Y

XX

YY

= = +

= =

2 2

,

α βθ

=p

Pivot Offset

c R R Assembled or Set Bore Clearance

c R R Pad Machined Clearance

b b J

p p J

= −= −

mc

c

b

p

= −⎛

⎝⎜⎜

⎞

⎠⎟⎟1

Load Capacity

Section summary:

• A convergent wedge, surface motion, and viscous lubricant arenecessary conditions to generate hydrodynamic pressure or loadcapacity.

• Typical pressure profiles, journal eccentricity ratios, and center-line loci are shown through examples of three common designs.

• Hydrodynamic forces in fixed geometry bearings have strongcross-coupled components. Such cross coupling is negligible intilting pad bearings due to the pads’ ability to rotate.

In the early 1880s, the underlying physics of how a fluid filmjournal bearing supports a loaded and rotating shaft was a mystery.Using some of the lessons learned from pioneers in this field, theauthors will examine the most important physical phenomenagoverning a bearing’s load capacity: hydrodynamic pressure.

The concept of hydrodynamic lubrication was born from theexperimental work of Beauchamp Tower (1883, 1885).Commissioned to study the frictional losses in railroad bearings(Pinkus, 1987), Tower encountered a persistent oil leak when hedecided to drill an oiler hole in his bearing (Figure 5). After a corkand wooden plug were blown out of the hole, Tower realized thatthe lubricating oil was becoming pressurized. Tower altered hisdesign such that the oil was supplied through two axial groovesthat allowed him to install pressure gauges on the bearing surface.Figure 6 shows an example of the resultant pressures that Towermeasured. Integrating this pressure distribution, Tower discoveredthat it equaled the load he applied on the bearing. In one experi-ment, Tower’s pressure profile integration yielded a film force of7988 lbf compared with the applied load of 8008 lbf, an amazinglyaccurate result (Dowson, 1998).

Figure 5. Tower’s Experimental Bearings. (Courtesy Tower, 1883)

While Tower was conducting his experiments, OsborneReynolds (1886) derived the theoretical justification for the loadcarrying capacity of such journal bearings. He found that a fluid’spressure would increase when it is dragged by a moving surfaceinto a decreasing clearance, like the plane slider situation shown inFigure 7(a). Such a situation demonstrates the governing principleof hydrodynamic lubrication. Without relative motion or a con-verging clearance, no pressure or load capacity will be developed.It is the pressure in the lubricant film that carries the external loadand separates the solid surfaces, which further confirmed Tower’sobservations.

Figure 6. Tower’s Pressure Measurements. (Courtesy Tower, 1885)

Figure 7. Examples of Hydrodynamic Lubrication.

Figures 7(b) and (c) show two other examples of fluid beingdragged into a convergent clearance. The journal creates such con-vergent clearance because of its eccentric operation and/or the radiidifference between it and the pad. For a perfectly centered journalwith a zero preloaded pad, the inlet film thickness equals theminimum or outlet film thickness (hi = ho) and no pressure wouldbe expected to be developed in the film. The phenomenon ofhydroplaning is another good example. Here the tire deformationcreates a converging clearance that generates enough pressure inthe water film to support the weight of the car. Both situations canbe treated like the plane slider with hi > ho.

To understand the pressure distributions and the film forcesdeveloped in some typical journal bearing designs, here the authorsexamine a two axial groove bearing [often referred to as a plainjournal bearing (Salamone, 1984)], a pressure dam bearing, and atilting pad bearing with four pads [Figure 8(a)]. Running at thesame diameter, axial length, bore clearance, preload, oil viscosity,and speed, Figure 8(b) displays each bearing’s circumferentialpressure distribution. Each bearing has its journal position fixeddownward halfway within the clearance at EX = 0.0 and EY =�0.5.

The two axial groove’s pressure distribution in Figure 8(b) has apeak pressure over 750 psig. It is important to notice that thepressure distribution is not symmetric about this peak.Furthermore, the peak pressure does not occur at the minimumfilm thickness position (270 degrees) where one would instinc-tively anticipate. These two pressure distribution characteristicsare fundamental to all bearing types where hydrodynamicpressures are developed.

No pressure is developed in the upper half of the bearingbecause of the diverging clearance and the relatively low oil supplypressure (20 psig). This condition, which exists in most fixedgeometry bearings, causes the film to cavitate and restricts thepressure in the film to the vapor pressure of the lubricant.Physically on a cavitated pad, the fluid film is ruptured and therotating shaft drags streamlets across this region (Heshmat, 1991).The film’s positive pressure area results in a 1374 lbf effectiveforce on the shaft at an angle of 56 degrees. Recalling that thejournal was displaced vertically downward only, one should noticethat the fluid film has now generated a responding force with a


Figure 8. Pressure Profiles for Three Bearing Designs.

significant horizontal component. Such a reaction is the mainreason plain journal bearings create such dynamic stabilityconcerns. A cross coupling is experienced since movement in onedirection causes a force component in the perpendicular direction.This behavior will be discussed further with respect to dynamiccharacteristics.

Since its lower half film profile is the same as that of the twoaxial groove bearing, the pressure dam bearing has an identicalpressure distribution in its lower half. However, because of thepresence of the dam, the upper half now has a converging wedgethat generates a positive pressure profile. The peak pressure occursat the dam location. With the upper half pressure counteracting thepressure developed in the lower half, the force exerted on the shafthas slightly reduced in magnitude to 1162 lbf and rotated more hor-izontally in direction.

Film pressure is developed on the bottom two pads of the tiltingpad bearing. At steady-state, the moments on a pad must bebalanced in order for the pad to reach an equilibrium tilt angle.When this occurs, the pad pressure force vector passes directlythrough the pivot. Unlike the two axial groove and pressure dambearings, the tilting pad has produced a resultant film force that isalmost purely vertical. In this case, a vertical displacement of thejournal has resulted in an almost directly vertical force. Thisdesirable characteristic totally relies on the pad’s ability to tilt eventhough the tilt angles are very small (on the order of 0.01 degree).Boyd and Raimondi (1953) were among the first to explain thisbehavior. Both they and Hagg (1946) realized the implicationsfrom the dynamics standpoint, which will be discussed later.

As a final point on Figure 8(b), one should notice the additionalreduction in the tilting pad bearing’s film force magnitude (1109lbf versus 1374 lbf for the two axial groove bearing). This isexpected since the pad area that carries load has been reduced. Thetwo axial groove bearing has 150 degrees of lower half pad arclength, while the tilting pad bearing has only 2 � 72 degrees = 144degrees with a supply groove in between.

While good for demonstrating film forces and their nonlinearnature, setting the journal at a fixed eccentricity like in Figure 8(b)does not represent a realistic operating condition. In reality, thebearing will adjust the shaft’s position till the hydrodynamic forcebalances the applied load W. For the same three bearings, Figure8(c) shows the pressure profiles when a constant load (W = 1100lbf) is applied downward at 270 degrees. Also shown are theresultant shaft eccentricity ratios where the shaft has reached itssteady-state equilibrium position.

Comparing Figures 8(b) and (c), the journal inside the two axialgroove bearing has now had to shift horizontally in order to createa film profile that only opposes the vertical load. This is evident inthe more vertical orientation of the pressure distribution. Similar

behavior is observed for the pressure dam bearing. However,because of the pressures created by the dam and its angular orien-tation, the shaft reaches a position of higher eccentricity andgreater attitude angle than the plain journal bearing. Both fixedgeometry bearings are able to support the load at a lower eccen-tricity than the tilting pad bearing. This higher load capacity isexpected when one recalls the resultant film forces created inFigure 8(b).

One should now have a basic feel for the pressures developed byhydrodynamic lubrication. Through different bearing geometries,one has seen how different converging wedges create differentpressure distributions, and, thus, various abilities to support load.What has not been emphasized is the importance of the lubricant’sviscosity that determines the pressure generation just as much asthe bearing’s geometry.

As the next section will describe in detail, the lubricant’sviscosity will decrease because of the internal heat generatedduring operation. Discussions and comparisons, so far, have keptthe viscosity constant or isoviscous. With this restriction removed,Figure 9 demonstrates that a bearing’s load capacity is a strongfunction of its operating condition. In Figure 9(a), the load is fixedat 1100 lbf and the shaft speed varies from 1000 rpm to 19,000rpm. When stationary, the shaft sits on the bottom with zeroattitude angle and unity eccentricity ratio (for the tilt pad bearing,the eccentricity is slightly higher because the shaft rests betweenthe pads). As the shaft accelerates, the journal is lifted higher andhigher by increasing hydrodynamic pressure.

Figure 9. Load Capacity Trends Allowing for Viscosity Degrada-tion Effects .

Although all three bearings exhibit this general trend, differentloci of the journal center are observed for each bearing. For the plainjournal bearing, the journal center moves approximately along acircular arc. With increasing speed, the journal gradually approachesthe bearing center because it requires less and less of a convergentwedge to produce a 1100 lbf hydrodynamic force. The pressure dambearing behaves similar to the plain journal bearing at low speeds.At high speeds, the dam generates significant hydrodynamicpressure that pushes the journal away from the bearing center.

For the tilting pad bearing, the journal center is directly lifted inthe vertical direction maintaining very little attitude angle. Suchsmall attitude angles are only possible because of the pads’ abilityto tilt. Figure 10 shows an example of what occurs when this tiltingability is lost. Here, a noticeable attitude angle was observed whenthe loaded pad was locked. When this is encountered, it may beattributable to pivot design, operating conditions, or even thermo-couple or resistance temperature detector (RTD) wiring problems.

Figure 9(b) reverses the situation, keeping speed constant andvarying load. Like a speed increase, load reduction allows thejournal position to reach a lower eccentricity. At 100 lbf, thejournal is almost perfectly centered for both the tilting pad and twoaxial groove bearings. Once again, because of the dam, thepressure dam bearing maintains a higher eccentricity ratio even atthis light loading.


(a) (b)

Figure 10. Shaft Centerline Test Data for a Tilting Pad Bearingwith a Locked and Unlocked Loaded Pad. (Courtesy Brechting,2002)

If the lubricant’s viscosity was not allowed to change, the cen-terline trends in Figure 9(a) and (b) would be identical. In otherwords, increasing speed and decreasing load would be equivalent.This is why the dimensionless Sommerfeld number S, whichcombines speed and load effects, was often used to define bearingsimilarity in early isoviscous studies. The Sommerfeld number isstill used today to compare bearings and is typically defined as:

(5)

One should note that it does not include all geometry factors suchas preload or pivot offset. Since thermal and deformation effectsare also absent, caution must be used when comparing bearing per-formance using this rather simplified relationship.

Since the external load is a vector that has direction, journalsteady-state position is as much dependent on the load’s directionas it is on its magnitude. Figure 11 shows the resultant pressureprofiles and eccentricity ratios when the 1100 lbf load is nowdirected horizontally. Because the load is now pushing toward theiraxial groove, both fixed geometry bearings’ pressure areas are dra-matically reduced. Likewise, their load capacity is reduced asevidenced by their higher shaft eccentricity ratios. Tower came tothe same realization during his experiments. The tilting padbearing, however, achieves the same eccentricity ratio as before.This can be attributed to its symmetry (four pads equally distrib-uted) and each pad’s ability to tilt and generate a load carryingpressure. If the load was directly on pad, one would expect somereduction in load capacity versus the between-pad loading (Boydand Raimondi, 1953; Jones and Martin, 1979).

Figure 11. Pressure Distributions with a Horizontal Load (W =1100 LBF, N = 7000 RPM).

Load capacity is of great concern in slow roll, turning gearoperation with rotational speeds around 10 to 15 rpm. At such lowspeeds, the lubricant is unable to generate much supportingpressure, resulting in a very thin film likely in the regime of

boundary lubrication. Compared to hydrodynamic lubrication, themating surface roughnesses in boundary lubrication becomeimportant and the lubricant film shows increased friction coeffi-cient (Elwell and Booser, 1972; Gardner, 1976). However, sincethe shaft typically does not vibrate at such low speed, boundarylubrication does not necessarily mean bearing failure. A generallyaccepted criterion is that the minimum film thickness must be atleast twice the surface roughness to ensure successful operation.

Viscous Shearing and Temperature Rise

Section summary:

• Viscous shearing causes temperature rise.

• Temperature rise affects bearing performance through lubricantviscosity reduction and solid deformations.

• Shaft speed is the primary operating factor compared to load.

While producing load carrying capacity, the lubricant film alsogenerates heat that causes temperature rise in operation. It is wellknown that a lubricant’s viscosity is extremely sensitive to temper-ature. Table 1 provides some indication for several commonturbine oils. Figure 9 has already shown some thermal effects: dueto different temperature rises, increasing speed and decreasing loadare not equivalent, and the centerline trend in Figure 9(a) differsfrom that in (b). To fully understand the thermal effects of journalbearings, one must grasp the principle of viscous shearing, whichis their heat generation mechanism.

Table 1. Lubricant Viscosity at Different Temperatures.

Figure 12(a) shows the flow of lubricant being sheared by twoparallel surfaces. Since the lubricant adheres to both surfaces, itremains stationary on the upper surface and moves at the samevelocity as the lower plate. For laminar flow, layers of lubricantmove smoothly and the velocity profile is a straight line. In case ofthe convergent film within a bearing, Figure 12(b) shows that theactual lubricant flow is a little more complex. Nevertheless, theshearing type flow is still dominant unless the journal eccentricityis very high. This shearing motion creates frictional stressesbetween the lubricant layers. Per Newton, the fundamental relationfor fluid friction (as a stress) takes the form τ = μ(du/dy). Using theparallel plate model, it can be simplified to τ = μU/h. Thus, increas-ing lubricant viscosity or shaft speed increases the viscous shearingand, consequently, heat generation. This heat generation due toviscous shearing impacts a bearing’s performance in several ways:

• Reduction in lubricant viscosity due to increased temperature

• Thermal growth and distortion of surrounding surfaces affectingthe film shape

• Heating of the lubricant and bearing materials toward theirthermal failure limits

Figure 13(a) demonstrates the influence of shaft speed onbearing temperature rise. For the two axial groove bearing, as theshaft accelerates from 1000 rpm to 19,000 rpm, the peak pad tem-perature substantially increases from 125°F to 230°F, which is nearthe failure limit. The operating viscosity is consequently reducedaccording to Table 1. Because of this viscosity reduction, speedincreases are less and less effective in producing hydrodynamicpressure to lift the journal. This is apparent in Figure 9(a).Meanwhile, since heavier load results in smaller h on the loadedpad, the external load also affects pad temperature. However, asshown in Figure 13(b), its thermal influence is substantiallyweaker compared to the shaft speed.


Absolute Viscosity (Reyns = lbf-s/in2) Temperature

(°F) ISO VG

32 ISO VG

46 ISO VG

68 104 3.75e-6 5.42e-6 8.06e-6 212 5.98e-7 7.68e-7 9.99e-7

SW

R

cU

= ⎛⎝⎜

⎞⎠⎟

μω 2

Figure 12. Shearing Flows in Lubricant Film.

Figure 13. Maximum Pad Temperature Versus Speed and Load.

Considering the thermal effects on lubricant viscosity, Figure 14presents pressures and journal positions of the same three bearingsunder the same operating condition. Compared to the isoviscousresults in Figure 8(c), the pressure profiles do not show significantchanges because the sums of the pressures must still equal the 1100lbf applied load. The thermal effects on load capacity are mostevidently shown by the new journal equilibrium positions. The twoaxial groove bearing’s eccentricity ratio has increased from 0.32 to0.53, and the attitude angle also decreased by about 10 degrees. Forthe tilting pad bearing, the journal has moved vertically downwardfrom a position of 0.5 eccentricity ratio to a 0.7 position. Thejournal position drop is the result of reduced load capacity becauseof viscosity reduction due to shearing heat generation. It can alsobe explained as the following: to generate the same force with aless viscous oil, the bottom pad needs to have a smaller clearanceand larger wedge ratio hi/ho, which is achieved by the increasedjournal eccentricity.

Power Loss

Section summary:

• Mechanical energy is converted into heat through viscousshearing.

• Shaft speed is the primary operating factor compared to load.

Figure 14. Pressure Distribution Comparison, Variable Viscosity(W =1100 LBF, N = 7000 RPM).

The increased temperature in the fluid film is the result ofmechanical work done by the shaft. In turn, friction caused by theshaft shearing the lubricant produces a resistive torque on theshaft and consumes mechanical power. This friction loss isclosely related to a bearing’s size, clearance, shaft speed, and oilviscosity. A bearing’s size dictates the area of shearing.Therefore, the partial arc design, which eliminates the top pad ofa plain journal bearing, is often used to minimize friction loss(Byrne and Allaire, 1999). As shown in Figure 15, power lossgrows with increasing shaft speed. And the partial arc bearingsaves noticeable amounts of horsepower, especially at highspeeds. In recent years, an industrial trend is to use directly lubri-cated bearings with reduced supply oil to decrease power loss.Similar to the partial arc design, this practice effectively reducesa bearing’s shearing area through starvation, which will bediscussed later in the modeling section.

Figure 15. Friction Power Loss Versus Shaft Speed.

Petrov (1883) conducted pioneering work on viscous frictionand proposed Petrov’s Law, which is still used as a quick estimatefor bearing power loss. He estimated the frictional torqueaccording to:

(6)

Power loss predictions using Petrov’s law turn out to be liberalbecause the shaft is assumed centered (unloaded) within thebearing clearance. Since a bearing is always loaded statically anddynamically, it has more friction loss according to Figure 13(b)(temperature rise is the result and indicator of mechanical energyloss). Turbulent flow also increases friction loss due to additionaleddy stresses.


U

(a) Lubricant Shearing Between Surfaces

U

Peak Pressure

Shear

ShearPressure

Pressure

u=Uu=U

u(y)u(y)

(b) Lubricant flows within a Bearing Convergent Wedge

Shaft Speed (E+3 rpm)

Tm

ax(D

eg

F)

0 2 4 6 8 10 12 14 16 18 20120

130

140

150

160

170

180

190

200

210

220

230

240

250 Maximum Pad TemperatureTwo Axial Groove Bearing, W=1,100 lbf

Load (lbf)

Tm

ax(D

eg

F)

0 1000 2000 3000 4000120

130

140

150

160

170

180

190

200

210

220

230

240

250 Maximum Pad TemperatureTwo Axial Groove Bearing, N=7,000 rpm

(a) (b)

Shaft Speed (E+3 rpm)0 2 4 6 8 10 12 14 16 18 20

0

1

2

3

4

5

6

7

8

9

10

2 Axial GroovePartial Arc

Friction Power Loss, W=1,100 lbf

( ) ( )[ ]( )

TR L U

c

RLU

cR Surface Area Shear Stress Moment Arm

friction =

= ⎛⎝⎜

⎞⎠⎟

⎡

⎣⎢

⎤

⎦⎥ ⋅ =

2

2

2π μ

π μ

Supply Flowrate

Section summary:

• Oil supply is necessary to maintain steady-state operation.

• Many older bearings work in a flooded condition.

• Many newer bearing designs use directed lubrication.

Another important static parameter is the oil supply flowratethat the bearing requires. Lubricant flows into a pad at its leadingedge, exiting at its trailing edge and axial ends. The majority ofthe oil leaving the trailing edge enters the next pad and continuesto circulate inside the bearing. However, the oil flowing axiallyleaves the bearing through the end seals and drains, and must bereplenished by fresh oil from the lube system. Therefore, thesupply flowrate needs to be at least equal to this side leakagerate. Using the previous two axial groove bearing as an example,Figure 16 plots the minimum required flowrate as functions ofshaft speed and applied load. Since either higher speed orheavier load leads to stronger hydrodynamic pressure, the sideleakage, driven by the film pressures, increases as a result. Inpractice, the supply flowrate is often larger than this minimumrequirement to maintain the temperature rise between the oilsupply and drain within recommended limits (typically between40 to 60°F). However, this is only helpful in reducing the mixedsump temperature in a flooded bearing. Methods to determinethe supply flowrate and inlet orifice size can be found inNicholas (1994).

Figure 16. Minimum Required Supply Flow Versus Speed andLoad.

One should also bear in mind that lubricant is dragged into thebearing clearance by shaft rotation, not pumped into it by highsupply pressure. Therefore, the function of the oil pump is simplyto send enough oil into the bearing and keep it circulating.Furthermore, the bearing clearance will accept only a finitequantity of oil; supply less and the bearing will be starved, supplymore and the extra oil will fall to the sides and be wasted. Theamount of oil that the bearing clearance requires is simply afunction of the shaft speed, clearance, and minimum filmthickness. Again, supplying more oil than the bearing clearancecan consume will not effectively reduce pad temperatures and issimply a waste of oil flow.

The basic working principles governing the steady-stateoperation of fluid film journal bearings are summarized in Figure17. A convergent wedge, a moving surface, and viscous lubricantare the three ingredients necessary to generate the film hydrody-namic forces to support the applied load. An accompanyingphenomenon is viscous shearing that causes temperature rise andpower loss. The temperature rise and power loss are relatedbecause the energy used to heat up the film is converted from theshaft mechanical energy. Increased temperatures lead to oilviscosity reduction and bearing deformation. In turn, the defor-mations also change the bearing geometry and, thus, the wedgeshape.

Figure 17. Basic Working Principles Within a Fluid Film Bearing.

Dynamic Performance

Section summary:

• A bearing is a component of an integrated dynamic system.

• A bearing can be dynamically represented as springs anddampers in a linearized model.

• Stiffness and damping coefficients have significant rotordy-namic implications.

• Dynamic coefficients are dependent on shaft speed and appliedload.

• There are two types of instability related to bearings: oil whirland shaft whip.

Desirable steady-state operation, where the bearing is runningwith sufficient load capacity and acceptable temperatures, helps toensure the long-term reliability of the bearing itself. However, thebearing’s dynamic properties must also be acceptable for theoverall machine’s reliability. This is because a bearing’s dynamicproperties, in conjunction with dynamics of the rest of the rotorsystem, govern all aspects of a machine’s vibrational performance.

The dynamic performance of journal bearings first came underscrutiny because of the vibration problems encountered byNewkirk and Taylor (1925). In this landmark paper, Newkirk andTaylor describe the first published account of a rotor goingunstable due to “oil whip.” Initially, they thought the vibration wascaused by improper shrink fits, which were the only known sourceof whipping instability (Newkirk, 1924). They eventually foundthat the bearing’s parameters such as clearance (Figure 18),loading, alignment, and oil supply (in some tests, the supply wascut off!!) controlled the instability.

Figure 18. Newkirk and Taylor’s Oil Whip Measurements.(Courtesy Newkirk and Taylor, 1925)

With the considerable development of steam turbine technology,the 1920s continued to provide evidence that a machine’s vibrationwas heavily linked to the operation of the bearings. In two papers,Stodola (1925) and one of his pupils, Hummel (1926), introduced


Rotational Speed (rpm)

Min

imu

mR

equ

ired

Su

pp

lyF

low

(gp

m)

0 5000 10000 15000 200000

0.5

1

1.5

2

2.5

3

3.5

4Minimum Required Supply Flow (Side Leakage)Two Axial Groove Bearing, W=1,100 lbf

Total Load (lbf)

Min

imu

mR

equ

ired

Su

pp

lyF

low

(gp

m)

0 1000 2000 3000 40000

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

Minimum Required Supply Flow (Side Leakage)Two Axial Groove Bearing, N=7,000 rpm

(a) (b)

Convergent Wedge Moving Surface Viscous Lubricant

Hydrodynamic Force Viscous Shearing

Temperature Rise Power Loss

Deformations Viscosity Reduction

the concept that a bearing’s oil film dynamically acts like a spring.They found that, when this oil spring’s stiffness was considered,their rotor critical speed calculations could be improved (Lund,1987). They also realized that the oil film stiffness could be verynonlinear, i.e., the film force variation was not directly propor-tional to the journal position variation. While many continue tostudy this nonlinear complication, most of the machines inoperation today were designed using linearized stiffness anddamping properties for the oil film.

As shown in Figure 19, the fluid film can be represented bysprings and dampers. The static load W, such as gravity, establishesthe journal’s steady-state equilibrium position. Then, some dynamicforce, such as rotor unbalance forces, pushes the journal away fromits equilibrium and causes it to whirl on an elliptical orbit (Figure20). To have an acceptable vibration level, the orbit’s size must berelatively small compared to the bearing clearance. When this is thesituation, the vibration is said to be in the linear range and the filmdynamic forces are directly proportional to the displacements (Δx,Δy) and associated velocities (Δx

., Δy

.). This relationship is given by:

(7)

Figure 19. Dynamic Properties of the Fluid Film.

Figure 20. Journal Steady-State Position and Orbit.

where the Kij and Cij are called linearized stiffness and dampingcoefficients, respectively. In other words, at an instantaneousjournal position, the horizontal and vertical forces due to the oilfilm can be obtained by expanding Equation (7):

(8)

where the negative sign implies that the force is acting on the rotor.To justify the use of linearized dynamic properties, one should

understand the other situation where the vibration levels are rela-tively large. Figure 21 shows the classic example of an unstableshaft where the orbit nearly fills up the entire bearing clearance.Here the rotor has almost reached the so-called “limit cycle.” Sincethis motion is large relative to the bearing clearance, linearizedcoefficients are inadequate to represent the film dynamic forces.Therefore, they cannot be used to predict the actual amplitudes forsuch large vibrations. However, the strength of the linearized coef-ficients is their ability to predict whether or not such unstablevibrations will occur. This ability, combined with their accuracy inpredicting vibration amplitudes within the range of interest [up to40 percent of the clearance according to Lund (1987)], enablesmodern rotordynamics to be firmly based on their use.

Figure 21. Unstable Rotor Exhibiting Large Vibrations.

Now, let us define and explain those linearized dynamic coeffi-cients in Equations (7) and (8). The following two stiffnesscoefficients are called principal or direct coefficients:

(9)

where each relates the change in force in one direction due to a dis-placement in the same direction. In other words, these directstiffnesses provide a restoring force that pushes the journal backtoward its steady-state equilibrium position. As shown in Figure22(a), a positive horizontal perturbation Δx generates a negativehorizontal force Fxx = �Kxx(Δx), a negative vertical perturbation�Δy yields a positive vertical force Fyy = �Kyy(�Δy). The com-bination is a radial force that tries to push the journal back to Os.

Figure 22. Dynamic Forces in the Fluid Film.


F

F

K K

K K

x

y

C C

C C

x

y

x

y

xx xy

yx yy

xx xy

yx yy

⎧⎨⎪

⎩⎪⎫⎬⎪

⎭⎪= −

⎡

⎣⎢⎢

⎤

⎦⎥⎥

⎧⎨⎩

⎫⎬⎭

−⎡

⎣⎢⎢

⎤

⎦⎥⎥

⎧⎨⎩

⎫⎬⎭

ΔΔ

ΔΔ�

�

( )( )

F K x K y C x C y

F K y K x C y C x

x xx xy xx xy

y yy yx yy yx

= − + + +

= − + + +

Δ Δ Δ Δ

Δ Δ Δ Δ

� �

� �

K F x Horizontal incipal Stiffness

K F y Vertical incipal Stiffness

xx x

yy y

==

Δ ΔΔ Δ

/ Pr

/ Pr

x

y

Kyy� y

Kxx� x

FK

a) DIRECT STIFFNESSFORCE

FC

b) DIRECT DAMPINGFORCE

x

y Kyx� x

Kxy� y

FCCK

c) CROSS COUPLED STIFFNESS FORCE

x�Δ

y�Δ

xCxx �Δ

yCyy �Δ

OS OS OS

The principal stiffnesses are extremely important with respect tothe machine’s vibration performance. Their magnitude, relative tothe shaft stiffness, governs the location and amplification of therotor’s critical speeds. They are equally important for stabilitypurposes. Large asymmetry of Kxx and Kyy is the main cause forsplit critical speeds (API Standard 684, 2005) and noncircular(elliptical) orbit shapes. Although such asymmetry can be verybeneficial with respect to stability (Nicholas, et al., 1978),symmetry of these coefficients is usually preferred for unbalanceresponse considerations.

Two principal or direct damping coefficients are also present:

(10)

Here the damping coefficients relate the change in force due to asmall change in velocity. Because of the rotor’s whirling motion,the combination of these two principal damping coefficientsproduces a force that is tangential to the vibration orbit.Furthermore, as shown in Figure 22(b), this direct damping forceacts against the whirling motion, helping to retard or slow it.

Like their direct stiffness counterparts, the principal dampingterms dictate much about the machine’s unbalance response andstability. They are often the predominant source of damping in theentire machine. However, their effectiveness in reducing criticalspeed amplification factors and preventing subsynchronous insta-bilities is determined also by the bearing’s direct stiffnesscoefficients as well as the shaft stiffness. For the direct damping tobe effective, the bearing cannot be overly stiff because thedamping force relies on journal motion. Also, contrary to one’sinitial instincts, large amounts of damping can actually be detri-mental. Barrett, et al. (1978), in an important fundamental paper,highlighted this fact and verified that the optimum amount ofbearing damping is a function of the bearing (direct) and shaft stiff-nesses.

The off-diagonal stiffness coefficients in Equation (7), Kxy andKyx, are the infamous cross-coupled stiffness coefficients. Themeaning of cross coupled becomes apparent when these stiffnessesare defined as:

(11)

As an example, the coefficient Kyx relates a vertical force due to ahorizontal displacement. Thus, the horizontal and vertical direc-tions have become coupled. This exactly corresponds to thebehavior highlighted in Figure 8 for the two fixed geometrybearing, where a displacement in one direction resulted in forcecomponent perpendicular to this displacement.

Almost all structures have such cross-coupled stiffness terms butmost are symmetric in nature where Kxy = Kyx. Rotor systems areunique in that this symmetry usually does not exist (Kxy ≠ Kyx andusually Kxy > 0, Kyx < 0). Fundamentally, their presence and theirasymmetry result from the various fluids rotating within a turbo-machine, such as oil in bearings and gas in labyrinth seals. Figure22(c) illustrates why asymmetric cross-coupled stiffnesses aredetrimental. Instead of opposing the rotor’s whirling motion likethe direct damping, the cross-coupled stiffnesses combine to createa force pointing in the whirl direction, promoting the shaftvibration.

When the direct damping force is unable to dissipate the energyinjected by the cross-coupled stiffness force, the natural frequency(typically the lowest one with forward whirling direction) willbecome unstable, causing the shaft to whirl at this frequency(Ehrich and Childs, 1984). This frequency will appear in thevibration spectrum, typically as a subsynchronous component.Such self-excited vibration is the reason why these cross-couplingcoefficients are of such concern for stability purposes.

Unlike their fixed geometry counterparts, tilting pad bearingsproduce very little cross-coupled stiffness, which explains theirpopularity. This fact was actually touched on earlier in Figures 8and 9 where the tilting pad bearing’s journal position moved onlyvertically under a vertical load. It consistently maintains a smallattitude angle, indicating a small amount of cross-coupled stiffnesspresent. As shown in Figure 9, the attitude angle of the two fixedgeometry bearings approaches 90 degrees at light loads, implyingthe cross-coupled stiffnesses are very large relative to their directcounterparts. Thus, instability is often encountered when runningfixed geometry bearings at light loads.

Like the static performance, the dynamic coefficients vary withshaft speed and external load. Figure 23 shows the two axialgroove bearing’s coefficients as functions of the shaft speed. Thevertical stiffness Kyy and damping Cyy decrease significantly as thespeed increases from 1000 to 10,000 rpm. Meanwhile, the bearingbecomes less stable because it loses considerable damping whileretaining a high cross-coupled stiffness. Therefore, an unbalanceresponse analysis must include the coefficients’ variations foraccurate prediction of critical speeds and vibration levels. Suchspeed dependency is also the reason why the amplification factorof the first critical speed should not be used as a measure ofstability for higher speeds like maximum continuous speed.

Figure 23. Two Axial Groove Bearing Dynamic Coefficients VersusSpeed.

The stabilities of the plain journal and pressure dam bearings arecompared in Figure 24. The tilting pad bearing is not presentedbecause its stability is not an issue. Here, stability is measured bythe rigid rotor threshold speed, which excludes the rotor effects andis solely dependent on the bearing properties and loading (Lund andSaibel, 1967). As shown in Figure 24, the plain bearing is predictedto be unstable at around 10,000 rpm using the bearing coefficientsat 700 rpm; using the coefficients at 10,000 rpm, instability ispredicted at 7500 rpm. Therefore, a rigid rotor would go unstable at7600 rpm where the curve intersects the 1� line. The pressure dambearing shows improved stability since the intersection is beyond10,000 rpm. Experimental results and more discussions can befound in Lanes, et al. (1981), and Zuck and Flack (1986).

Figure 24. Rigid Rotor Stability Threshold Speeds Versus Shaft Speed.


X X XX X X X X X X

Shaft Speed (rpm)

Stiff

ne

ss

(lb

f/in

)

0 5000 10000 15000 200000

500000

1E+06

1.5E+06

2E+06

2.5E+06

Kxx

Kxy

-Kyx

Kyy

X

Two Axial Groove BearingConstant Load: 1100 lbf at 270 deg

X

XX X X X X X X X

Shaft Speed (rpm)

Da

mp

ing

(lb

f-se

c/in

)

0 5000 10000 15000 20000

5000

10000

15000

20000

25000

30000

Cxx

-Cxy = -Cyx

Cyy

X

Two Axial Groove BearingConstant Load: 1100 lbf at 270 deg

Shaft Speed (E+3 rpm)

Th

resh

old

Spe

ed(E

+3

rpm

)

0 2 4 6 8 100

2

4

6

8

10

12

14

16

18

20

Two Axial Groove

Pressure Dam

1x

7600 rpm

C F x Horizontal incipal Damping

C F y Vertical incipal Damping

xx x

yy y

==

Δ ΔΔ Δ

/ � Pr

/ � Pr

K F y

K F x

xy x

yx y

=

=

Δ Δ

Δ Δ

/

/

Examining all the ways a bearing’s dynamic properties caninfluence a machine’s rotordynamics is beyond the scope of thistutorial. The literature on the subject is extensive and the reader isencouraged to examine API Standard 684 (2005) for further expla-nation and references. Fundamentally, a machine’s rotordynamicperformance becomes an interplay of how the dynamics of variouscomponents (bearings, shaft, seals, supports, etc.) interact whencombined together as a system. In other words, the overall dynamicperformance is governed by the system, not one particularcomponent in general.

Oil whirl is one exception where the bearing’s dynamic proper-ties dominate the rotordynamic behavior of the system. Firstobserved by Newkirk and Taylor (1925) who called it “journalwhirl,” this instability phenomenon has received considerableinterest even though its occurrence is rare in most machinery appli-cations. Some exceptions are gearboxes and internally gearedcompressors operating at low power and resulting in small gearforces.

As shown in Figure 25, the frequency of the system’s firstforward mode follows the 0.5� line at low shaft speeds. When thesystem becomes unstable at such low speed, the frequency of thesubsynchronous vibration equals half of the running speed andtracks it as the rotor accelerates. Since the shaft does not experi-ence much bending, it can be regarded as a rigid body or just amass inside the bearing. Therefore, the oil whirl instability isdominated by the dynamic properties of the bearing. With increas-ing shaft speed, the unstable mode steps into the territory of shaftwhip where the subsynchronous frequency is locked at a constantvalue. Unlike oil whirl, the rotor’s mode shape in whip undergoesnoticeable bending and its flexibility plays a significant role in thesystem’s overall dynamics.

Figure 25. Campbell Diagram Showing Oil Whirl and Shaft Whip.

In real life, most unstable machines exhibit shaft whip directlywithout exhibiting oil whirl behavior. To produce the 0.5� oilwhirl, the bearing must be unloaded, allowing it to operate at verylow eccentricity. Hamrock (1994) theoretically deduced that oilwhirl would occur if the bearing had a constant pressure (zero)throughout the film. Obviously, a constant (zero) pressure can onlybe achieved with a centered or unloaded shaft. The unloading maybe caused by the lack of gravity load like Newkirk and Taylor’s

vertical rotor (1925), the use of some “centering device”(Muszynska and Bently, 1995), misalignment, overhung masseffects, or the presence of external forces from gearing or partialarc steam admission forces that can negate the gravity loading.Again, oil whirl is driven by large bearing cross coupling, and thusoccurs only with fixed geometry bearings.

MODELING

Accurate evaluation of a bearing’s performance has become avital factor in the design, operation, and troubleshooting of rotatingmachinery. A number of computer programs have been developedto accomplish this task. This section presents the major aspects ofbearing modeling, the mainstream techniques used in thosecomputer codes as well as their potential limitations. First, thegeneral areas that are required in most modern bearing analysiswill be covered. Such areas include:

• Hydrodynamic pressure

• Temperature

• Deformations

• Turbulence

• Dynamic coefficients

The next step is to assemble those components into a functionalcomputer algorithm. Then, discussion will switch to special situa-tions such as direct lubrication and starvation. Some modelingdifficulties and challenges will be addressed at the end. Theobjective is to shed some light on those computer tools, and, thus,help engineers to use them properly.

Hydrodynamic Pressure

Section summary:

• Hydrodynamic pressure is the primary physical phenomenon tomodel.

• The Reynolds equation is the governing equation for thinlubricant film.

Hydrodynamic pressure modeling is the foundation of anaccurate bearing analysis. In general fluid dynamics, the film’spressure, and velocity distributions are governed by the coupledcontinuity equation and the momentum equations. The continuityequation comes from the basic law of mass conservation. Each ofthe three momentum equations, known as Navier-Stokes equationsfor incompressible flow, is essentially Newton’s second law ineach direction of the three-dimensional space. Thus, a simple the-oretical analysis requires simultaneous solution of four equations,which is not trivial because iterations must be employed and themomentum equations are nonlinear. If other parameters such astemperature and turbulence are considered, more equations mustbe added to the formulation and the solution procedure quicklybecomes very complex. Fortunately, such a procedure can beavoided in a bearing analysis and the hydrodynamic pressure canbe directly calculated from the following linear equation.

(12)

Equation (12) is the classic Reynolds equation (Reynolds,1886). During his derivation, Reynolds had to make severalassumptions. Most of all, he utilized the fact that the film thicknessis much smaller than the bearing’s diameter (the typical c/D ratiois on the order of 10�3). Consequently, the momentum equationscan be significantly simplified by neglecting the small terms. Fromthese simplified equations, the pressure across the film is shown tobe constant and the velocity components can be directly solved.


F

F

F

F

FF

F F F FB B B B B

Shaft Speed (rpm)0 2000 4000 6000 8000 10000 12000 14000

0

1000

2000

3000

4000

5000

6000 0.5x1x

Oil Whirl Shaft Whip

∂∂ μ

∂∂

∂∂ μ

∂∂

∂∂x

h p

x z

h p

z

U h

x

x essure z essureShear

3 3

12 12 2

⎛

⎝⎜⎜

⎞

⎠⎟⎟ +

⎛

⎝⎜⎜

⎞

⎠⎟⎟ =

Pr Pr� ��

Then, the Reynolds equation is obtained by substituting thevelocity components into the continuity equation and integratingacross the film (Szeri, 1979). This classic Reynolds equation alsoassumes that viscosity is invariant across the film and the flow islaminar. However, Equation (12) can be made more general byrelaxing these two conditions (Constantinescu, 1959; Dowson,1962).

The left hand side of Equation (12) includes the pressure flowterms that represent the net flowrates due to pressure gradientswithin the lubricant area; the right hand side is the shear flow termthat describes the net entraining flowrate due to the surfacevelocity. If the shaft is stationary (U = 0), the right hand side equalszero and no lubricant can enter the bearing clearance. The pressuregradients on the left hand side must be zero to satisfy the flow con-tinuity. Similarly, if the pairing surfaces are parallel (∂h/∂x = 0),the right hand side also becomes zero and no hydrodynamicpressure can be developed. Therefore, it is shown from theReynolds equation that the rotating shaft and convergent wedge arethe necessary conditions to generate hydrodynamic pressure. If theclearance is divergent (∂h/∂x > 0), the Reynolds equation will giveartificially negative pressure. However, since the lubricant cannotexpand to fill the increasing space, cavitation occurs in this regionand the divergent clearance is occupied by streamlets and vapor-liquid mixture (Heshmat, 1991).

Figure 26 shows a two axial groove bearing and the bottom padpressure distribution solved from the Reynolds equation.Hydrodynamic pressure is smoothly developed in the area of con-vergent clearance. Axially, the pressure distribution is symmetricabout the midplane and goes down to the ambient pressure at theedges. No hydrodynamic pressure is generated in the cavitatedregion that exists near the trailing edge of the bottom pad and onthe entire top pad.

Figure 26. Pressure Distribution on the Bottom Pad of a Two-AxialGroove Bearing.

Temperature

Section summary:

• Including temperature effects is critical for accurate bearing per-formance predictions.

• The modeling involves shaft, fluid film, and bearing pads.

• Energy equation is the governing equation.

One early idea to model the thermal effects is the approach ofeffective viscosity (Raimondi and Boyd, 1958). This methodemploys an empirical equation to calculate an effective tempera-ture. From the effective temperature, an effective viscosity isdetermined and used in the Reynolds equation. While this simpleidea recognizes the viscosity reduction due to temperature rise, itseffectiveness is very limited and it fails to give the maximum padtemperature, which is an important operation parameter.

To accurately model the thermal effects, the temperature distri-bution must be solved from the governing energy equation. Similar

to the momentum equations, the energy equation for bearinganalysis has been substantially simplified because of small filmthickness. The three-dimensional energy equation for laminar flowis usually written in the form of:

(13)

As shown in Equation (13), the steady-state temperature is deter-mined by three terms. The dissipation term describes the internalheat generation due to viscous shearing. As would be expected, theheating intensity is shown to be proportional to the lubricantviscosity. The heat convection term describes the rate of heattransfer due to the lubricant’s motion. And the conduction termdetermines the heat transfer between the lubricant and surroundingsurfaces. It can be shown by dimensional analysis that the heatconvection term is usually much larger than the conduction term.Thus, the film physically constitutes a heat source; while some ofthat heat is conducted away through the solid surfaces, the majorityof it is carried away by the flowing lubricant.

To achieve better computational efficiency, two simplified formsof Equation (13) are often used in practice. The first one is theadiabatic equation that is obtained by neglecting the conductionterm in Equation (13). The adiabatic energy equation was derivedby Cope in 1949 and has been widely used for a long time. Itimplies that no heat is transferred to the solids and the film tem-perature is constant radially. Figure 27 shows the typical adiabatictemperature solution for a smooth pad that has convergent-divergent clearance. Most of the temperature rise takes place in theconvergent clearance section where significant viscous shearingoccurs. In the divergent region, the temperature rise is significantlyreduced due to weak heat generation in the vapor-liquid mixture.Axially, the temperature is almost invariant, showing only slightincrease at the edges.

Figure 27. Adiabatic Temperature Solution for a Convergent-Divergent Film.

For many years, bearing designs had used adiabatic theory andisoviscous theory to bracket a bearing’s actual performance.However, this notion was later invalidated by a number of studies.It became clear in the 1960s that the radial temperature variationmust be taken into account for accurate bearing modeling(McCallion, et al., 1970; Seireg and Ezzat, 1973; Dowson andHudson, 1963). Moreover, in such a situation as a steam turbinewhere the hot shaft conducts heat into the film, the adiabaticassumption is clearly inappropriate. Therefore, another form of the


ρ ∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

μ ∂∂

∂∂

C uT

xv

T

yw

T

z

xk

T

x yk

T

y zk

T

z

u

y

w

y

p

Convection

Conduction Dissipation

+ +⎛⎝⎜

⎞⎠⎟ =

⎛⎝⎜

⎞⎠⎟

+⎛⎝⎜

⎞⎠⎟ + ⎛

⎝⎜⎞⎠⎟

+⎛⎝⎜

⎞⎠⎟ +

⎛⎝⎜

⎞⎠⎟

⎡

⎣⎢⎢

⎤

⎦⎥⎥

� ��

� ��

2 2

x (Circumferential)z (Axial)

Divergent

simplified energy equation, which includes the radial heat conduc-tion, has become more popular in modern bearing analysis. Sincethe temperature varies little in the axial direction, as shown inFigure 27, the axial heat transfer can be eliminated (∂T/∂z = 0) andthe original three-dimensional energy equation is reduced to two-dimensional. As shown in Figure 28, this two-dimensionalequation solves the temperature on the x-y plane, which is perpen-dicular to that of the adiabatic energy equation and the Reynoldsequation.

Figure 28. Numerical Meshes for the Governing Equations.

Figure 29 presents the temperature contour obtained from thistwo-dimensional energy equation. The lower rectangular section isthe bearing pad and the upper section is the fluid film. For bettervisualization, the film thickness is enlarged 103 times and the pad-film interface is highlighted by a bold line. The convergent-divergent clearance is clearly shown on the upper boundary. In thecircumferential direction, the film and pad temperatures increasefrom the leading edge, arrive at the maximum value around theminimum film thickness location, and decrease near the trailingedge as the result of heat conduction. The radial temperaturevariation is shown to be significant with a hot spot close to the padsurface. This trend is generally true for most bearings regardless ofthe specific contour values in this example.

Figure 29. Temperature Contour from the 2-D Energy Equationwith Conduction.

In some situations, neither radial nor axial temperature profilecan be assumed constant and the full three-dimensional energyequation must be solved. For example, in a pressure dam bearing,the temperature inside the pocket is much lower than that in theland regions (He, et al., 2004). Or if the bearing is misaligned withrespect to the shaft, the temperature is higher at one axial edgewhere the film thickness is minimum.

Most theoretical algorithms solve one form of the energyequation or another. Usually, the Reynolds equation is solved firstusing assumed lubricant viscosity. After the pressure distribution isobtained, the velocity components can be derived and the energyequation is solved to give the temperature distribution. Then, theviscosity is recalculated and the Reynolds equation is solved againusing the updated viscosity. This procedure continues till the dif-ference between two consecutive iterations is sufficiently small.

One may have noticed that the energy equation does not directlygovern the temperature in the solid pad. One way to obtain the padtemperature is to solve a separate heat conduction equation. Sincethe temperature and heat flux must be continuous at the film-padinterface, the heat conduction equation is coupled with the energyequation and they can be solved through iterations. The secondapproach is to extend the energy equation into the solid pad. Influid film, the energy equation has the convection, conduction, anddissipation terms. In solid pad, convection and dissipation termsare set to zero, leaving the heat conduction equation. In order tosatisfy the heat flux continuity, harmonic averaging is employed tomodify the heat conductivity on the film-pad interface (Paranjpeand Han, 1994).

The thermal effects on the predicted bearing performance aredemonstrated through the example of a two axial groove bearingreported in Fitzgerald and Neal (1992). All thermohydrodynamic(THD) results are calculated using the two-dimensional energyequation including heat conduction. Figure 30 compares the padsurface temperatures along the axial centerline. The theoreticalresults have close agreement with the experimental data. Figure 31shows the journal eccentricity ratio under various loads andspeeds. The THD analysis consistently gives more accurate resultscompared to the isoviscous hydrodynamic (HD) analysis, espe-cially in the case of 8000 rpm that has relatively high temperaturerise. The predicted vertical stiffness coefficients Kyy are plotted inFigure 32. The difference due to the inclusion of the thermal effectscan be as much as 30 percent at high speeds.

Figure 30. Pad Surface Temperature Comparison, L/D=0.5,N=8000 RPM, W=5.43 kN.

Deformations

Section summary:

• Elasticity modeling has become more and more important as aresult of increasing high speed, heavy load applications.

• It involves deformations of bearing pad, pivot, shaft, and shell.

• Theoretical models require caution in use.

Deformations change a bearing operating geometry, and, conse-quently, affect all aspects of a bearing performance. Because of itsflexible assembly, a tilting pad under high speed and/or heavy load


Fluid

Solid

Mesh for the Reynolds Equation & the Adiabatic Energy Equation

Mesh for the 2D Energy Equation with Conduction

60

6363

66

66

70

70

73

73

7676

76

79

79

82 85

85

89

89

92

9294

Circumferential Direction0 50 100 150

-10

0

10

20

30

40

50

60

70

80

90

100

Pad

Film

Angular Location (deg)

Te

mp

era

ture

(C)

0 90 180 270 36050

60

70

80

90

100

Load

Figure 31. Thermal Effects on the Predicted Eccentricity Ratio.

Figure 32. Thermal Effects on the Predicted Direct Stiffness.

is subject to deformations that are composed of two parts: mechan-ical deformation due to pressure and thermal deformation due totemperature rise. The simplest elastic model treats a tilting pad asa one-dimensional curved beam (Ettles, 1980; Lund and Pedersen,1987). If the deformed pad is assumed circular, the clearancevariation ΔC can be calculated and the bearing is modeled with amodified clearance c = co + Δc. Alternatively, the beam equationcan be numerically integrated and the nodal displacements are usedto correct the film thickness.

A more advanced approach is to formulate the problem based onthe principle of virtual work and solve it using the finite differ-ences or finite element method (Brugier and Pascal, 1989;Desbordes, et al., 1994). Although the actual pad is three-dimen-sional, two-dimensional plain strain approximation is often usedsince the deformations are primarily on the x-y plane. Figure 33(a)shows the shape of a tilting pad under mechanical deformation.The finite element grid before deformation is plotted with thedashed lines and the deformed pad is plotted with the solid lines.The mechanical deformation is shown to be mainly in the radialdirection. Since the displacements around the pivot are smallerthan those near the ends, mechanical deformation effectivelyincreases more cp than cb. Consequently, the pad preload, m, mayincrease or decrease depending on the relative cp and cb variations.Figure 33(b) shows the pad deformations under both mechanicaland thermal loads. Since the thermal deformation is dominant inthis example, the total deformation is shown as largely thermalgrowth with decreased cb and cp. The pad preload also varies

because the temperature rise is not uniform and the pivot con-strains the deformations near the pad center. In this particularexample, the mechanical load had little effect on the preload whilethe thermal deformation increases it.

Figure 33. Pad Deformations Obtained by 2-D Finite ElementMethod.

In addition to the pad deformations, the journal and bearing shellalso experience thermal growth in operation. Due to the rotation,the journal temperature is usually assumed constant and its defor-mation is modeled as free thermal growth at uniform temperature(Kim, et al., 1994). The bearing outer shell can also be modeled ina similar fashion. In addition, the pivot deformation under heavymechanical load can be calculated from the Hertian contact theory(Kirk and Reedy, 1988; Nicholas and Wygant, 1995).

Figure 34 presents the temperature predictions of a four-padtilting pad bearing experimentally investigated by Fillon, et al.(1992). The thermoelastohydrodynamic (TEHD) analysis takesinto account both the thermal and elastic effects. As shown in thisfigure, the inclusion of deformations brings the theoretical resultscloser to the experimental data. While this example shows that theinclusion of elasticity can improve the predictions, the deformationmodels, especially the journal and shell models, must be used verycarefully. In fact, it is often inadequate to model the journal andshell thermal growth as free expansions. Since the journal is part ofthe entire shaft, its growth is not “free” and its proper modelingrequires the knowledge of the entire shaft temperature distribution.Meanwhile, the bearing shell also cannot expand freely because itis constrained by the bearing housing. Its deformation is signifi-cantly affected by the housing conditions, including its stiffness,temperature, and shrink fit interference. A poor evaluation of thejournal and shell deformations can introduce very large errors inthe modeling predictions. Although the physics seem straightfor-ward, accurate modeling of elasticity is one of the most difficulttasks in a bearing analysis.

Turbulence

Section summary:

• Turbulent bearings have different behaviors compared tolaminar bearings.

• The turbulence effects must be included in modeling.

Two different types of flow may exist in fluid film bearings:laminar and turbulent. In laminar flow, the fluid particles aremoving in layers with one layer gliding smoothly over the adjacentlayers. In turbulent flow, the fluid particles have irregular motionand the flow properties, such as pressure and velocity, show erraticfluctuations with time and with position. Since it is impossible totrack the instantaneous flow properties, their statistical mean values


Load (KN)

Ecc

entr

icity

Rat

io

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Exp., 8000 rpmExp., 2000 rpmTHD, 8000 rpmTHD, 2000 rpmHD, 8000 rpmHD, 2000 rpm

Speed(RPM)

Kyy

(E+

8N

/m)

5000 100000

2

4

6

8

10

12

14

THDHD

W=9.43 kN

Original Deformed

Cb (in) 0.00310 0.00312

Cp (in) 0.00585 0.00589

m 0.47 0.47

Pivot

(a) Mechanical Deformation

Original Deformed

Cb (in) 0.00310 0.00274

Cp (in) 0.00585 0.00571

m 0.47 0.52

Pivot

(b) Mechanical and Thermal Deformations

Figure 34. Elastic Effects on the Pad Temperature Calculations.

are sought in a turbulent flow calculation. The flow regime isusually indicated by the Reynolds number defined as Re = ρUh/μ.The flow is laminar at low Reynolds numbers. As Re increases, theflow becomes unstable and partially turbulent, and eventuallyevolves into full turbulence at high Re. From the definition of Re,one can deduce that turbulence is likely to occur in large bearingsdue to high surface velocity and relatively large clearance. Aturbulent bearing exhibits increased power consumption alongwith a sharp change of bearing eccentricity (Wilcock, 1950).

Compared to laminar flow, turbulent flow has increased stressdue to the fluctuating motion. Since stress is proportional toviscosity, turbulent flow can be treated as laminar flow withincreased effective viscosity, which is defined as the superpositionof turbulent (eddy) viscosity and the actual viscosity of thelubricant. Thus, the Reynolds equation can be extended into theturbulent regime using effective viscosity values. Several modelshave been developed to evaluate the eddy viscosity. The models inConstantinescu (1959), Ng and Pan (1965), Elrod and Ng (1967),Safar and Szeri (1974) are similar in that they all utilize the “law ofwall” in which the eddy viscosity is assumed as a function of wallshear stress and distance away from the wall. On the wall surface,the flow is laminar and there is no eddy viscosity contribution. Theeddy viscosity increases as the position moves from the wall to thecore of the film. The specific formula used to quantify the eddyviscosity is different in those references. A distinct alternative is thebulk flow theory developed by Hirs (1973). Ignoring the detailedturbulence structure, his theory directly correlates the wall shearstress with the mean flow parameters using an empirical drag law.In addition, the fluctuating motion also enhances the heat transferacross the film. Analogous to the effective viscosity, an effectiveheat conductivity can be defined and employed to generalize theenergy equation into turbulent flow regime.

Figure 35 shows maximum temperature and power loss asfunctions of shaft rotational speed. The experimental data repre-sented by the discrete symbols are taken from Taniguchi, et al.(1990). As shown in this figure, the Tmax curve has a shift that cor-responds to the flow regime transition: when the flow is laminar,Tmax increases smoothly with the increasing shaft speed; Tmaxstays flat or even shows slight decrease during the flow regimetransition; Tmax resumes smooth increase after the transition iscompleted. Due to the increased effective heat conductivity, theanalysis including turbulent effects yields lower and more accurateTmax predictions. The inclusion of turbulence also significantlyimproves the friction loss prediction. Bouard, et al. (1996),compared three popular turbulent models: the Ng and Pan model,the Elrod and Ng model, and the Constantinescu model. Theyconcluded that, if a bearing is turbulent, the turbulent effects mustbe taken into account and these three models gave similar results.

Figure 35 Comparisons of the Results from Turbulent and LaminarTheories.

Dynamic Coefficients

Section summary:

• The reduced coefficients of a tilting pad bearing are dependenton the shaft’s precession or whirl frequency.

The bearing dynamic coefficients can be calculated by numeri-cally perturbing the journal position or by solving the perturbedReynolds equations. The first approach is straightforward. Afterestablishing the steady-state journal position, the hydrodynamicforce is calculated at a slightly different position. Since the force issomewhat different at this new journal position, the force variationdue to the small displacement is obtained and a stiffness coefficientis easily calculated from the definition of ΔF/Δx. A damping coef-ficient is similarly calculated with a velocity perturbation. Thesecond approach involves more mathematics because theperturbed Reynolds equations must be theoretically derived. Then,the dynamic coefficients are obtained by directly integrating thepressure solutions from those perturbed equations.

A fixed geometry bearing has eight dynamic coefficientsbecause such journal-bearing system has only two degrees offreedom (the journal translation in X and Y). However, thedynamic system of a tilting pad bearing has more degrees offreedom because the pads can rotate. These extra degrees offreedom lead to additional dynamic coefficients that are related tothe pads’ tilting motion. For example, a five-pad tilting pad bearinghas 58 dynamic coefficients. In practice, it is convenient to reducethese coefficients to eight equivalent ones that are related tojournal’s X and Y motions [Equation (7)]. This procedure is calleddynamic reduction or dynamic condensation. As shown in Figure36, the reduced coefficients are not constants, but dependent on thefrequency of the shaft whirl, which means the shaft perceivesdifferent bearing stiffness and damping at each vibrationfrequency. A widely debated topic, more discussions on thisfrequency dependency can be found in Lund (1964), Parsell, et al.(1983), and API Standard 684 (2005).

The Coupled Algorithm

Figure 37 shows the structure of a comprehensive thermoelasto-hydrodynamic algorithm that assembles the various modelsdiscussed above. The basic block is the classic hydrodynamicanalysis. Since the film thickness h is required in the Reynoldsequation, the journal operating position must be known in order tocalculate the hydrodynamic pressure. However, we only know that the journal is operating at equilibrium where the resulting



Te

mp

era

ture

(C)

0 90 180 270 36040

45

50

55

60

65

70

75

80

85

90Experiment

THD Analysis

TEHD Analysis

Load

Rotational Speed (rpm)0 1000 2000 3000 4000 5000

30

35

40

45

50

55

60

65

70

75

80

85

90

0

100

200

300

400

500

600

Tmax

HP Loss

Laminar Theory

Turbulent Theory

Laminar Theory

Figure 36. Frequency Dependent Stiffness and Damping, TiltingPad Bearing.

hydrodynamic force balances the external load. Therefore, theReynolds equation is initially solved with assumed journal positionand the actual position is searched through iterations. If a pad cantilt, its tilt angle also needs to be iteratively determined using thefact that, at equilibrium, the moment about the pivot must be zero.

Figure 37. Structure of a Sample TEHD Algorithm.

From the HD block, the algorithm can be expanded to a higherlevel that includes the thermal effects on the lubricant viscosity. Asmentioned earlier, the energy equation must be added to the for-mulation and iteratively solved with the HD block. In addition, thejournal and pad inlet temperatures also need to be calculated asimportant boundary conditions. During a revolution, a point on thejournal surface travels across the hot and cool sections of the fluid

film. Thus, it is reasonable to assume that the journal acquires theaverage film temperature and is constant. According to this model,heat flows into the journal in the hot sections, dumped back intothe fluid film in the cool sections, and the journal is adiabatic in abulk sense. The pad inlet temperature is determined in thepreceding oil groove (Heshmat and Pinkus, 1986). As shown inFigure 38, two streams of lubricant are mixed in the groove: coollubricant from the supply line and hot lubricant carried over fromthe previous pad. Therefore, at the pad inlet, the lubricant temper-ature is at some mixing value, which can be calculated by applyingenergy conservation to the groove control volume. Includingvarious deformation models, the THD block can be furtherextended to a complex thermoelastohydrodynamic analysis. Itshould be pointed out that the structure shown in Figure 37 is notunique. People have used a variety of structures to achieve thesame objective. However, regardless of the specific structure, acomputation always begins with a group of assumed initial values,and ends after convergence has been reached for every iterationloop.

Figure 38. Mixing in an Oil Groove.

Special Situations

In some special applications, the TEHD models presented aboveare not sufficient to predict a bearing’s properties. Those specialsituations require additional modeling efforts in order to achievesatisfactory theoretical predictions. However, these special situa-tions are often not modeled although they should.

Direct Lubrication

In recent years, exceedingly high pad temperature has becomean increasing problem in rotating machinery operations. Onesolution to this problem is the use of direct lubrication designs,such as the inlet pocket and spray bar. As suggested by the name,the idea is to directly supply cool oil into the pad clearance andblock hot oil carryover from the previous pad. According to Figure38, more Qsupply and less Qout will lead to lower mixing tempera-ture Tin, and consequently, lower temperature on the ensuing pad.Such direct lubrication designs have been successfully used andare gaining popularity in industry (Edney, et al., 1998).

Following this idea, Brockwell, et al. (1994), developed a newgroove mixing model assuming all cool oil in the inlet pocketenters the film. Since such model yields significantly reduced inlettemperature, lower pad temperature is predicted in their THDanalysis. The predicted peak temperatures also have goodagreement with their experimental data. Later, He, et al. (2002),noticed several interesting trends in the same group of test data.First, compared to a conventional pad, an inlet pocket pad does notalways have lower temperature near its leading edge; instead, itconsistently shows a smaller temperature gradient in the circum-ferential direction, which leads to the reduced peak temperature.This trend is clearly displayed in Figure 39. Second, on the curvesof maximum temperature versus shaft speed, some flat sections are observed, and before those flat sections, the pocketed and


Whirl Frequency (cpm)0 2000 4000 6000 8000 10000 12000

8.0E+04

1.0E+05

1.2E+05

1.4E+05

1.6E+05

1.8E+05

2.0E+05

100

120

140

160

180

200

220

5 Pad, Load Between PadCenter Offset, L/D = 0.750.3 Preload, S = 2.2

Kxx

KyyCyy

Cxx

Rotor Speed = 10,000 rpm

Pressure

Pad Tilt Angle

Journal Position

Initial Values

Film & Pad Temperatures

Journal Temperature

Inlet Temperature

Deformations

Output

HD

THD

TEHD

conventional pads often have similar peak temperatures. As shownin Figure 35, the flat sections are likely the indicator of flowregime transition. To simulate these trends, He, et al. (2002),proposed a different theory that attributes the cooling effects toturbulent flow that elevates heat transfer. According to their model,the inlet pocket destabilizes the flow and causes early turbulenceonset. Figure 39 shows the theoretical results that employed theirtriggered turbulence model. The predicted pad temperatures haveclose agreement with Brockwell’s experimental data. However,He, et al. (2002), did not identify the trigger that prompts turbu-lence on an inlet pocket pad.

Figure 39. Pad Temperature Comparisons Between Conventionaland Inlet Pocket Bearings, Experimental and TEHD results.

Since the direct lubrication designs are relatively new, theircooling mechanisms are still not clear and subject to debate. Morework needs to be done to understand their underlying physics andimprove their theoretical modeling.

Starvation

During operation, lubricant must be continuously supplied intoa bearing to replenish the side leakage. Many bearings areoperating in flooded condition in which the amount of supply oil ismore than what the bearing actually needs. With enough oil, con-tinuous fluid film is always established at the leading edge of a padthat has convergent clearance. The flooded lubrication condition isschematically shown in Figure 40(a). A bearing can also beworking in a starved condition in which the amount of oil is notenough to fill the pad leading edge clearance and the inlet region iscavitated. As shown in Figure 40(b), the continuous film is formeda certain distance away from the inlet where the clearance is suffi-ciently reduced. Figure 40 also shows conventional cavitation thatis caused by divergent clearance near the pad trailing edge.Examples of starved applications include ring-lubricated bearingsand direct lubrication designs that may be starved to minimizepower consumption (Heshmat and Pinkus, 1985; Brockwell, et al.,1994).

To model starvation, the main task is to determine the continu-ous film onset angle θf. If θf is known, the bearing can be analyzedusing the standard models and the effective arc length from θf tothe trailing edge. θf can be iteratively determined by comparing theavailable and required flowrates: at a certain location, if morelubricant is available to fill the clearance, the predicted θf shouldbe upstream where the larger clearance can accommodate the extrafluid; otherwise, the available lubricant can only fill a smallerspace and θf should be predicted further downstream (He, et al.,2003). Clearly, this search is coupled with the search of journalposition.

Figure 40. Flooded and Starved Lubrication Conditions.

For a multipad bearing, the level of starvation is different oneach pad because a loaded pad has smaller operating clearancecompared to an unloaded one. Therefore, starvation tends to occuron the unloaded pads first, and gradually spread onto the loadedpads. It also means that required pad flow is a function of eccen-tricity (or load) and speed, and different for each pad. Since padflow is usually controlled by inlet orifices preceding each pad, theflow to each pad can be tailored, but only for a single load/speedcondition. Also, note that the unloaded pads require more oil flowthan the loaded pads.

As shown in Figure 41, when the bearing has 100 percent supplyflow (flooded), hydrodynamic pressure is developed on all padsbecause the unloaded top pads have 0.6 offset pivots. The hydro-dynamic forces on those pads are labeled as F1 to F5, respectively.When the total supply flowrate to the entire bearing is cut by half,pad #3 and #4 are totally starved and pad #5 exhibits a 6.7 percentstarvation region, the two bottom pads are still flooded. When theflowrate is reduced to 40 percent, the starvation region on pad #5is expanded to 10 percent and pad #2 has a 6.7 percent starvationregion. If the flowrate is further reduced to 30 percent, pad #5becomes 100 percent starved and a 13.3 percent starvation areashows up on pad #2.

Figure 41. Development of Starvation in a Five-Pad Tilting PadBearing. (Courtesy He, et al., 2003)


X

Y Y

X

Continous Film Cavitated Film

(a) (b)

F1

F2

F3

F4

F5

Pad #1 Pad #2

Pad #3

Pad #4

Pad #5

X

Y

Shaft

W=1200 lb

F1

F2

F5

Pad #1 Pad #2

Pad #3

Pad #4

Pad #5

X

Y

Shaft

W=1200 lb

100% Flow 50% Flow

F1

F2

F5

Pad #1 Pad #2

Pad #3

Pad #4

Pad #5

X

Y

Shaft

W=1200 lb

F1

F2

Pad #1 Pad #2

Pad #3

Pad #4

Pad #5

X

Y

Shaft

W=1200 lb

40% Flow 30% Flow

A starved bearing exhibits increased temperature and decreasedfriction loss. As shown in Figures 42 and 43, the pad temperatureis increasing as the bearing becomes more and more starved.Meanwhile, since starvation leads to reduced continuous film area,the power loss due to viscous shearing is decreased. Besides highertemperature, starvation also reduces a bearing’s load capacity andstiffness. An example in He, at al. (2003), shows that a bearing’shorizontal stiffness Kxx can quickly diminish as the result of overlyreduced flowrate. In addition, starvation may result in dry frictionrubs (which may excite shaft vibration) and pad flutter (which maydamage the fluttering pads).

Figure 42. Pad Temperature Versus Flowrate. (Experimental Resultsfrom Brockwell, et al., 1994)

Figure 43. Power Savings Versus Flowrate. (Experimental Resultsfrom Brockwell, et al., 1994)

Additional Comments on Modeling

The modern TEHD theories generally give fairly accurate pre-dictions for a bearing’s performance parameters. A variety ofcomputer programs have been developed and successfully used inbearing design and analysis. Although significant progress hasbeen made since Osborne Reynolds, bearing modeling still faces anumber of challenges. To name a few:

• Temperature boundary conditions—In thermal analysis, thedifficult task is not to write down the equations, but to assign appro-priate boundary conditions. The most important one is the film inlet

temperature that is governed by groove mixing shown in Figure 38.A detailed modeling of the three-dimensional flow is impractical,and would involve turbulence, heat exchange with the solids, andpossible two-phase flow of liquid and air. Therefore, as mentionedearlier, a simple equation based on energy balance is used as apractical approximation. In this model, a hot oil carryover factor isrequired to address the fact that not all exit flow Qout enters the nextpad. The hot oil carryover factor, which is a function of the bearingdesign and operation condition, cannot be accurately obtained.Instead, it is usually estimated between 75 to 100 percent based onexperience. Therefore, significant error can be introduced as theresult of a poor estimate. Errors are also introduced on the back of apad where heat convection boundary condition is usually applied.Similar to the hot oil carryover factor, the convection coefficient isunknown and often specified at an engineers’ best estimate. In somesituations, such as misaligned shaft and bearing, the axial tempera-ture cannot be assumed constant. The full three-dimensional energyequation must be employed, which leads to the difficulty of deter-mining the boundary conditions at the axial ends.

• Deformation boundary conditions—As discussed above, toaccurately model the journal and outer shell deformations, theshaft and bearing housing need to be taken into account.However, the shaft and housing conditions are difficult to obtainand they are dependent on a machine’s specific design andoperation. Their modeling essentially goes beyond the scope of abearing analysis.

• Flow regime transition—To analyze a possibly turbulent bearing,the difficult question is when to apply the turbulence model. In mostanalyses, two critical Reynolds numbers are employed to determineflow regime transition. If the actual Re in the bearing is smaller thanthe lower critical Reynolds number, the flow is considered laminar;if Re is larger than the upper critical Reynolds number, the flow ismodeled as full turbulence; if Re is between those two thresholdnumbers, the flow is transitional and the eddy viscosity is scaled bya percentage factor (Suganami and Szeri, 1979). However, there isno reliable way to determine those critical Re’s. Although they areusually prescribed as constants, studies have indicated that they arefunctions of bearing geometry and operating condition (Xu and Zhu,1993). Again, large errors can be introduced if a modeling is basedon incorrect flow regime type.

• Complex geometries—These include the inlet pockets, spraybars, bypass cooling grooves, and hydrostatic lift pockets forstartup. Future research is needed to investigate these unconven-tional designs.

CONCLUSIONS

In this tutorial, major areas of journal bearings’ operation andmodeling are discussed. With respect to the operational aspects, wehave learned that:

• A bearing’s load carrying capacity comes from the hydrody-namic pressure developed in the fluid film.

• A convergent wedge, a moving surface, and a viscous lubricantare necessary to generate hydrodynamic force.

• Hydrodynamic forces have cross-coupled components that leadto large attitude angle and stability issues for fixed geometrybearings.

• Due to their pads’ ability to tilt, tilting pad bearings haveminimum cross-coupled forces and stiffnesses, which lead to theirsuperior performance.

• Viscous shearing generates heat in film, which leads to temper-ature rise and viscosity reduction.

• Hydrodynamic pressure and temperature rise causes elasticdeformations that change the film shape.

• Viscous shearing also results in mechanical power loss.



Te

mp

era

ture

(C)

0 72 144 216 288 36050

60

70

80

90

100

110

120

130

Experiment, 100% Flow



THD Results, 100% Flow



Load

% Flow20 40 60 80 100

0

5

10

15

20

25

30

35

40

45

50 Experiment, LOPTHD Results, LOPExperiment, LBPTHD Results, LBP

• There is a minimum required flowrate for oil supply. However,many bearings work in a flooded condition with more oil than theminimum.

• For relatively small vibrations like those normally encountered,a bearing’s dynamic properties can be represented by linear springsand dampers.

• The direct stiffness, direct damping, and cross-coupled stiffnesscoefficients all have significant rotordynamic implications.

• Dynamically, it is important to remember that a bearing is partof a global system involving rotor, seals, supports, etc.

• There are two predominant types of vibration instability: oilwhirl and shaft whip. Associated only with fixed geometrybearings, the former is largely determined by the bearing proper-ties and load. Applicable to either type of bearing, shaft whip isgoverned by the combined system.

• Both static and dynamic characteristics are speed and loaddependent.

To predict a bearing’s performance, theoretical models havebeen developed and successfully used in industry. The majormodels and mainstream techniques can be summarized as follows:

• A theoretical model that includes pressure, temperature and elas-ticity effects is often called thermoelastohydrodynamic algorithm.When properly used, a TEHD analysis can yield good prediction ofa bearing’s performance.

• Pressure calculations are the foundation of a TEHD algorithm.The Reynolds equation is usually employed in computer programs.

• For most analysis, thermal effects must be taken into account. Acomputer code usually solves some form of the energy equation.

• The elastic deformations should be included in the analysis ofhigh speed, heavily loaded bearings. However, the models need tobe used with caution.

• A turbulence model must be available to accurately predict theproperties of a turbulent bearing. Turbulent flow is usually associatedwith large bearing size, high shaft speed, and low lubricant viscosity.

• For a tilting pad bearing, the dynamic coefficients can be highlyfrequency dependent.

• Most TEHD algorithms cannot be applied to predict direct lubri-cated or starved bearings. These special cases require additionalenhancements.

• Current state-of-the-art bearing modeling still faces a variety ofdifficulties and challenges.

NOMENCLATURE

c = Bearing clearance cb = Assembled clearancecp = Pad machined clearanceco = Nominal bearing clearanceΔc = Clearance variation due to deformationCij = Damping coefficients, i, j = X or YCp = Lubricant specific heatD = Journal diametere = Journal eccentricityeX = Journal eccentricity projected on the horizontal (X) axiseY = Journal eccentricity projected on the vertical (Y) axisE = Journal eccentricity ratioEX = Journal eccentricity ratio projected on the horizontal (X) axisEY = Journal eccentricity ratio projected on the vertical (Y) axisFx = Film force in horizontal (X) directionFy = Film force in vertical (Y) directionh = Film thicknesshi = Film thickness at wedge inletho = Film thickness at wedge outlet

Kij = Stiffness coefficients, i, j = X or YL = Bearing axial lengthm = Pad preloadOb = Bearing centerOp = Pad arc centerOJ = Journal centerp = PressureQ = FlowrateR = Journal radiusRb = Bearing set bore radiusRp = Pad set bore radiusRJ = Journal radiusRe = Reynolds numberS = Sommerfeld numberT = TemperatureTmax = Maximum pad temperatureU = Journal surface velocityu = Fluid velocity in circumferential (x) directionv = Fluid velocity in radial (y) directionW = Applied loadWU = Unit load [WU = W/(L.D)]w = Fluid velocity in axial (z) directionX = Horizontal directionY = Vertical directionx = Circumferential direction along a pady = Radial direction across filmz = Axial direction along a padα = Pad offset factorβ = Pad arc measured from leading edge to the pivot locationΦ = Journal attitude angleκ = Heat conductivityμ = Lubricant viscosityθp = Pad arc lengthρ = Lubricant densityτ = Shear stressω = Shaft rotational speed

REFERENCES

API Standard 684, 2005, “Tutorial on Rotordynamics: LateralCritical, Unbalance Response, Stability, Train Torsional andRotor Balancing,” Second Edition, American PetroleumInstitute, Washington, D.C.

Barrett, L. E., Gunter, E. J., and Allaire, P. E., 1978, “OptimumBearing and Support Damping for Unbalance Response andStability of Rotating Machinery,” ASME Journal ofEngineering for Power, 100, (1), pp. 89-94.

Bouard, L., Fillon, M., and Frene, J., 1996, “Comparison BetweenThree Turbulent Models—Application to Thermohydro-dynamic Performances of Tilting-Pad Journal Bearings,”Tribology International, 29, pp. 11-18.

Boyd, J. and Raimondi, A. A., 1953, “An Analysis of the Pivoted-Pad Journal Bearing,” Mechanical Engineering, 75, pp. 380-386.

Brechting, R., 2002, “Static and Dynamic Testing of Tilting PadJournal Bearings as a Function of Load Angle and JournalSpeed,” Master Thesis, University of Virginia.

Brockwell, K., Dmochowski, W., and DeCamillo, S. M., 1994,“Analysis and Testing of the LEG Tilting Pad JournalBearing—A New Design for Increasing Load Capacity,Reducing Operating Temperatures and Conserving Energy,”Proceedings of the Twenty-Third Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 46-56.

Brugier, D. and Pascal, M., 1989, “Influence of ElasticDeformations of Turbo-Generator Tilting Pad Bearing on theStatic Behavior and on the Dynamic Coefficients in DifferentDesigns,” Journal of Tribology, 111, pp. 364-371.


Byrne, J. M. and Allaire, P. E., 1999, “Optimal Design of FixedPad Fluid Film Bearings for Load Capacity, Power Loss, andRigid Rotor Stability,” Report No.UVA/643092/MAE98/530,ROMAC Laboratories, University of Virginia.

Constantinescu, V., 1959, “On Turbulent Lubrication,” Pro-ceedings of the Institution of Mechanical Engineers, 173, pp.881-900.

Cope, W., 1949, “The Hydrodynamic Theory of Film Lubrication,”Proceedings of the Royal Society, London, United Kingdom,Series A, 197, pp. 201-217.

Desbordes, H., Fillon, M., Wai, C., and Frene, J., 1994, “DynamicAnalysis of Tilting-Pad Journal Bearing—Influence of PadDeformations,” Journal of Tribology, 116, pp. 621-628.

Dowson, D., 1962, “A Generalized Reynolds Equation for FluidFilm Lubrication,” International Journal of MechanicalScience, 4, pp. 159-170.

Dowson, D., 1998, History of Tribology, Second Edition, Suffolk,United Kingdom: Profession Engineering Publishing Ltd.

Dowson, D. and Hudson, J., 1963, “Thermo-HydrodynamicAnalysis of the Infinite Slider Bearing: Part II—The ParallelSurface Bearing,” Lubrication and Ware Convention,Institution of Mechanical Engineering, Paper 4 and 5.

Edney, S., Heitland, G. B., and Decamillo, S., 1998, “Testing,Analysis, and CFD Modeling of a Profiled Leading EdgeGroove Tilting Pad Journal Bearing,” Presented at theInternational Gas Turbine & Aeroengine Congress &Exhibition, Stockholm, Sweden.

Ehrich, F. F. and Childs, D. W., 1984, “Self-Excited Vibration inHigh-Performance Turbomachinery,” Mechanical Engineer-ing, May, pp. 66-79.

Elrod, H. and Ng, C., 1967, “A Theory for Turbulent Fluid Filmsand Its Application to Bearings,” Journal of LubricationTechnology, 89, pp. 346-362.

Elwell, R. C. and Booser, E. R., 1972, “Low-Speed Limit ofLubrication—Part One: What is a ‘Too Slow’ Bearing?”Machine Design, June 15, pp. 129-133.

Ettles, C., 1980, “The Analysis and Performance of Pivoted PadJournal Bearings Considering Thermal and Elastic Effects,”Journal of Lubrication Technology, 102, pp. 182-192.

Fillon, M., Bligoud, J., and Frene, J., 1992, “Experimental Studyof Tilting-Pad Journal Bearings—Comparison withTheoretical Thermoelastohydrodynamic Results,” Journal ofTribology, 114, pp. 579-587.

Fitzgerald, M. and Neal, P., 1992, “Temperature Distributions andHeat Transfer in Journal Bearings,” Journal of Tribology, 114,pp. 122-130.

Gardner, W. W., 1976, “Journal Bearing Operation at LowSommerfeld Numbers,” ASLE Transactions, 19, (3), pp. 187-194.

Hagg, A. C., 1946, “The Influence of Oil-Film Journal Bearings onthe Stability of Rotating Machinery,” ASME Transactions,Journal of Applied Mechanics, 68, pp. A211-A220.

Hamrock, B. J., 1994, Fundamentals of Fluid Film Lubrication,New York, New York: McGraw-Hill.

He, M., Allaire, P., Barrett, L., and Nicholas, J., 2002, “TEHDModeling of Leading Edge Groove Journal Bearings,”IFToMM, Proceedings of 6th International Conference onRotor Dynamics, 2, pp. 674-681.

He, M., Allaire, P., Barrett, L., and Nicholas, J., 2003, “THDModeling of Leading Edge Groove Bearings under StarvationCondition,” The 58th STLE Annual Meeting, New York, NewYork, 2003.

He, M., Allaire, P., Cloud, C. H., and Nicholas, J., 2004, “APressure Dam Bearing Analysis with Adiabatic ThermalEffects,” Tribology Transactions, 47, pp. 70-76.

Heshmat, H., 1991, “The Mechanism of Cavitation inHydrodynamic Lubrication,” Tribology Transactions, 34, pp.177-186.

Heshmat, H., and Pinkus, O., 1985, “Performance of StarvedJournal Bearings with Oil Ring Lubrication,” Journal ofTribology, 107, pp. 23-32.

Heshmat, H. and Pinkus, O., 1986, “Mixing Inlet Temperatures inHydrodynamic Bearings,” Journal of Tribology, 108, pp. 231-248.

Hirs, G., 1973, “A Bulk-Flow Theory for Turbulence in LubricantFilm,” Journal of Lubrication Technology, 95, pp. 137-146.

Hummel, C., 1926, “Kristische Drehzahlen als Folge derNachgiebigkeit des Schmiermittels im Lager,” VDI-Forschungsheft 287.

Jones, G. J. and Martin, F. A., 1979, “Geometry Effects in Tilting-Pad Journal Bearings,” STLE Tribology Transactions, 22, pp.227-244.

Kim, J., Palazzolo, A., and Gadangi, R., 1994, “TEHD Analysis forTilting-Pad Journal Bearings Using Upwind Finite ElementMethod,” Tribology Transactions, 37, pp. 771-783.

Kirk, R. and Reedy, S., 1988, “Evaluation of Pivot Stiffness forTypical Tilting-Pad Journal Bearing Design,” Journal ofVibration, Acoustics, Stress, and Reliability in Design, 110, pp.165-171.

Lanes, R. F., Flack, R. D., and Lewis, D. W., 1981, “Experimentson the Stability and Response of a Flexible Rotor in ThreeTypes of Journal Bearings,” ASLE Transactions, 25, (3), pp.289-298.

Lund, J., 1964, “Spring and Damping Coefficients for the TiltingPad Journal Bearing,” ASLE Transactions, 7, pp. 342-352.

Lund, J. W., 1987, “Review of the Concept of DynamicCoefficients for Fluid Film Journal Bearings,” ASME Journalof Tribology, 109, pp. 37-41.

Lund, J. and Pedersen, L., 1987, “The Influence of Pad Flexibilityon the Dynamic Coefficients of a Tilting-Pad Journal Bearing,”Journal of Tribology, 109, pp. 65-70.

Lund J. and Saibel, E., 1967, “Oil Whip Whirl Orbits of a Rotor inSleeve Bearings,” Journal of Engineering for Industry, 89, pp.813-823.

McCallion, H., Yousif, F., and Lloyd, T., 1970, “The Analysis ofThermal Effects in a Full Journal Bearing,” Journal ofLubrication Technology, 92, pp. 578-587.

Muszynska, A. and Bently, D. E., 1995, “Fluid-Induced Instabilityof Rotors: Whirl and Whip—Summary of Results,” Noise andVibration Conference, Pretoria, South Africa.

Newkirk, B. L., 1924, “Shaft Whipping,” General Electric Review,27, (3), pp. 169-178.

Newkirk, B. L. and Taylor, H. D., 1925, “Shaft Whipping Due toOil Action in Journal Bearings,” General Electric Review, 28,(8), pp. 559-568.

Ng, C. and Pan, C., 1965, “A Linearized Turbulent LubricationTheory,” Journal of Basic Engineering Series D, 87, pp. 675-688.

Nicholas, J. C. and Wygant, K. D., 1995, “Tilting Pad JournalBearing Pivot Design for High Load Applications,”Proceedings of the Twenty-Fourth Turbomachinery Sympo-sium, Turbomachinery Laboratory, Texas A&M University,College Station, Texas, pp. 33-48.


Nicholas, J. C., 1994, “Tilting-Pad Bearing Design,” Proceedingsof the Twenty-Third Turbomachinery Symposium, Turboma-chinery Laboratory, Texas A&M University, College Station,Texas, pp. 179-194.

Nicholas, J. C., Gunter, E. J., and Barrett, L. E., 1978, “TheInfluence of Tilting Pad Bearing Characteristics on theStability of High-Speed Rotor Bearing Systems,” ReportNo.UVA/643092/MAE81/141, ROMAC Laboratories,University of Virginia.

Paranjpe, R. S. and Han, T., 1994, “A Study of theThermohydrodynamic Performance of Steadily LoadedJournal Bearings,” Tribology Transactions, 37, pp. 679-690.

Parsell, J. K., Allaire, P. E., and Barrett, L. E., 1983, “FrequencyEffects in Tilting-Pad Journal Bearing Dynamic Coefficients,”ASLE Transactions, 26, pp. 222-227.

Petrov, N., 1883, “Friction in Machines and the Effect of theLubricant,” Inzhenernii Zhurnal, St. Petersburg, Russia, 1, pp.71-104, 2, pp. 227-279, 3, pp. 377-436, 4, pp. 435-464 (InRussian).

Pinkus, O., 1987, “The Reynolds Centennial: A Brief History ofthe Theory of Hydrodynamic Lubrication,” ASME Journal ofTribology, 109, pp. 2-20.

Raimondi, A. and Boyd, J., 1958, “A Solution of the Finite JournalBearing and Its Application to Analysis and Design—Part III,”Transactions of ASLE, 1, pp. 159-209.

Reynolds, O., 1886, “On the Theory of Lubrication and ItsApplications to Mr. Beauchamp Tower’s ExperimentsIncluding an Experimental Determination of the Viscosity ofOlive Oil,” Philosophical Transactions, 177, pp. 157-234.

Safar, Z. and Szeri, A. Z., 1974, “ThermohydrodynamicLubrication in Laminar and Turbulent Regimes,” Journal ofLubrication Technology, 96, pp. 48-56.

Salamone, D. J., 1984, “Journal Bearing Design Types and TheirApplications to Turbomachinery,” Proceedings of theThirteenth Turbomachinery Symposium, TurbomachineryLaboratory, Texas A&M University, College Station, Texas,pp. 179-192.

Seireg, A. and Ezzat, H., 1973, “ThermohydrodynamicPhenomena in Fluid Film Lubrication,” Journal of LubricationTechnology, pp. 187-198.

Simmons, J. and Lawrence, C., 1996, “Performance Experimentswith 200 mm, Offset Pivot Journal Pad Bearing,” TribologyTransactions, 39, pp. 969-973.

Stodola, A., 1925, “Kristische Wellenstorung infolge derNachgiebigkeit des Oelpolsters im Lager,” SchweizerischeBauzeiting, 85, pp. 265-266.

Suganami, T. and Szeri, A. Z., 1979, “A ThermohydrodynamicAnalysis of Journal Bearings,” Journal of LubricationTechnology, 101, pp. 21-27.

Szeri, A., 1979, “Tribology—Friction, Lubrication, and Wear,”New York, New York: McGraw-Hill Book Company.

Taniguchi, S., Makino, T., Takeshita, K., and Ichimura, T., 1990,“A Thermohydrodynamic Analysis of Large Tilting-PadJournal Bearing in Laminar and Turbulent Flow Regimes withMixing,” Journal of Tribology, 112, pp. 542-550.

Tower, B., 1883, “First Report on Friction Experiments,”Proceedings Institute of Mechanical Engineering, pp. 632-666.

Tower, B., 1885, “Second Report on Friction Experiments,”Proceedings Institute of Mechanical Engineering, pp. 58-70.

Wilcock, D., 1950, “Turbulence in High Speed Journal Bearings,”Transactions of the ASME, 72, pp. 825-834.

Xu, H. and Zhu, J., 1993, “Research of Fluid Flow and FlowTransition Criteria from Laminar to Turbulent in a JournalBearing,” Journal of Xi’An Jiaotong University, 27, pp. 7-14.

Zuck, C. and Flack, R., 1986, “Experiments on the Stability of anOverhung Rotor in Pressure-Dam and Multilobe Bearings,”ASLE Transactions, 30, pp. 225-232.

BIBLIOGRAPHY

Allaire, P. E., 1997, Course Notes: Lubrication Theory and Design,Department of Mechanical and Aerospace Engineering,University of Virginia, Charlottesville, Virginia.

Hagg, A. C. and Sankey, G. O., 1956, “Some Dynamic Propertiesof Oil-Film Journal Bearings with Reference to the UnbalanceVibration of Rotors,” ASME Transactions, Journal of AppliedMechanics, 23, pp. 302-306.

Lund, J. W., 1974, “Stability and Damped Critical Speeds of aFlexible Rotor in Fluid-Film Bearings,” ASME Journal ofEngineering for Industry, Transactions ASME, 96, (2), pp.509-517.

Wills, J. G., 1980, Lubrication Fundamentals, Mobil Oil Corpora-tion, New York, New York: Marcel Dekker, Inc.


Charles R. (Charlie) Rutan is SeniorEngineering Advisor, Specialty Engineer-ing, with Lyondell Chemical Company, inAlvin, Texas. His expertise is in the field ofrotating equipment, hot tapping/plugging,and special problem resolution. He hasthree patents and has consulted on turbo-machinery, hot tapping, and pluggingproblems all over the world in chemical,petrochemical, power generation, andpolymer facilities.

Mr. Rutan received his B.S. degree (Mechanical Engineering,1973) from Texas Tech University. He is a member of the AdvisoryCommittee of the Turbomachinery Symposium, and has publishedand/or presented many articles.

ABSTRACT

There are several methods of analysis to define the reliability ofthe critical rotating equipment at various facilities within acompany. Part I of this tutorial is intended to present several ofthese methods. The following Part II presents the calculations.

INTRODUCTION

In the mid 1980s, the author’s supervisor gave him an “opportu-nity” to better define the critical machinery reliability asturnarounds were extended. The task was to develop an equationthat would produce a number to be used as a guide such that theplant management would have a feeling as to the risk for postpon-ing the next olefin unit turnaround or minimizing the amount ofinspection performed on the major rotating equipment, e.g., minoror major overhaul(s). Based on the critical machinery on thehistory, design, vibration, and process conditions he developed aweighted number that he tried to use as an indicator to justify theoverhaul requirements. At the time, this was not accepted well bythe plant management because other facilities in the company hadlonger/shorter shutdown intervals as well as published reports ofother companies in the same commodity chemical industry and the“Solomon” report that was published every two years did not agreewith his conclusions derived from this number. During this periodof time there were two methods:

• Kepner-Tregoe’s® Problem Solving and Decision Making

• Managerial Analytics, a Monsanto Chemical Company eventanalysis system

In later years several other methods of potential problem and/orroot cause analysis—hazard and operability analysis (HAZOP),Federal Emergency Management Agency (FEMA), failure modeeffect and criticality analysis (FMECA), event-tree analysis (ETA),Delphi, method organization for a systematic analysis of risks

(MOSAR), management oversight risk tree (MORT), WeibullAnalysis, and Six Sigma—have been developed to aid in quantify-ing the risk of extending the operational time of the criticalturbomachinery.

KEPNER-TREGOE®

Kepner-Tregoe® Problem Solving and Decision Making(PSDM) is a step-by-step process that helps people resolvebusiness. Used in organizations worldwide, PSDM helps individu-als, groups, and/or teams efficiently organize and analyze vastamounts of information and take the appropriate action.

This process provides a framework for problem solving anddecision making that can be integrated into standard operating pro-cedures. It is used to enhance other operational improvement toolssuch as Six Sigma, Lean Manufacturing, and others.

PSDM comprises four distinct processes:

• Situation Appraisal is used to separate, clarify, and prioritizeconcerns. When confusion is mounting, the correct approach isunclear, or priorities overwhelm plans, Situation Appraisal can beused.

• Problem Analysis is used to find the cause of a positive ornegative deviation. When people, machinery, systems, or processesare not performing as expected, Problem Analysis points to therelevant information and leads the way to the root cause.

• Decision Analysis is used for making a choice; it is intended toclarify the purpose and balances risks and benefits to arrive at asupported choice.

• Potential Problem/Opportunity Analysis is used to protect andleverage actions or plans. Potential Problem Analysis shoulddefine the driving factors and identifies ways to lower risk. Whenone action is taken, new opportunities, good or bad, may arise.These opportunities must be recognized and acted on to maximizethe benefits and minimize the risks.

MANAGERIAL ANALYTICS

Managerial Analytics (MA) is an analytical identificationprocess developed by the Monsanto Company. Managing changeand using change to manage requires the use of some combinationof the five basic analytical processes.

• Event Analysis is a systematic process to identify events andevents of change that impact the reliability of the turbomachinery.Events and events of change from the past and present have animpact on the present and future reliability. Events of change couldbe the stopping of wash oil injection or the rising of the sodiumconcentration in the steam or not repairing the spare rotor.

Step 1. First person responsibilityStep 2. Recognize events and their relationshipsStep 3. Establish priorityStep 4. Separate and sequence components

193

APPLIED RISK AND RELIABILITY FOR TURBOMACHINERY

PART I—RELIABILITY PROCESSES

byCharles R. Rutan

Senior Engineering Advisor, Specialty Engineering

Lyondell Chemical Company

Alvin, Texas

Step 5. Identify intended nominal first person resultsStep 6. Identify resolution resources, threats, and opportunitiesStep 7. Identify intended nominal first person actionStep 8. Select the analytical processesStep 9. Statement for the first analysis

• Deviation Analysis is a process to help determine the unknowncause of an observed effect deviating from a standard effect inorder to decide on action. This is used for both positive andnegative deviations and in other words events of change that havehappened.

Step 1. Statement of the deviation effectStep 2. Deviation effect specificationsStep 3. Unique characteristics of the “involved” dimensionsStep 4. Change eventsStep 5. Possible causesStep 6. Test possible causesStep 7. Set priority on possible causesStep 8. Verification of cause

• Action Analysis is used to select a single course of action fromseveral courses of action. This is based on the projected perform-ance of the action to attain a set of desired effects and assessing theimpacts of and to that action if it were taken.

Step 1. Statement of the intended nominal actionStep 2. Set the desired effectsStep 3. Classify the desired effectsStep 4. Weight the want effectsStep 5. Generate action alternativesStep 6. Filter the action alternatives through “must” effectsStep 7. Score action alternatives to the “wanted” effectsStep 8. Impact evaluation

Step a. Identify an action alternativeStep b. Identify impactsStep c. Identify potential deviationsStep d. Assess impacts on the environmentStep e. Summarize potential deviations and their likely causesStep f. Plan actions to manage likely causesStep g. Plan actions to manage potential deviations

Step 9. Make best balanced selection

• Action Planning helps to decide “What will be done?”, “Whowill do it?”, and “When it will be done” to reach one or moredesired future effects that are not expected to occur unlesssomething is done. It involves a set of interrelated and interde-pendent actions.

Step 1. Define the action requiredStep 2. Define person(s) who have the prime responsibility tocomplete the required actionStep 3. Define support resourcesStep 4. Define the date and time to initiate the actionStep 5. Define the projected time of completion of the action

• Potential Deviation Analysis is a process used to examine aplanned action or a future event of change for significant impactsand deviations, and to plan additional actions to manage theseresults.

Section 1. Statement of the intended nominal actionSection 2. Identification of potential deviations

Step a. Identify a planned actionStep b. Identify impactsStep c. Identify potential deviationsStep d. Assess impacts to the environmentStep e. Summarize potential deviations and their likely causes

Section 3. Action planning for potential deviationsStep a. Plan actions to manage likely causeStep b. Plan action to manage potential deviationStep c. Revise the action plan

ROOT CAUSE ANALYSIS (RCA)

Define the Problem

Cause and Effect

There are five elements of a cause and effect chart.

1. Primary effect:a. A singular effect of consequence that we wish to eliminate or

mitigateb. The “what” in the problem definitionc. It is always the most present cause in the analysis and fre-

quently the most significant.d. It is the point at which we begin to ask “why.”

2. Actions and condition causes:a. They are both causes, actions are momentary and conditions

exist over time.b. Actions can become conditions and conditions can become

actions.c. Actions and conditions interact to create effect.

3. Casual connection caused by:a. Forces that cause going from present to past.b. Elicits a more specific responsec. Minimizes storytellingd. If a cause “connects” it adds to the visual dialogue and

therefore has value.

4. Evidence is the data that supports a conclusion and is presentedas:

a. Sensed—It is processed through our senses of sight, sound,smell, taste, and touch. Sensed is the highest quality evidence.

b. Inferred—The ability to infer is derived from our understand-ing of known and repeated casual relationships. Inference is thenext highest quality of evidence.

c. Evidence is important because:i. It supports the reality of any single cause.ii. Solutions should only be applied to evidence-based causes.iii. It minimizes the influence of politics and power plays.

5. “Stop” or a “?”

Problem Definition

There are four elements of problem definition.

1. What—What is the problem?2. When—When did it happen?3. Where—Where did it happen?4. Significance—Why are we working on this?

Identify Effective Solutions

The root cause is not what we seek, it is effective solutions.

1. Challenge each cause.2. Offer possible solutions for each cause.

Implement the Best Solutions

1. Prevents recurrencesa. Prevents or mitigates this problemb. Prevents similar problemsc. Does not create additional problems or unacceptable situa-

tions

2. Within your controla. Your control may be you, your department, your company,

your suppliers, or your customers.b. Nature is not within your control.c. The facilitator is rarely the problem owner.

3. Meets your goals and objectivesa. The goals of the overall organizationb. The goals of your department or group


c. Your individual goals and objectivesd. Must provide reasonable value

RELIABILITY ASSESSMENTOF ROTATING EQUIPMENT

Reliability assessment of rotating equipment (RARE) is one toolof an overall maintenance and turnaround strategy of criticalrotating equipment.

Origin of Initial Program and Concept

The Hartford Steam Boiler Inspection & Insurance Company(HSB) expected the companies they insured to follow the steamturbine overhaul intervals recommended by the original equipmentmanufacturer (OEM). Due to world market pressure and techno-logical advancements companies began extending the intervalsbetween major unit turnarounds. These companies depended onthe expertise of their rotating equipment engineers to help definehow long they could reliably operate between turnarounds. Theirgoal was to lengthen the interval between major overhaul outagesto coincide with the need for other inspections and testingmandated by government agencies or process needs. During the70s and 80s the chemical, petrochemical, and refining industrieswould perform a major overhaul of their critical machinery everyfour to six years as a standard practice. As operating companiesextended their turnaround intervals HSB became concerned aboutthe increased risk their customers were taking. At this point intime, data showing the effects or risks of stretching run timesbetween overhauls were not available. HSB decided to develop aprogram that would quantify these risks for their customers. M&MEngineering, a part of the HSB company at that time, was given theassignment to develop a tool to assess the reliability of steamturbines. The tool is called Steam Turbine Reliability AssessmentProgram (STRAP). HSB did not plan to use this program tostructure premiums. The vision was to require a STRAP analysiswhenever a company chose to run longer than the OEM recom-mended practice. If STRAP showed that they were at low risk thenHSB would authorize or accept the company’s plans, but if STRAPshowed the company to have a high risk they were told that theyneeded to implement some risk reduction procedures or improve-ments to mitigate some of these risks.

Consortium of Companies

With this vision, M&M Engineering pulled together a group ofrecognized industry experts in the turbomachinery field. Theexperts brought with them their individual company’s reliabilityassessment techniques, industry standards (API, ASME, etc.), rec-ognized industry best practices, new technologies, and mostimportant their personal experience.

Starting with a blank sheet of paper the group spent countlesshours defining, developing, categorizing, and weighing questionsand responses. The caveat for this group was when completed theywould take this program back to their respective companies anduse it as another tool to justify turbine improvements that wouldimprove reliability and extend run lengths, which then resulted ina significant savings of money.

The turbine data accumulated, which consisted of originaldesign specifications, history, uprates, upgrades, failures, and sitespecific information, were then entered into the programdeveloped by M&M Engineering. The algorithms in the programtook the weighted turbine data and generated a risk index number(RIN). The RIN is the number of days that the turbine will bedown, during the run extension. It is based on statistical informa-tion that is calculated from the data contained in the database forthe same general type of turbine. The program gives 25 specificitems to be addressed and calculates a return on the investment(ROI) using site pricing data. Armed with this information therotating equipment engineer and managers can make informeddecisions to accept the projected risk or to take corrective actions

to minimize or mitigate the risks that were defined. Like home andcar insurance, risk can never be eliminated but most risks can beminimized, then it becomes a decision on what level of risk theplant/company is willing to accept. With this information long-term shutdown strategy can be developed that is based on theoperation, maintenance, and reliability practices of the unit.

STEAM TURBINE RISK ASSESSMENT PROGRAM

1. General Information Data Sheeta. Plant specifics?b. Size or class of the turbine (five categories)?c. Age of the turbine?d. Manufacturer of the turbine?e. What is the turbine driving?f. When the turbine was last dismantled?g. Etc.?

2. Turbine Performance (Design and Actual)a. Horsepower?b. Speed?c. Inlet flow?d. Temperature?e. Pressure?f. Etc.?

3. Site and Utility Dataa. Location?b. Steam generation?c. Etc.?

4. Construction Featuresa. Type of turbine?b. Critical speed?c. Control system?d. Etc.?

5. Sparesa. Complete turbines?b. Bearing sets?c. Complete rotor sets?d. Stationary diaphragms?e. Case?f. Nozzles blocks?g. Where are they stored?h. Couplings?i. Labyrinth seal sets?j. Control valve parts/governor valve assemblies?k. How many days are required to prepare the rotor for installa-

tion?l. Is there a periodic inspection for signs of corrosion on the

spare parts?

6. Maintenance and Repairsa. How many hours to the repair shop?b. How often do you drain water from the oil reservoir?c. Do you drain it at startup?d. Turbine overhauls?e. Are there documented detailed overhaul procedures for

this/similar turbines?f. How is the lube oil system cleaned during overhauls?g. Qualifications?h. Foundation inspection?i. Alignment?j. Inspections?k. Oil systems?

7. Turbine Operationa. Procedures?

i. Do the plant’s written steam turbine operating proceduresinclude:

(1) Prestartup checklist?(2) Overspeed tests?

APPLIED RISK AND RELIABILITY FOR TURBOMACHINERYPART I—RELIABILITY PROCESSES

195

(3) Putting turbine on slow roll?(4) Normal operations?

ii. Do you have a procedure for lining up sealing steam?iii. Is there a formal management of change?

b. Tests?i. When is the governor or valve rack exercised online?ii. When do you test the trip and throttle valve?iii. When do you test the nonreturn valve?

c. Qualifications?i. Do operators have the authority to shut down the turbine?

8. Monitoring and Protectiona. Which of the following parameters are monitored and

trended?i. Turbine steam inlet temperature?ii. Bearing metal temperatures?iii. Thrust?iv. Vibration?v. Lube oil pressure?

b. Do the following parameters trigger an operating–specificalarm?

i. Turbine steam inlet temperatures?ii. Bearing metal temperatures?iii. Thrust?iv. Vibration—radial?

c. Do the following parameters trigger an alarm followed by atrip?

i. Vibration—radial?ii. Thrust?

9. Upgrades—Have the following components been upgraded?a. Rotor assembly?b. Bearings?c. Casing?d. Coupling?e. Seals?f. Trip and throttle valve?g. Miscellaneous?

10. Steam Systema. What supplies steam to the turbine?b. What type of make up water does it use?c. What type of condensate polishing does this unit use?d. Is steam purity monitored?e. Etc.?

11. Past Failures/Problemsa. Turbine internal components?b. Bearings?c. Casing?d. Coupling?e. Seals?f. Trip and throttle valve and its components?g. Governor?h. Fouling?

12. Consequence Dataa. Plant production in dollars per day (minimum/maximum)?b. What is the cost if the turbine goes down?c. What other costs are there if the turbine shuts down?d. How does the plant typically handle a rub during the startup

of this turbine?e. How would the plant typically handle fouling of this turbine

during normal operation?

The total number of possible questions is about 3000, but for alarge steam turbine 350 to 400 questions are normally answered.

Question Weighting

The team of experts then weighted each of the 3000+ questionsand their respective answers based on their experience and

knowledge. As part of the development of failure probabilities, theteam needed to set a baseline interval for overhaul outages.

They decided to use a six year dismantle overhaul schedulefrequency for Class 1 to 4 turbines and a five year overhaulschedule frequency for Class 5 turbines. Thus, the baseline prob-abilities developed were based on the turbine operating for sixyears or five years without opening the case. Since risk is calcu-lated by multiplying probability times the consequence, thecalculated risk would be for a six year/five year interval. Thus,the risk calculated would be in days of lost production over a fiveor six year interval. Because some STRAP users do not divulgefinancial data about production revenue, converting the days inlost production to dollars cannot be performed for all turbines. Itwas decided to create the term “risk index number,” which wouldallow all turbines to be compared to each other. The RIN is therisk of failure in days of lost production over a six year intervalfor Class 1 to 4 turbines and over a five year interval for Class 5turbines.

On the basis of the input data and the likelihood-consequenceinformation, a risk for operation of the turbine may be calculatedas a function of time between the dismantle inspections. In eachcase, a quantified list of recommendations to mitigate the risk willalso be reported based on the greatest contribution to the risk.Inspection outage plans then may be tailored to optimize the timebetween overhauls on the basis of acceptable level of risk. The riskindex number is a number generated by the program based on:

• The questions and their corresponding answers.• Industry standards (ASME, API, etc.).• Accepted industry practices.• Latest technology.• Relative to other turbines in the company and/or the number ofturbines of the same design in the database.

Probability

The probability of failure of a component is the risk of failure indays of lost production (RIN) divided by the consequence of thatcomponent. Since risk is the product of probability time’s conse-quence there are questions that will significantly affect the RIN ifanswered with a poor option or not answered at all. An example ofthis is that the question of “testing of overspeed trip” and ananswer of “never tested” would have a significant impact on theRIN and the recommendations by increasing the probability offailure.

Program Aids

The program has several aids for the user in an effort to makethe output meet the user needs while providing the best output.Some of the aids are:

• Check for missing answers.• Check for inconsistent answers.• Etc.

Results and Comparisons

STRAP will compare the turbines with:

• Other turbines in the company.• All the turbines in the database.• Turbines in the same class.• Turbines by the same manufacturer.• Turbines in the same industry.• Etc.

Recommendations

STRAP makes recommendations to improve the reliability ofthe turbine(s) based on return of investment. It is then up to theengineers to decide which recommendation can be executed, inwhat order, and during an outage and/or online.


Impact

Based upon the consequence data of the operating unit that hasbeen input to the program, the ROI can be calculated. Some of thebasic questions are:

• Unit production rate?• Effect on the plant if a unit goes down per day?• Impact of a trip of the turbine?• Etc.?

RELIABILITY ASSESSMENT OF COMPRESSORS

The reliability assessment of compressors (RAC) program hasbeen developed in a very similar manner to the STRAP program.Compressors are divided into the types of service to which they arebeing applied. In addition the program requires all the constituents ona percent molecular weight basis. In reality the RAC program is farmore complicated than the STRAP program. There are a number of:

A. General Information1. What group this compressor belongs to?

a. Charge gas/cracked gas?b. Air?c. Ammonia?d. Chlorine?e. Oxygen?f. Refrigeration (clean gas)?g. Other?

2. What industry environment this compressor operates in?a. Chemical/petrochemical?b. Gas?c. Refining?d. Other?

3. Compressor General Details?a. Compressor manufacturer (there are a multitude of manu-

facturers)?b. Model number?c. Serial number?d. Date manufactured?e. Compressor duty?f. Driver?g. Number of years in service?

B. Construction1. Type of end seals?2. Type of interstage seals?3. Internal coatings?4. Has the rotor been high-speed balanced?5. Do you have a lube and oil system designed API 614?6. Type of bearings?7. Materials of construction?8. Type of coupling(s)?9. What is the number of impellers?

C. Past Failures/Problems1. Has the compressor ever had past failures or problems?2. Has the compressor ever operated in reverse?

3. How often does the compressor have vibration problems?4. Has the lube oil system been contaminated?5. Has the seal oil system been contaminated?6. Has the buffer gas consumption increased?7. Have you experienced compressor trips due to instrumenta-

tion problems?8. Has the compressor ever been oversped?9. Have you ever had problems with kinking in the past?

D. Design Versus Actual1. Inlet

a. Pressure?b. Temperature?c. Molecular weight?

2. Dischargea. Pressure?b. Temperature?c. Molecular weight?

3. Brake horsepower required?4. Speed?5. Estimated surge ICFM?

E. Process Gas Data1. Is the process dry or wet?2. Is the process gas corrosive?3. Does the process foul?4. Do you monitor gas composition online?5. What molecular weight was the compressor designed for?6. What is the current molecular weight of the process?7. Process stream

a. Air (MW 28.966)b. Carbon monoxide (MW 28.010)c. Ethylene (MW 28.052)d. Propane (MW 44.094)

F. Site DataG. Control SystemsH. Lube Oil SystemsI. SparesJ. Maintenance and RepairsK. OperationL. Monitoring and ProtectionM. Rerates and UpgradesN. Seal Fluid SystemO. Environment and Business Consequence Data

CONCLUSION

The Reliability Assessment of Rotating Equipment (RARE) isthe combination of both the STRAP and RAC programs.Depending on the maintenance strategy of the company and/orthe facility, either/or STRAP and RARE programs can be a sig-nificant benefit in assessing the critical machinery performanceand identifying the areas that could or should be addressed toimprove the reliability and define the obstacles to extending therun time between minor and major overhauls with an acceptablerisk.

APPLIED RISK AND RELIABILITY FOR TURBOMACHINERYPART I—RELIABILITY PROCESSES

197

Shiraz A. Pradhan is a ConsultingEngineer, with ExxonMobil ChemicalCompany, in Baytown, Texas. His recentexperiences include the design, commis-sioning, and startups of oxoalcohol,halobutyle, polyethylene, polypropylene,and fluids’ projects in the Far East, SouthAmerica, and the U.S. He previouslyworked with Esso/Imperial Oil in Canadaas Machinery Engineer and with BritishGas Corporation in the United Kingdom as

Senior Project Leader. He has been involved in reliability auditsand service factor improvement projects in ammonia, fertilizer,PVC, olefins and polyolfins plants, and oil and gas pipelineprojects internationally.

Mr. Pradhan has a degree (Mechanical Engineering) from theUniversity of Nairobi, and an M.S. degree from Lehigh Universityin Pennsylvania. He holds the title of European Engineer (FEENI,Paris), is a fellow of the Institution of Mechanical Engineers, U.K.,and is a registered Professional Engineer in the Province ofOntario, Canada.

INTRODUCTION

Reliability is critical for all industries. For the petrochemicalindustry it assumes added significance because much equipment isunspared or has minimal redundancy. Table 1 shows a comparisonbetween the commercial airlines, nuclear, and petrochemicalindustries.

Table 1. System Characteristics for Different Industries.

COMPARISON BETWEENELECTRONICS AND MECHANICAL SYSTEMS

In electronic component reliability assessment the concept ofconstant failure rate is used (Ireson, et al., 1996). This is not thecase for mechanical and machinery components. There are manyreasons for this. Machinery components follow:

• Have increasing failure rate pattern• Are not standardized like electrical components

• Have more failure modes than electronic components

Fundamental to reliability assessment of mechanical compo-nents is the need for failure distribution and supporting data thatdescribe the behavior of the components in the real world. This ismore easily said than done.

Cumulative Distribution Function

For reliability prediction one would like to know the probabilityof a failure occurring before a time t. This can be derived by theequation:

F (t), Probability of a failure before time t =

(1)

As t approaches infinity, F(t) approaches 1.

Reliability Function

Reliability function is complementary to the cumulative distri-bution function and gives the probability of survival of acomponent or system to specified time t.

(2)

Failure Rate or Hazard Function

This function allows the determination of the failure probabilityof a system or component in a small increment of time Δt, havingsurvived to time t.

(3)

Failure Distributions for Mechanical Systems

Exponential Distribution

This distribution is used extensively in the electronic industryand for some mechanical system’s reliability assessment as well:

(4)

where λ = failure rate per unit time

(5)

where MTBF = mean time between failure.The reliability function for exponential distribution is:

(6)


APPLIED RISK AND RELIABILITY FOR TURBOMACHINERY

PART II—RELIABILITY CALCULATIONS

byShiraz A. PradhanConsulting Engineer

ExxonMobil Chemical Company

Baytown, Texas

System CommercialAirlines

Nuclear Petrochemical

Mission Length, Hr <50 <5000* 8760> T > 70000

Access DuringMission

NIL NIL NIL

Access BetweenMission

Full Limited FULL if a planned IRD**

* Depending on mandated maintenance** Inspection, Repair Downtime

( )f t dt

t

−∞∫

( ) ( ) ( )R t F t f t dt

t

= − =−∞

∫1

( ) ( )( )h t

f t

R t=

( ) ( )f t e for tt= >−λ λ 0

and MTBFλ = 1 /

( ) ( )R t e t= − λ

For equipment that follows exponential distribution, the probabil-ity of having exactly k failures by time T is given by the Micro-soft® Excel function:

(7)

Some characteristics of the exponential distribution are:

• Applies to situations where failure events are random and notdue to wear, age, or deterioration• Is a memoryless distribution. This means that the probability offailure is the same in all intervals of time.• Has constant hazard rate• This distribution is often applied to systems that are repairable.

Normal Distribution

This distribution is applied to situations where the failures aredue to wear. However mathematics of failure rate or hazard rate arecomplex.

Log Normal Distribution

This distribution has wide applicability to mechanical systemswhere failures are due to crack propagation, corrosion, and stress-temperature phenomenon.

Weibull Distribution

Weibull distribution (Dodson, 1994) is one of the most versatileof the failure distributions and has wide applicability for mechani-cal systems. It is defined by two parameters: η called thecharacteristic life or scale factor and a constant β called the shapeparameter. The Weibull probability density function is:

(8)

The Weibull reliability function is:

(9)

The characteristic life, η, is the age at which 63.2 percent of thepopulation will have failed.

The shape parameter β has several cases of interest in mechani-cal reliability assessment.

• When β <1—This indicates a decreasing hazard rate. Inmechanical systems this is the initial run-in phase where faultycomponents with defects fail. With time these early failuresdiminish. This phase is often called the infant mortality or burn-inphase.

• When β = 1—This is a special case of Weibull distribution whenit becomes an exponential distribution. As previously noted, for theexponential distribution the hazard rate is constant and failures arerandom. This phase designates the useful life of the component.The failure rate is reciprocal of the MTBF.

• When β = 2—the hazard rate is increasing linearly with time.This case is known as Rayleigh distribution.

• When β = 2.5—the hazard rate is increasing and the distributionapproximates the log normal distribution.

• When β = 3.5—For this case the hazard rate is increasing andthe distribution approximates a normal distribution.

Figure 1 shows these various cases.

Bathtub Curve

A bathtub curve is a plot of hazard rate against time. Figure 2shows the curves for electronic and mechanical systems.

Figure 1. The Weibull Probability Density Function.

Figure 2. Bathtub Curve for Mechanical and ElectronicComponents.

APPLICATIONS OF FAILURE DISTRIBUTIONS

Example 1

In this example there are two identical pumps in parallel asshown in Figure 3. They have negative exponential failure distri-bution and therefore:

(10)

Figure 3. Two Pumps in Parallel.

Survival of one pump is sufficient to assure the success of thesystem. For this case the reliability of the system is given by:

(11)

(12)

APPLIED RISK AND RELIABILITY FOR TURBOMACHINERYPART II—RELIABILITY CALCULATIONS

199

( )POISSON k t False, ,λ

( ) ( )f t t e t= − −βηβ

ββ η1 /

( ) ( )R t e t= − /η β

λ λ λA B= =

R R R R Rt A B A B= + − ×

= −− −2 2e et tλ λ

A

B

• Assume pump MTBF = 36 months (3 years)Failure rate, λ = 1/ MTBF = 1/3 = 0.333 failures/yearMission time = 1 year

• For a single pump, the reliability is:R = e�0.333�1

= 71 percent and failure probability is 29 percent

• For a parallel pump, system reliability is:R = 2e�0.333�1 � e�2�0.333�1

= 1.4335 � 0.5137= 92 percent and failure probability is 8 percent

We will now evaluate the situation when one pump fails and thesister pump is operating without backup (Figure 4). Assume thatpump mean time to repair (MTTR) = six days. Evaluating theprobability of success for six days when the spare pump isoperating without backup:

• Repair time = 6 days = 6/365 = 0.0164 yearsR = e�0.333�0.0164

= 99.45 percent and failure probability is 0.55 percent

Installing a spare pump reduces the probability of system failurefrom 29 percent to less than 1 percent.

Figure 4. Only One Pump in Service.

Example 2

A screw compressor has a failure rate of 0.0666 failure/year.What is the probability of two failures in exactly five years? Thisproblem is solved by Equation (7).

• POISSON(k,λT,False)ΛT = 0.0666 � 5 = 0.333k = 2

The POISSON expression answer for the above values is 3.9percent.

Risk Based Maintenance

With a spared system one is often faced with deciding if therepair of failed equipment should be carried out in an expedited oremergency basis. A typical situation is the repair strategy for, say,boiler feedwater pumps when the main pump has failed. Should itbe repaired on an emergency basis? Plant operators have no confi-dence in the operating pump. Figure 5 can be used to aid in thisdecision. It relates the MTBF of the spare pump with MTTR ordays unavailable and reliability.

How long can the spared pump be out for repair so as not tocompromise the target reliability of 99 percent. The following factswill aid in the decision:

• The spare pump is running satisfactorily.• Its MTBF is 15 months.

In Figure 5 the x-axis is entered at 15 months and intersects thereliability line of 99 percent at a horizontal line that corresponds toan allowable outage of nine days. In this case there is no need todo emergency repair.

Example 3

In this example we will use a commercially available Weibullprogram to plot the Weibull for a set of reactor pump seals. There

Figure 5. Reliability Versus Mean Time Between Failure andRepair Time. (Courtesy Bloch and Geitner, 1990)

are two pumps in service and seven seal failures. Inputting thetimes to failure in the program yields the Weibull plot shown inFigure 6.

• Each Weibull distribution is applicable to a single failure modeof the equipment.• Weibull plots as a straight line• Statistically more data points give greater accuracy to the plot.• The r2 in the plot gives an indication of the good fit. r2 = 1 isbest.• In engineering sometimes we are forced to work with fewer datapoints. This increases the uncertainty of the plot.

Figure 6. Weibull Plot for Reactor Pump Seals.

The Beta (β) or the shape factor = 3.95 and the characteristic life(η) = 28.3 months. This suggests that for these pump seals hazardfunction is increasing and the seals have a wear-out mode.

Table 2 is an example of a Microsoft® Excel function reliabilitycalculator that can be programmed. It takes Beta (β) and the char-acteristic life (η) from the Weibull plot and calculates theprobability of failure and its converse, the reliability of the pumpseals, based on the Weibull reliability function [Equation (9)] for arange of desired operating months.


A

Input Data: Reactor Seal Failures

Failure # Time to Failure 1 17 2 18 3 23 4 24 5 31 6 33 7 34

Table 2. Reliability Calculator for Reactor Pump Seals, Based onWeibull.

In Table 3, the Weibull equation is solved for critical values.From this table it is possible to get the expected time to failure forany desired reliability.

Table 3. Inverse Weibull: Time to Failure for Desired Reliability.

Reliability Simulations

In multicomponent systems the mathematics gets too complexfor closed form solutions. In such cases simulation is the next bestapproach. With the availability of computers, simulations havebecome relatively easy. The objective of the simulations is topredict the central tendency of a given variable. In this case it is topredict the probability of failure of a complex system. Thealgorithm for Monte Carlo simulations for mechanical systems isdeveloped from the basic Weibull reliability function [Equation(8)], and noting that the failure function is:

(13)

Taking logarithm of both sides, it yields the equation for the timeto failure, t.

(14)

Inputting a random number for the F(t) in Equation (14) yieldsan estimate of time to failure, t, for a given set of the Beta and Eta.With modern computers it is possible to execute several thousandsimulations.

Example 4

Reliability/Risk and Maintenance Planning

A machinery engineer is faced with a decision to recommend tomanagement if a turbine, which has operated successfully for fouryears, should be overhauled in the fifth or the sixth year ofoperation? In this example only three components of the turbinefor which Weibull data are available from similar machines areconsidered. A commercially available reliability program was usedfor the simulation.

Figure 7 shows the reliability block diagram (RBD) of theturbine and Table 4 shows the input data for the program.

Figure 7. Reliability Block Diagram of the Turbine.

Table 4. Inputs for Turbine Components.

The methodology for making a risk-based decision is as follows:

• Step 1—Assemble failure distribution for each component. Thedata for the turbine components are from the maintenance/failurehistories from the operating plant, original equipment manufac-turer’s (OEM) data, and public domain databases (Table 4).

• Step 2—Make program simulations:• Number of program simulations: 1000• Mission times: five years and six years. The simulation

program is run from time t = 0 to t = 6.

• Step 3—From the event log feature of the reliability program,assemble histograms of the number of failures for each componentof the turbine for each successive year of mission time until thesixth year. The histograms essentially confirm the trend in thewear-out modes of the turbine components. Figure 8 shows asample output from the event log and Figure 9 shows a histogramfor the turbine components.

Figure 8. Sample Output from Event Log for Run #15.

• Step 4—Calculate the reliability of each component for amission time of five and six years having survived four years.

Use the Weibull reliability function [Equation (9)]:


201

Beta 3.948 Months Failure Probability

Reliability

Eta (Months) 28.3 28 0.62 0.38

27 0.57 0.43

26 0.51 0.49

25 0.46 0.54

24 0.41 0.59

23 0.36 0.64

22 0.31 0.69

21 0.27 0.73

20 0.22 0.78

19 0.19 0.81

18 0.15 0.85

17 0.13 0.87

16 0.10 0.90

15 0.08 0.92

14 0.06 0.94

13 0.05 0.95

12 0.03 0.97

11 0.02 0.98

10 0.02 0.98

9 0.01 0.99

8 0.01 0.99

Beta = 3.98 Reliability Time to Failure (Months) Eta = 28.28 Months 0.01 41.637

0.1 34.932

0.5 25.773

0.75 20.627

0.8 19.341

0.9 15.993

0.99 8.820

( ) ( )F t R t= −1

( )Time to Failure InF t

=−

⎛

⎝⎜

⎞

⎠⎟

⎡

⎣⎢⎢

⎤

⎦⎥⎥

ηβ

1

1

1

Mon Apr 25 14:06:48 2005

Block Name Failure Distro Param1 Param2 Param3 _________________________________ _(Beta)_____(Eta)___________ BladeErosion Weibull 4.20000 8.50000 0.00000 Controller-HydroCU Weibull 1.50000 11.0000 0.00000 TTV Weibull 4.70000 8.50000 0.00000 Valve-Control Weibull 1.50000 10.0000 0.00000

Starting Run 15 Time= Years 2.619166 Valve-Control Failed , TimeOperated=2.619166 System=Red 2.634584 Valve-Control Repaired, RepairTime=0.015419 System=Green 4.239981 Controller-HydroCU Failed , TimeOperated=4.224562 System=Red 4.243281 Controller-HydroCU Repaired, RepairTime=0.003300 System=Green 4.535766 TTV Failed , TimeOperated=4.517048 System=Red 4.548038 TTV Repaired, RepairTime=0.012272 System=Green 5.423212 BladeErosion Failed , TimeOperated=5.392222 System=Red 5.510067 BladeErosion Repaired, RepairTime=0.086855 System=Green

6.000000 Simulation Terminated End of Run #15

Figure 9. Histograms of Component Failures with Time.

(15)

• Sample calculations for trip and throttle valve (TTV): Beta = 3.8Eta = 9 years

• Reliability at time = 4 yearsR(4 yr) = e�(4/9)3.8 = 95.5 percent


Thus, R (5yr/having survived 4 yr) = R(5 yr)/R(4 yr) = 89.8/95.5 =94.0 percentFailure probability = 1 � 94.0 = 6 percent


Thus, R (6 yr/ having survived 4 yr) = R(6 yr)/R(4 yr) = 80.7/95.5= 84.5 percentFailure probability = 1 � 84.5 = 15.5 percent

Table 5 and Figure 10 show the relative reliabilities and failureprobabilities for the turbine components.

Table 5. Relative Reliabilities and Probabilities of Failure forTurbine Components.

Figure 10. Relative Probability of Failure for Turbine Components(Example 3).

The analysis shows that there is an 11 percent additional risk inextending the operation from the fifth to the sixth year. The totalrisk is 19 percent for the control valve.

Example 5

A gas turbine generator (GTG) system is arranged in parallel asshown in Figure 11. Survival of one GTG line is sufficient forsystem success. All components have negative exponential distri-bution. The desire is to:

• Forecast the system reliability and availability for a mission timeof two years.• Assess system vulnerability when one GTG is out for plannedmaintenance for 10 days.

Figure 11. Two GTGs in Parallel.

The reliability program model comprises all the components ofthe GTGs including the gearboxs, turbine and compressor sections,the generator, and the auxiliaries such as cooling water, lubesystem, control system, and fire suppression system.

The results (Table 6) indicate that for two GTGs in parallel andfor a mission time of two years, the system reliability will be 99percent with a mean system failure rate of 0.01 failure in two years.The mean availability is >99 percent.

Table 6. Results of Simulation: GTGs in Parallel, Mission Time =Two Years.

When one GTG is down for planned maintenance of 10 days, theresults (Table 7) show that the system is vulnerable to failurewithin the 10 days. For the duration of the10 days the system reli-ability is only 97 percent and there is a potential of a mean failureof 0.03.

REFERENCES

Ireson, W. G., Coombs, C. F., Jr., and Moss, R. Y., 1996, Handbookof Reliability Engineering and Management, Second Edition,New York, New York: McGraw Hill.

Dodson, B., 1994, Weibull Analysis, Milwaukee, Wisconsin: ASQPress.

Bloch, H. and Geitner, F., 1990, Machinery Reliability Assessment,New York, New York: Van Nostrand Reinhold.


Failure Trend of TTVAccumulated Over 20 Simulations

02468

101214

1 2 3 4 5 6 7 8 9 10

Years

# F

ailu

res

Failure Trend of Blade ErosionAccumulated Over 20 Simulations

0

1

2

3

4

5

6

1 2 3 4 5 6 7 8 9

Years

# F

ailu

res

Failure Trend of Hydraulic Control UnitAccumulated Over 20 Simulations

01234567

1 2 3 4 5 6 7 8 9 10

Years

# F

ailu

res

Failure Trend of Valve ControlAccumulated Over 20 Simulations

0

1

2

3

4

5

1 2 3 4 5 6 7 8 9

Years

# F

ailu

res

( ) ( )R t e t n= − / β

Component Reliability (5 yr /having survived 4 yr)

Failure Prob. (5 yr /having

survived 4 yr)

Reliability (6 yr /having

survived 4 yr)

Failure Prob. (5 yr /having survived 4 yr)

Control Valve 90.43 9.57 80.9 19.1 Controller Hydraulic CU 91.65 8.35 83.2 16.8 Trip and Throttle Valve 94.05 5.95 84.5 15.5 Blade Erosion 98.7 1.3 96.4 3.6

Risk Assessment Probability of Failure

(Having Survived 4 Yrs.)

05

10152025

5 Yr. Run

6 Yr. Run

Mon Apr 25 15:20:35 2005

Results from 100 run(s):

Parameter Minimum Mean Maximum Standard Deviation Total Costs 66.00 67.62 72.66 1.37 Ao 0.991366829 0.999913668 1.0000000000 0.000858989 MTBDE 1.982734 >1.999827 >2.000000 n/a MDT (1 runs) 0.017266 0.017266 0.017266 n/a MTBM 0.283248 >1.439166 >2.000000 n/a MRT (79 runs) 0.003004 0.013159 0.030500 0.006948 %Green Time 95.527907 98.950059 100.000000 1.002106 %Yellow Time 0.000000 1.041308 4.065847 0.975965 % Red Time 0.000000 0.008633 0.863317 0.085899 System Failures 0 0.010000 1 0.099499

R(t=2.000000) =0.990000

Table 7. Results of Simulation: One GTG Out for Overhaul,Mission Time = 10 Days.


203

Mon Apr 25 15:27:36 2005

Results from 100 run(s):

Parameter Minimum Mean Maximum Standard Deviation Total Costs 11.30 11.33 12.50 0.17 Ao 0.729199029 0.993634398 1.0000000000 0.037137561 MTBDE 0.019973 >0.027216 >0.027390 n/a MDT (3 runs) 0.004199 0.005812 0.007417 0.001314 MTBM 0.019973 >0.027216 >0.027390 n/a MRT (3 runs) 0.000000 0.003339 0.005819 0.002452 %Green Time 72.919903 99.363440 100.000000 3.713756 %Yellow Time 0.000000 0.000000 0.000000 0.000000 % Red Time 0.000000 0.636560 27.080097 3.713756 System Failures 0 0.030000 1 0.170587

R(t=0.027390) =0.970000

John C. Nicholas is owner and Presidentof Rotating Machinery Technology,Incorporated, in Wellsville, New York, acompany that repairs and services turbo-machinery, and manufactures bearings andseals. He has worked in the turbomachineryindustry for 31 years in the rotor andbearing dynamics areas, including fiveyears at Ingersoll-Rand and five years asSupervisor of the Rotordynamics Group at

the Steam Turbine Division of Dresser-Rand.Dr. Nicholas received a B.S. degree (Mechanical Engineering,

1968) from the University of Pittsburgh and a Ph.D. degree in rotorand bearing dynamics (1977) from the University of Virginia. Heholds several patents including one for a spray-bar blocker designfor tilting pad journal bearings and another concerning by-passcooling technology for tilting pad journal and thrust bearings. Dr.Nicholas, a member of ASME, STLE, and the Vibration Institute,has authored over 40 technical papers concerning rotordynamicsand tilting pad journal bearing design and application.

Greg Elliott is a Senior Project Engineerin the Power Transmission Division ofLufkin Industries in Lufkin, Texas. He worksprimarily in development, analysis, anddesign of high speed gear drives. He alsoprovides support when finite elementanalysis, fatigue analysis, vibration analysis,or other assistance is needed in machinerydesign or problem solving. Previous activitieshave included development of Lufkin’s

current “N-D” high speed gear product line.Mr. Elliott received a B.S. degree and an M.S. degree

(Agricultural Engineering, 1982, 1990) from Texas A&M University.

Thomas P. Shoup is the Chief Engineerand Vice President of Operationsat Rotating Machinery Technology,Incorporated, in Wellsville, New York. Hehas worked in the turbomachinery industryfor 20 years in rotor and bearing systemdynamics, including two years at the SteamTurbine Division of Dresser-Rand, fiveyears at Jacobs/Sverdrup Technology, Inc.,and 12 years at Siemens Demag Delaval

Turbomachinery, Inc. Mr. Shoup is a member of ASME.

ABSTRACT

Improved turbomachinery aerodynamic performance requirementshave increased journal bearing operating speeds and loads wellabove traditionally acceptable values. For example, for highperformance gearboxes, pinion bearing surface speed requirementsare often over 325 f/s with bearing unit loadings in the 500 psirange. In order to meet the design challenges for these severeapplications, evacuated bearing housings have been utilized as aneffective means of reducing journal bearing operating temperatures.Unfortunately, the use of evacuated housing designs has

introduced a new and troubling phenomena—journal bearingstarvation. This was never a problem with flooded designs withpressurized housings since any additional oil that may be requiredis simply drawn from the captured oil inside of the bearinghousing. With the new evacuated housing designs, all required oilmust be supplied by the oil inlet orifices. Often times, the amountof supply oil required to keep all pads from starving is well beyondreasonable. Thus, due to practicality, starvation in some form isallowed in almost all evacuated designs.This paper discusses evacuated journal bearing starvation and its

possible detrimental effects on rotordynamics. Specifically, the

1

TILTING PAD JOURNAL BEARING STARVATION EFFECTS

byJohn C. NicholasOwner and President


Wellsville, NewYork

Greg ElliottSenior Project Engineer

Lufkin Industries

Lufkin Texas

Thomas P. ShoupChief Engineer and Vice President of Operations


Wellsville, NewYork

andEd Martin

Project Engineer

Lufkin Industries

Lufkin Texas

effect of starvation on journal bearing stiffness and damping isinvestigated. A case history is presented showing the effect ofincreasing oil flow on the location and amplification of a gearboxpinion critical speed during near zero load mechanical testing. Asflow increased and the bearing became less starved, the location ofthe critical increased while the amplification decreased indicatinga strong dependency of bearing stiffness and damping on oil flow.Concurrently, a similar but smaller bearing was tested under zeroload starvation conditions. Essentially no effect on stiffness anddamping was evident. From these results, the authors conclude thatalthough increasing the oil flow solved the problem, starvation initself was not the cause.

INTRODUCTION

Improved turbomachinery aerodynamic performance requirementshave increased journal bearing operating speeds and loadswell above traditionally acceptable values. For example, for highperformance gearboxes, pinion bearing surface speed requirementsare often over 325 f/s with bearing unit loadings in the 500 psirange. In recent years, many gearbox applications have been above350 f/s. Within the lead author’s experience, the fastest journalbearing surface velocity for an American Petroleum Institute (API)gearbox is 389 f/s. Again, within the lead author’s experience,faster surface velocities have been successfully achieved for highspeed balancing applications with speeds up to 575 f/s.Achieving these extremely high surface velocities would not be

possible with a 1970s vintage tilting pad journal bearing (TPJB).Early journal bearing designs were almost exclusively flooded.That is, the exit area for the oil was less than the oil inlet area. Thiscreated a positive pressure inside the bearing housing, therebyflooding the bearing with oil (Nicholas, 1994).In order to meet the design challenges for these severe

applications with excessive surface velocities, evacuated bearinghousings have been utilized as an effective means of reducingjournal bearing operating temperatures. Tanaka (1991) presentedexperimental operating temperature data for a tilting pad journalbearing with bearing end seals (flooded and pressurized) andwithout bearing end seals (evacuated, nonpressurized). The bearingoperated at lower temperatures without the end seals. Since then,many designs have been developed adopting the evacuated housingconcept including Gardner (1994), Brockwell, et al. (1994), Balland Byrne (1998), and Nicholas (2003).Unfortunately, the use of evacuated housing designs has

introduced a new and troubling phenomena—tilting pad journalbearing starvation. This was never a problem with flooded designswith pressurized housings since any additional oil that may berequired is simply drawn from the captured oil inside of the bearinghousing. With the new evacuated housing designs, all requiredoil must be supplied by the oil inlet orifices. Depending on theefficiency of the inlet oil supply mechanism, some oil escapes thebearing directly without participating in lubricating the pads. Thiscertainly exacerbates the problem.Another issue with evacuated housing tilting pad journal

bearings is the unloaded pads. For heavy loads, the loaded pads,with a much smaller journal-to-pad leading edge entrance area,require much less oil compared to the unloaded pads that have amuch larger entrance area. In most cases, the amount of supply oilrequired to keep all pads, including the unloaded pads, fromstarving is well beyond reasonable.Finally, for high performance gearboxes, journal bearings are

sized for peak performance at full load. The bearing is “oversized”for operation at near zero load during mechanical acceptancetesting. Oil flow requirements for a full film on the loaded pads atfull load results in starvation for all pads at near zero load. Again,the amount of supply oil required to keep all pads from starvingat near zero load when the film thickness is larger is beyondreasonable. Thus, due to practicality, starvation in some form isallowed in almost all evacuated designs.

This paper discusses evacuated tilting pad journal bearingstarvation and its possible detrimental effects on rotordynamics.Specifically, the effect of starvation on journal bearing stiffnessand damping is investigated. A case history is presented showingthe effect of increasing oil flow on the location and amplificationof a gearbox pinion critical speed during no-load mechanicaltesting. Tilting pad journal bearing stiffness and damping testresults will be presented for a zero load case with varying degreesof starvation (Harris and Childs, 2008). From these test results andthe gearbox case history, the authors conclude that althoughincreasing the oil flow solved the gear box problem, starvation initself was not the cause.

OIL FLOW REQUIREMENTS

This author’s tilting pad journal bearing severe applicationexperience plot is shown in Figure 1. The blue dots inside of the redbox are API gearbox applications. Outside of the red box are teststand and high speed balance applications. Almost all applicationsshown on the plot are evacuated housing designs.

Figure 1. Tilting Pad Journal Bearing Severe Application ExperiencePlot.

The steps used to determine the oil flow requirements for all ofthese applications are summarized below:

1. Assume that the hot oil carryover from pad-to-pad is equal to theamount of oil that passes through the pads minimum film thickness(refer to ACKNOWLEDGEMENT section).

2. Using this hot oil carryover amount, determine the minimumlubricating flow requirement for a full film on the loaded pad atfull load and at full speed (i.e., no loaded pad starvation) and thenmultiply by the number of pads. This is the minimum per bearinglubricating oil flow requirement.

3. Increase the flow as necessary from the calculated minimum tomeet the bearing operating temperature requirements.

This results in a full film on the loaded pads at full load. It alsoresults in starvation for the unloaded pads at any load condition andstarvation for all pads during the no-load mechanical test. Thisdesign methodology worked successfully for all of the indicatedFigure 1 applications except the one shown with the red triangle.Notice that it is safely within a batch of other successful applicationsas opposed to standing alone near the edge of the red box. Indeed,it is not the fastest application nor the most heavily loaded.

THE PROBLEM GEARBOX

The problem gearbox indicated by the red triangle in Figure 1 isa 24 MW, double helical, speed increaser driving three centrifugalcompressors in offshore gas reinjection service. A photo of the boxduring mechanical testing is shown in Figure 2. The maximum

PROCEEDINGS OF THE THIRTY-SEVENTH TURBOMACHINERY SYMPOSIUM • 20082

continuous pinion speed is 12,700 rpm. The 6.0 inch diametertilting pad pinion bearing’s surface velocity is 333 f/s with a fullload bearing unit load of 489 psi. The bearing’s geometric propertiesare summarized in Table 1. The actual bearing is shown in Figure3. Note that this design does not use end seals. Also of note is thehuge discharge opening between pads to enable the oil to easilyexit the bearing. This bearing design is described in detail inNicholas (2003).

Figure 2. Problem Gearbox During Mechanical Testing.

Figure 3. Tilting Pad Journal Pinion Bearing—Evacuated HousingDesign.

Table 1. Pinion and Test Tilting Pad Bearing Geometric Properties.

Using the design steps outlined in the previous section, theminimum lubricating oil flow requirement for a full film on theloaded pads at full load and full speed is 26 gpm. The oil flow wasincreased to 34 gpm of lubricating oil plus 5 gpm of by-passcooling oil (Nicholas, 2003) to properly cool the bearing for a totalof 39 gpm. However, due to pressure from the customer to reduceoil flow, the bearings were shipped to the gear manufacturer with atotal oil flow rate of 34 gpm (includes lubricating oil plus by-passcooling oil).For reference, at zero load (i.e., gravity load only), the calculated

full film lubricating oil flow requirement is 44 gpm. Adding in the5 gpm of by-pass cooling flow, the total flow requirement for a fullfilm at zero load is 49 gpm. Thus, it was anticipated that thebearing would be partially starved during the no-load mechanicaltest. Table 2 summarizes these results.

Table 2. Pinion Bearing Oil Flow.

Running a pinion bearing partially starved during mechanicaltesting should not be a problem from the standpoint of loadcapacity. Since the bearing is sized for full load, it is obviouslyoversized at no-load. Partial starvation would, in effect, reduce thesize of the bearing. At the leading edge of a partially starvedbearing pad, there is not enough oil to fill the pad-to-journal gap.Thus, air is drawn into the pad and the initial section of pad islubricated with an air-oil mixture. As this mixture moves fartherinto the pad, the film thickness decreases to the point where thereis enough oil to fill the gap and a full film results. Thus, part of thepad’s leading edge is ineffective in providing load capacity duringpartial starvation. Since the bearing is operating in a zero loadcondition, this reduction in load capacity will not be a problem.However, the starved part of the pad’s leading edge will also notparticipate fully in providing the bearing’s stiffness and dampingproperties. Thus, some degradation in the bearing’s stiffness anddamping is expected for partial starvation.Although the problem described in this paper would not

mechanically harm the pinion or bull gear, it was significantlogistically and commercially as it did cause this gearbox to fail theAPI mechanical acceptance test. Everyone involved may believethat this problem is only an artifact of the test conditions and thatit would not manifest itself under real operating conditions.Nevertheless, to meet the API test specifications (API 613, 2003),and ship the gearbox, it was necessary to make the changesdescribed in this paper.It is important to note that this gearbox ran flawlessly during the

loaded string test. There was no indication of the problemdescribed herein.

MECHANICAL TESTING

A speed-amplitude plot for the pinion from the initial no-loadmechanical test run is shown in Figure 4 with a pinion bearing totaloil flow of 34 gpm per bearing. For this plot and all other test resultspresented herein, an unbalance weight was placed on the couplinghub in order to excite and locate the pinion’s first critical speed. FromFigure 4, note that the pinion critical speed is evident at about N1 =13,500 rpm with an amplification factor of A1 = 13.5. This does notmeet the API 613 (2003) acceptance criteria. Note that the terms

3TILTING PAD JOURNAL BEARING STARVATION EFFECTS

“no-load” and “zero load” are used herein to describe the mechanicaltest load, which, in actuality, was close to 5 percent of full load.

Figure 4. Initial Pinion Response, Q = 34 gpm, Cb = 9.0 mils,Fabricated Baseplate.

Increasing the inlet oil pressure appeared to push the critical upsomewhat so the total bearing oil flow was increased to 45 gpm perbearing and the gearbox was retested. At the same time, sincegearbox support stiffness was believed to be a significant factor, astiffer solid baseplate replaced the original hollow, fabricatedbaseplate. The resulting speed-amplitude plot is shown in Figure 5.Now the pinion critical speed is at 14,200 rpm with a correspondingamplification factor of 14.2.

Figure 5. Interim Pinion Response, Pin = 25 psi, Q = 45 gpm, Cb= 9.0 mils, Solid Baseplate.

Since this still does not meet the API 613 (2003) acceptancecriteria, the inlet pressure was increased from the design value of25 psi to 45 psi in an attempt to easily check the effects of increasingoil flow. The corresponding per bearing total oil inlet flow increasewas from 45 to 60 gpm. The resulting response run is shown inFigure 6 with N1 greater than 15,200 rpm.

Figure 6. Interim Pinion Response, Pin = 45 psi, Q = 60 gpm, Cb= 9.0 mils, Solid Baseplate.

Based on the favorable Figure 6 results, another increase in oilinlet flow seemed appropriate. Further increasing the total oil flowto 55 gpm per bearing (50 gpm lubricating plus 5 gpm by-pass)

and, at the same time, decreasing the bearing clearance by 1.5 milsdiametral (from 9.0 to 7.5 mils diametral, nominal), results inthe speed-amplitude plot shown in Figure 7. The pinion criticalappears to be at 15,250 rpm minimum with A1 = 10.9. Thiscondition finally meets the API 613 (2003) acceptance criteria.These results are summarized in Tables 2 and 3.

Figure 7. Final Pinion Response, Q = 55 gpm, Cb = 7.5 mils, SolidBaseplate.

Table 3. Summary of Mechanical Test Results.

It is important to note that a similar problem also occurred on thelow speed shaft with a maximum continuous speed of 6865 rpm.The low speed shaft problem was solved in a similar manner asdescribed above. Due to space constraints, only the high speedpinion problem will be discussed herein. Additionally, the problemalso occurred on the spare gearbox.Finally, subsynchronous vibration for both the pinion and the

bull gear shafts was not an issue. Spectrum plots show very lowsubsynchronous vibration levels throughout the no-load mechanicaland full load string tests.

ANALYTICAL CORRELATION

In an attempt to match the no-load test results of Figure 4 (N1 =13,600 rpm, A1 = 13.5), a rotordynamics analysis was performedon the pinion. Initially, the tilting pad bearing analysis assumed afull film as the bearing code used in the analysis did not have thecapability to calculate any starvation effects explicitly (Nicholas,et al., 1979). A reasonable gearbox case support stiffness value ofKs = 7.0×106 was assumed. The nominal as-shipped bearingclearance of Cb = 9.0 mils diametral was also used. The resultingspeed-amplitude plot is presented in Figure 8. The pinion critical ispredicted at 13,700 rpm with an amplification factor of 7.6. Thefrequency is a good match but the amplification is predicted to beabout 50 percent of the actual value indicating far less damping inthe system than anticipated.

Figure 8. Predicted Pinion Response, Full Film, Cb = 9.0 mils, Ks= 7.0×106 lbf/in.


Figure 5 test results (N1 = 14,200 rpm, A1 = 14.2) were obtainedwith 45 gpm of total per bearing oil flow plus the stiffer, solidbaseplate. The original fabricated, hollow baseplate is shown inFigure 9. In an attempt to increase the pinion’s critical speed byincreasing the gearbox support stiffness, a new solid baseplate,Figure 10, replaced the original prior to the Figure 5 test. A rap testwas performed on the gearbox with the new, solid baseplate.Results indicated a dynamic support stiffness of 20.0×106 lbf/in ata frequency of 14,000 cpm. Using Ks = 20.0×106 lbs/in and thenominal as-shipped bearing clearance of Cb = 9.0 mils diametral,results in the speed-amplitude plot shown in Figure 11. The pinioncritical is now predicted at 17,200 rpm with an amplification factorof 5.5. The frequency is over predicted by 3,000 rpm and theamplification factor is under predicted by a factor of 2.6. Theseresults are summarized in Table 4.

Figure 9. Original Fabricated Hollow Baseplate.

Figure 10. New Solid Baseplate.

Figure 11. Predicted Pinion Response, Full Film, Cb = 9.0 mils, Ks= 20.0×106 lbf/in.

Table 4. Summary of Analytical Results.

STARVATION MODELING

In an attempt to better match the no-load test results with analyticalpredictions, starvation is included in the tilting pad journal bearinganalysis.A tilting pad journal bearing computer code developed by He(2003) was used to predict the angle from the pad’s leading edgewhere a full film would occur. This predicted angle is 20 degrees. Asimplistic approach would be to assume that the pad arc length iseffectively reduced by 20 degrees, from 70 degrees to 52 degrees. Thisalso reduces the pad pivot offset from the as-machined 65 percent toan effective value of 50 percent. Using these effective values in theoriginal bearing code, Nicholas, et al. (1979), along with Cb = 9.0 milsdiametral and Ks = 20.0×106 lbs/in results in the speed-amplitudecurves shown in Figure 12. Now the pinion critical is criticallydamped. Clearly, this model predicts too much damping.

Figure 12. Predicted Pinion Response, Starvation Model with 52Degree PadArc Length, 50 Percent Pad Pivot Offset, Cb = 9.0 mils,Ks = 20.0×106 lbf/in.

Artificially decreasing bearing stiffness and damping independentof each other until the results match Figure 5 (N1 = 14,200 rpm, A1= 14.2) with Ks = 20.0×106 lbs/in produces the speed-amplitudecurve shown in Figure 13. Now N1 = 14,700 rpm and A1 = 11.3, areasonable match to test results. To obtain this match, the bearingstiffness was decreased to 70 percent of the full film value, which isreasonable, but the bearing damping was decreased to 15 percent of thefull film value, an 85 percent decrease, which is quite unreasonable.

Figure 13. Predicted Pinion Response, Starvation Model with K =70 Percent and C = 15 Percent of Full Film Values, Cb = 9.0 mils,Ks = 20.0×106 lbf/in.


Returning to the code by He (2003), plotting the normalizedbearing stiffness, K, and damping, C, as a function of lubricating oilflow results in the plot shown in Figure 14. The code does predict adrastic decline in K and C, but they decrease at about the same rate.The dots shown on the plot indicate a stiffness value that is 70percent of full film and a damping value that is 15 percent of fullfilm. The K value that is 70 percent of full film occurs at a predictedlubricating flow rate of 65 gpm while the C value that is 15 percentof full film extrapolates out to a lubricating flow rate of 36 gpm.Since both flow rates cannot occur at the same time, 70 percent offull film K and 15 percent of full film C also cannot occur at thesame time. But, they would have to occur together in order to matchthe test results. Thus, it is difficult to envision a starvation model thatmatches the test results. These results are summarized in Table 4.

Figure 14. Starvation Model, Normalized K and CVersus Oil FlowShowing K = 70 Percent and C = 15 Percent of Full Film Values.

STARVATION TESTING

Coincidently, as the gearbox was experiencing problems with a6.0 inch tilting pad journal bearing with an evacuated housing, a4.0 inch diameter evacuated housing bearing was undergoinglaboratory testing (Harris and Childs, 2008). Except for the testbearing being geometrically smaller, the designs were identical.Some of the geometric parameters are four pads, load betweenpivots, 65 percent pad pivot offset and L/D = 1.0 (Table 1). Aspecial test was requested at 12,000 rpm and at zero load with thebearing flow rates varying from a full film to a starved condition(Table 5). These results are shown in Figures 15, direct stiffness,and 16, direct damping.

Table 5. Test Bearing Oil Flow.

Figure 15. 4×4 Inch TPJB Zero Load Test Data—PrincipalStiffness Versus Oil Flow.

Figure 16. 4×4 Inch TPJB Zero Load Test Data—PrincipalDamping Versus Oil Flow.

From Figure 15, the test curves show a barely perceivabledecline in bearing stiffness as the total oil flow decreases from afull film value of 16 gpm to a starved value of 10 gpm.From Figure 16, the test results indicate that the direct damping

increases slightly and then declines as flow decreases. Certainly, nodecrease in damping is evident from Figure 16 that approaches the15 percent of full film value discussed previously.Also notice that full film analytical predictions from Nicholas,

et al. (1979), are included on both plots. The nominal as-builtbearing clearance was used for the analysis. Pivot stiffness wasincluded and calculated by the Hertzian method from Kirk andReedy (1988) and Nicholas and Wygant (1995).

ULTRA HIGH SPEED APPLICATION

Soon after the problem gearboxes were shipped, a similar evacuatedtilting pad bearing design with a 6.65 inch journal diameter was usedfor a high speed balance of a magnetic bearing rotor. This applicationis shown on the plot of Figure 1 in the extreme lower right-hand corner,575 f/s surface velocity, 26 psi unit loading. The tilting pad bearings,used for the high speed balance only, supported the rotor on themagnetic bearing laminations. Since the laminations had a relativelylarge outside diameter, the tilting pad bearing surface velocity wasextremely high with the evacuated bearing running up to 435 f/s. Theactual bearing is shown in Figure 17. The journal surface velocityversus metal temperature plot from one of the initial test runs using theevacuated design is shown in Figure 18, blue line with solid bluecircles (refer to ACKNOWLEDGEMENT section).

Figure 17. Ultra High SpeedApplication—Evacuated TPJ Bearing.

Figure 18. Evacuated Versus Flooded TPJB, High Speed Balanceon Laminations.


The minimum oil lubricating flow requirement for operation at435 f/s with the evacuated design is 36 gpm calculated as describedin the OIL FLOW REQUIREMENTS section. Because of flowrestrictions in the high speed balance facility, the evacuatedbearings were designed for 28 gpm of lubricating flow. After thetest resulting in the evacuated bearing data shown in Figure 18, itwas determined that one of the oil pumps was not operational.Thus, the resulting lubricating flow was 20 gpm and the bearingran 56 percent starved.Furthermore, because of a misunderstanding, it was believed

during the run that the oil drains were pressurized. This was not thecase and the evacuated bearing operated in a vacuum, furtherstarving the bearing due to oil atomization. Even in this extremestarvation condition, the rotor critical was located as anticipatedwith a reasonable amplification factor.

POSSIBLE CAUSES

As stated previously, this gearbox pinion rotor experienced avibration problem during no-load mechanical testing. The locationof the pinion’s critical speed was well below predicted and theamplification factor was well above predicted. Increasing oil flowincreased the critical speed location but had a minor effect onreducing the amplification factor. A similar problem occurred onthe low speed shaft. It was solved in a similar manner. In theauthor’s experience with dozens of gearboxes built in a similarmanner with essentially the same evacuated tilting pad bearing, thiswas the first gearbox to exhibit this problem. However, it must benoted that unbalance testing was not conducted on all of thegearboxes and similar problems of this type may have been missed.Clearly, increasing oil flow had a large influence on the critical

speed location. This seems to indicate that starvation was the causeof the problem. However, there are contrary issues that seem tonegate this conclusion:

• With all of the experience shown in Figure 1, it would seemlikely that this problem would have manifested itself previouslysince all of the bearings were similar in design, have evacuatedhousings, and were designed to operate partially starved at no-load.

• A similar problem did not occur during the magnetic bearingrotor high speed balance with a similar evacuated tilting padbearing design, at very light bearing loads, and in an extremestarvation condition.

• It is a highly unlikely coincidence that the problem finallyshowed itself on both rotors at the same time on the same gearbox.

• Increasing the total per bearing oil flow from 34 to 55 gpm, frompartial starvation to a full film, increased the critical speed frequencybut the amplification factor remained unreasonably high.

• An 85 percent decrease in full film bearing damping isnecessary to match the test data.

• Laboratory test results at no-load for a similar but smallerbearing did not show a dramatic decrease in bearing damping orstiffness as starvation increased.

• Analytical starvation modeling does not predict this dramaticdecrease in bearing damping necessary to match the test stand results.

The conclusion is that increasing oil flow helped to solve theproblem but was not the cause. It may have been a contributor,but not the sole cause of the problem. Other possible causes orcontributors are outlined below.

Air Entrainment

Air entrainment is a condition where air bubbles are trapped orentrained in the lubricating oil. It is a well-known phenomenon forsqueeze film dampers. It has been shown that air entrainment candrastically decrease the damping provided by a squeeze filmdamper (refer to Figures 4 and 6, Tao, et al., 2000).

For an evacuated housing journal bearing, this type of entrainmentmay occur in the starvation region at the leading edge of a tiltingpad. However, if there is inadequate dwell time for the oil inthe reservoir, air bubbles will be present in the oil when it isreintroduced to the bearing. Several weeks after the gearbox wasshipped, the amount of air entrainment present in the oil cooler wasmeasured at less than 5 percent. From Figures 4 and 6 (from Tao,et al., 2000), this is not nearly enough to account for an 85 percentreduction in full film damping.

Mesh Oil and Air Impingement

It may be possible that the oil that exits the gear mesh may jetinto the bearing and interfere with the inlet oil or the lubricatingoil film. Another factor that may affect the bearing is the windagefrom the gear mesh. Since the housing is evacuated, there are noend seals on the bearing. Thus, the journal-to-pad interface andthe lubricating film are exposed to the oil that is forced out of thegear mesh and from mesh windage. On this gearbox, the oil exitsthe gear mesh at a velocity of around 1000 f/s. This may besufficient to interfere with the lubricating film thereby affectingthe bearing’s stiffness and damping properties. Whether thiseffect is sufficient to cause an 85 percent damping decrease fromfull film levels is unknown.To prevent these phenomena from occurring, some gearboxes

have a shield at the end of the gear mesh. A mesh shield was notpresent on the problem gearbox nor has it been used on most ofthe applications shown in Figure 1.Another way to eliminate mesh oil or air impingement is to

place an end shield on the mesh side of the journal bearings. Noshields were present on the bearings for this gearbox. However,the inlet oil is protected from mesh impingement by side shieldson the spray-bar blocker as shown in Figure 19.

Figure 19. Spray-bar blocker Side Shields Protecting the Inlet Oil.

Support Stiffness

From Figure 8, with a reasonable gear case support stiffness of7.0×106 lbf/in, the pinion critical is predicted at 13,700 rpm.However, the predicted amplification factor remains well belowactual. Regardless, with the solid baseplate, the support stiffnesswas measured at 20×106 lbf/in. This is a relatively high valueand, therefore, the problem cause is not likely to be a soft gearcase support.


Pivot Stiffness

To obtain all of the analytical results, the spherical pivotstiffness was included in the calculations. Using the methodoutlined in Kirk and Reedy (1988) and Nicholas and Wygant(1995), the calculated Hertzian spherical pivot stiffness is Kp =24.7×106 lbf/in. The bearing’s spherical pivot was reconstructedand its stiffness measured using a hydraulic press to simulate thepivot load. The measured pivot stiffness was 17.4×106 lbf/in.Using the measured value for Kp, all of the analytical results showvirtually no change. Thus, the problem cause is not likely to be asoft pivot.

Pad Flutter

Pad flutter is a tilting pad bearing phenomenon that may occuron the pads that are located opposite of the load vector. If thesepads become unloaded (Nicholas, 1994), they may not be able tofind an equilibrium position. Pad flutter may cause babbitt damagebut rarely leads to high synchronous vibration. Furthermore, padflutter occurs under heavy loads. Since this problem occurs underlight loads, pad flutter is not a consideration.

Increasing Gearbox Performance Requirements

Gearbox performance requirements have been steadily increasing.As gearbox transmitted torques have increased, gear mesh facewidths have also increased to carry the load while keepinggearbox shaft center distances small and gear pitch line velocitiesdown. Wider face widths lead to longer rotors. Additionally,carburized gearing is often used to meet these higher power andtorque levels. The higher tooth surface hardness achieved bycarburizing permits considerably more torque to be transmittedthan would be possible with a through-hardened gear of the samephysical size. Bearing technology advancements are permittingmore load to be carried by smaller bearings with reduced oil flowand power loss. These factors combine to result in heaviercouplings for torque transmittal without a corresponding increasein the rotor diameter. Longer rotors and heavier couplingsdecrease the rotor’s critical speeds. At the same time, operatingspeeds are increasing. All of these factors apply to this gearbox.Further compounding the problem, the flexible couplings on thisgearbox utilize hydraulic taper fits, resulting in even heaviercouplings and higher overhung moments, further reducing thecritical speeds.

PROPOSED SOLUTIONS AND RECOMMENDATIONS

While the exact cause of the problem is unclear, some solutionsand recommendations are suggested below. Keep in mind that,while gearbox mechanical tests are run at loads in the range of 5 to10 percent of full load, this light load condition for centrifugalcompressor trains is not realistic in the field. Specifically,operation at 10 percent of full load and at maximum speed isunrealistic for centrifugal compressor trains as the minimum loadat maximum speed would be much greater than 10 percent. Thesolutions and recommendations follow:

• The use of an integrally flanged coupling with a low overhungmoment is highly recommended. From the pinion mode shapeillustrated in Figure 20, the pinion critical is clearly controlled bythe coupling end overhang moment. This pinion couplingemployed a shrunk-on hub. An integral flange attachment wouldgreatly reduce the overhung moment. From Figure 13, N1 = 14,700rpm and A1 = 11.3 with a coupling hub attachment. This analysisused the artificially degraded bearing stiffness and dampingproperties to match the test results. Replacing the hub attachmentwith an integral flange attachment with a lower overhung moment,results in Figure 21. Now, N1 = 16,200 rpm and A1 = 9.0 and thisproblem would not have materialized. For this gearbox, there wereno engineering reasons to use a hub attachment instead of anintegral flange.

Figure 20. Pinion Mode Shape with Coupling Hub Attachment.

Figure 21. Predicted Pinion Response, Starvation Model with K =70 Percent and C = 15 Percent of Full Film Values, Cb = 9.0 mils,Ks = 20.0×106 lbf/in, with Integral Flange.

• Another possibility to eliminate this problem is to use a properlycooled flooded bearing. The solution for the low speed shaft on thisgearbox included switching to a flooded bearing design. While theoperating temperatures increased compared to the evacuateddesign, they were still within the customer’s specification. Also, itwas estimated that the gearbox power loss increased by roughly 15percent. A flooded design can be successful in severe application ifproper cooling features are employed. An example is shown inFigure 18, green line, solid green triangles. This was the magneticbearing rotor high speed balance application discussed previouslyin the ULTRA HIGH SPEED APPLICATION section. After someinitial runs up to 435 f/s with the evacuated design, the bearingswere changed to a flooded design with special cooling features.The special flooded tilting pad bearings ran successfully up to asurface velocity of just over 500 f/s. As expected, the floodeddesign ran slightly hotter. For example, at 435 f/s, the floodedbearing ran 20�F hotter compared to the evacuated bearing.

• When using an evacuated housing design, properly size thebearing oil flow. The final configuration for this pinion bearingended up with 55 gpm of total per bearing oil flow. While this is114 percent of the minimum lubricating flow for a full film at zeroload, it is a staggering 192 percent of the minimum flow for a fullfilm at full load. The authors suggest sizing the flow for 150percent of the minimum flow for a full film at full load.

• Use a realistic gear case support stiffness value in the forcedresponse analysis.

• Include the pivot stiffness in the bearing dynamic analysis.• When evacuated housing bearings are used, a conservativeanalytical separation margin should be employed.

CONCLUSIONS

• The subject gearbox pinion rotor experienced a vibrationproblem during no-load mechanical testing.


• The location of the pinion’s critical speed was well belowpredicted and the amplification factor was well above predicted.Increasing oil flow increased the critical speed location but had aminor effect on reducing the amplification factor.

• Increasing oil flow had a large influence on the critical speedlocation. This seems to indicate that starvation was the cause of theproblem. However, there are contrary issues that seem to negatethis conclusion. The major issues are outlined below:

• With all of the experience shown in Figure 1, it wouldseem likely that this problem would have manifested itselfpreviously since all of the bearings were similar in design, hadevacuated housings, and were designed to operate partiallystarved at no-load. However, not all of the gearboxes wereunbalance tested, and this problem may have been overlooked onpast gearboxes.

• A similar problem did not occur during the magnetic bearingrotor high speed balance with a similar lightly loaded, evacuatedtilting pad bearing design in an extreme starvation condition.

• An 85 percent decrease in full film bearing damping isnecessary to analytically match the test data.

- Analytical starvation modeling does not predict thisdramatic decrease in bearing damping necessary to match the teststand results.

• Laboratory test results at no-load for a similar but smallerbearing did not show a dramatic decrease in bearing damping orstiffness as starvation increased during no-load testing.

• From the above, it is concluded that increasing oil flow helpedto solve the problem but was not the cause. It may have been acontributor, but not the sole cause of the problem.

• Other possible causes or contributors include:• Mesh oil and air impingement on the oil film. Mesh oil

impingement is more likely to be a cause for concern with singlehelical gearboxes whereas the problem gearbox has a doublehelical mesh. Air impingement from mesh windage is obviouslypresent in all gearboxes.

• Air entrainment in the lubricating oil.

• The use of an integrally flanged coupling with a low overhungmoment is highly recommended. If one were used on this application,this problem would not have occurred.

• An unbalance test is recommended for all gearboxes with pinionspeeds above 8000 rpm to locate both the bull gear and pinion rotorcritical speeds.

• Consider the use of a properly cooled flooded bearing.• This was the resulting configuration for the low speed

shaft bearings.

• Operating temperatures will increase.

• Power loss will increase.

• Proven successful in severe applications if properly designedwith appropriate cooling features.

• When using an evacuated housing design, properly size thebearing oil flow. A design flow equal to 150 percent of the minimumlubricating flow for a full film at full load is a suggested target.

• Use a realistic gear case support stiffness and include the pivotstiffness in the bearing and forced response analyses.

• When evacuated housing bearings are used, a conservativeanalytical separation margin should be employed.

• Bearing manufacturers, gear vendors, compressor manufacturers,and the end users all need to work together to help prevent similarproblems and to provide the best possible system. To this end:

• Acknowledge that operation at 10 percent of full load atmaximum speed is unrealistic for field operation of centrifugalcompressor trains.

• Relax the no-load mechanical test acceptance criteria. Thiswill allow the bearing designers to design for full load and not forthe no-load mechanical test. It would permit the elimination ofotherwise unnecessary oil flow and power loss, and thus reduce theoperating cost of the equipment, without reducing reliability.

• This gearbox performed flawlessly during the full load string test.No vibration problems were experienced. The bearing temperatureswere all well within specifications.

In summary, the subject gearbox pinion and bull gear rotorsexperienced vibration problems during no-load mechanicaltesting with evacuated housing tilting pad journal bearings. Thebearings were operating at essentially zero load in a partiallystarved condition. Test results indicate that the location ofboth rotor critical speeds were well below predicted with theamplification factors well above predicted. Increasing the oil flowincreased the location of both critical speeds thereby solving thevibration problem. From this, one may conclude that bearingstarvation was the problem cause. However, independent bearingtesting in a starved condition at zero load did not induce asignificant bearing stiffness or damping decrease. Thus, it isconcluded that starvation alone did not cause the problem. Themost probable cause was starvation in conjunction with anothergearbox and/or test stand related phenomena: either mesh airimpingement, mesh oil impingement, or air entrainment in thelubricating oil.

NOMENCLATURE

A1 = First critical speed amplification factor (rpm)Cb = Bearing diametral clearance (mils)C = Bearing principal damping (lbf-s/in)Cxx, Cyy = Horizontal, vertical principle bearing damping,

kN-s/m (lbf-s/in)Dj = Journal diameter (in)K = Bearing principal stiffness (lbf/in)Kp = Pivot stiffness (lbf/in)Ks = Gear case support stiffness (lbf/in)Kxx, Kyy = Horizontal, vertical principle bearing principle stiffness,

MN/m (lbf/in)m = Pad preloadN1 = First critical speed (rpm)Pin = Oil inlet pressure (psi)Q = Oil flow (gpm)Tmax = Maximum bearing operating metal temperature (�F)Wp = Pivot load (lbf)

REFERENCES


Ball, J. H. and Byrne, T. R., 1998, “Tilting Pad HydrodynamicBearing for Rotating Machinery,” US Patent No. 5,795,076,Orion Corporation, Grafton, Wisconsin.

Brockwell, K., Dmochowski, W., and DeCamillo, S. M., 1994,“Analysis and Testing of the LEG Tilting Pad JournalBearing—A New Design for Increasing Load Capacity,Reducing Operating Temperatures and Conserving Energy,”Proceedings of the Twenty-Third Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 46-56.

Gardner, W.W., 1994, “Tilting Pad Journal Bearing Using DirectedLubrication,” US Patent No. 5,288,153, Waukesha BearingCorporation, Waukesha, Wisconsin.


Harris, J. and Childs, D., 2008, “Static Performance Characteristicsand Rotordynamic Coefficients for a Four-Pad Ball-in-SocketTilting Pad Journal Bearing,” Proceedings of ASME TurboExpo 2008: Power for Land, Sea and Air, GT2008-50063.


Kirk, R. G. and Reedy, S. W., 1988, “Evaluation of Pivot Stiffnessfor Typical Tilting-Pad Journal Bearings Designs,” ASMEJournal of Vibration, Acoustics, Stress and Reliability inDesign, 110, (2), pp. 165-171.

Nicholas, J. C., 1994, “Tilting Pad Bearing Design,” Proceedings ofthe Twenty-Third Turbomachinery Symposium, TurbomachineryLaboratory, Texas A&M University, College Station, Texas,pp. 179-194.

Nicholas, J. C., 2003, “Tilting Pad Journal Bearings with Spray-BarBlockers and By-Pass Cooling for High Speed, High LoadApplications,” Proceedings of the Thirty-Second TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&M University,College Station, Texas, pp. 27-38.

Nicholas, J. C. andWygant, K. D., 1995, “Tilting Pad Journal BearingPivot Design for High Load Applications,” Proceedings of theTwenty-Fourth Turbomachinery Symposium, TurbomachineryLaboratory, Texas A&M University, College Station, Texas,pp. 33-48.

Nicholas, J. C., Gunter, E. J., andAllaire, P. E., 1979, “Stiffness andDamping Coefficients for the Five Pad Tilting Pad Bearing,”ASLE Transactions, 22, (2), pp. 112-124.

Tanaka, M., 1991, “Thermohydrodynamic Performance of a TiltingPad Journal Bearing with Spot Lubrication,” ASME Journal ofTribology, 113, (3), pp. 615-619.

Tao, L., Diaz, S., San Andrés, L., and Rajagopal, K. R., 2000,“Analysis of Squeeze Film Dampers Operating with BubblyLubricants,” ASME Journal of Tribology, 122, (1), pp. 205-210.

BIBLIOGRAPHY

Nicholas, J. C., Whalen, J. K., and Franklin, S. D., 1986,“Improving Critical Speed Calculations Using FlexibleBearing Support FRF Compliance Data,” Proceedings of theFifteenth Turbomachinery Symposium, TurbomachineryLaboratory, Texas A&M University, College Station, Texas,pp. 69-78.

ACKNOWLEDGEMENT

The authors recognize Lufkin Industries for their assistance andsupport, Texas A&M Turbomachinery Laboratory for the tiltingpad journal bearing test data, and GE Oil & Gas, Nuovo Pignone,for the magnetic bearing rotor, on-lamination, high speed balancingtest data. Also, the authors thank Marty Maier, Dresser-Rand, forsuggesting that the maximum amount of hot oil carryover musttraverse the bearing pad’s minimum film thickness. Finally,thanks to Dr. John Kocur, ExxonMobil, for providing some of thestarvation bearing data.


Scan M. DeCamillo is Manager ofResearch and Development forKingsbury, Inc., in Philadelphia,Pennsylvania. He is responsible fordesign, analysis, and development ofKingsbury fluid film bearings forworldwide industrial and militaryapplications. He began work in thisfield in 1975 and has since providedengineering support to industry

regarding application and performance of hydrodynamicbearings. Mr. DeCamillo has developed performance andstructural bearing analysis tools during his career, establishingdesign criteria used in many publications and specifications.He has patents and has authored several papers onbearing research, which is currently focused on advancinghydrodynamic bearing technology in high-speed turbomachinery.Mr. DeCamillo received his B.S. degree (Mechanical

Engineering, 1975) from Drexel University. He is a registeredProfessional Engineer in the State of Pennsylvania and a memberof STLE, ASME, and the Vibration Institute.

Minhui He is a Machinery Specialistwith BRG Machinery Consulting LLC, inCharlottesville, Virginia. His responsibilitiesinclude vibration troubleshooting, rotor-dynamic analysis, as well as bearing andseal analysis and design. He is a member ofSTLE, and is also conducting research onrotordynamics and hydrodynamic bearings.Dr. He received his B.S. degree

(Chemical Machinery Engineering,1994) from Sichuan University. From 1996 to 2003, heconducted research on fluid film journal bearings inthe ROMAC Laboratories at the University of Virginia,receiving his Ph. D. (Mechanical and AerospaceEngineering, 2003).

C. Hunter Cloud is President of BRGMachinery Consulting, LLC, inCharlottesville, Virginia, a companyproviding a full range of rotatingmachinery technical services. He began hiscareer with Mobil Research andDevelopment Corporation in Princeton, NJ,as a turbomachinery specialist responsiblefor application engineering, commissioning,and troubleshooting for production,

refining, and chemical facilities. During his 11 years at Mobil, heworked on numerous projects, including several offshore gasinjection platforms in Nigeria, as well as serving as reliabilitymanager at a large US refinery.Dr. Cloud received his B.S. (Mechanical Engineering, 1991) and

Ph.D. (Mechanical and Aerospace Engineering, 2007) from theUniversity of Virginia. He is a member of ASME, the VibrationInstitute, and the API 684 rotordynamics task force.

James M. Byrne is currently a member ofthe BRG Machinery Consulting team, inCharlottesville, Virginia. BRG performsresearch and analysis in the fields of fluidfilm bearings, magnetic bearings, androtordynamics. Mr. Byrne began his careerdesigning internally geared centrifugalcompressors for Carrier in Syracuse, NewYork. He continued his career at Pratt andWhitney aircraft engines and became a

technical leader for rotordynamics. Later Mr. Byrne became aprogram manager for Pratt and Whitney Power Systems managingthe development of new gas turbine products. From 2001 to 2007,he was President of Rotating Machinery Technology, a manufacturerof tilting pad bearings.Mr. Byrne holds a BSME degree from Syracuse University, an

MSME degree from the University of Virginia, and an MBA fromCarnegie Mellon University.

JOURNAL BEARINGVIBRATIONAND SSV HASH

byScan M. DeCamillo

Manager, Research and Development

Kingsbury, Inc.

Philadelphia, Pennsylvania

Minhui HeMachinery Specialist

C. Hunter CloudPresident

andJames M. Byrne

Machinery Consultant

BRG Machinery Consulting, LLC

Charlottesville, Virginia 22903 USA

11

ABSTRACT

Peculiar, low-frequency, radial vibrations have been observed invarious turbomachinery using tilt-pad journal bearings. Unlikediscrete subsynchronous spikes that often indicate a seriousproblem, the vibrations are indiscrete and of low frequency andamplitude. The low level shaft indications have raised concern inwitness tests of critical machinery, even in cases that comply withAmerican Petroleum Institute (API) limits, owing to uncertaintyregarding the cause and nature of the vibrations.This paper presents shaft and pad vibration data from various

tilt-pad journal bearing tests that were undertaken to investigate andbetter understand these subsynchronous indications. The vibrationcharacteristics are defined and compared under the influence ofspeed, load, oil flow, and bearing orientation. Results are presentedfor conventional and direct lube tilt-pad bearing designs, along withdiscussions of parameters and methods that were successful inreducing and eliminating these low level vibrations.The test results indicate that the low-frequency shaft indications

are caused by pad vibration. Hypotheses and analyses arepresented and discussed in relation to the test observations.

INTRODUCTION

Many early technical papers report on the inherent stability oftilt-pad journal bearings in overcoming oil whirl and oil whiplimitations of fixed geometry bearings. There are, however, othervibration phenomena associated with tilt-pad bearings that aretopics of past and present research. In the late 70s, extensivebabbitt fatigue cracking of upper, unloaded pads was a majorproblem in large, conventional tilt-pad journal bearings. Thesewere flooded designs, which incorporate end seals to restrict oiloutlet and flood the bearing cavity. A well-referenced study byAdams and Payandeh (1983) attributes the damage to self-excited,subsynchronous pad vibration.Trends for larger and higher speed turbomachinery impose

greater demands on bearings and rotor dynamics. Direct lubebearings have evolved to reduce the higher power losses, oil flowrequirements, and pad temperatures associated with higher surfacespeeds. Direct lubrication is not a new concept as designs have beenused in special applications with a long history of reliability. Use ofdirect lube journal bearings became more prevalent as machine sizeand speed increased, eventually spawning several papers in the early90s investigating their steady-state performance. These includeresearch by Booser (1990), Tanaka (1991), Harongozo, et al.(1991), Brockwell, et al. (1992), and Fillon, et al. (1993).The references document pad temperature reductions derived

from the efficient evacuation of hot oil from the bearing cavity.Direct lube bearings are therefore typically designed for evacuatedoperation, accomplished by opening up end seal and oil outletrestrictions. A direct application of oil to the journal surfaceprevents oil from bypassing the films, which can occur inconventional bearings when outlet restrictions are removed. Use ofnozzles to spray oil on the journal surface between pads is onemethod of direct lubrication. An additional pad temperatureadvantage is gained by direct lubrication features.Literature on vibration characteristics of direct lube journal

bearings is not as prevalent. DeCamillo and Clayton (1997) presentrotordynamic data for large, 18 inch (457 mm) generator bearings.The tests showed comparable vibration response for conventionaland direct lube designs. Edney, et al. (1996), provide similarinformation for a small, high-speed, multistage steam turbine with4 inch (102 mm) diameter journal bearings. Peculiar, low-frequency,radial vibrations were observed during these steam turbine tests, butlevels were low and acceptable. However, similar vibrations in ahigh-speed compressor during acceptance tests in 1999 did encroachupon acceptable API limits, and significant time and resources wereexpended to address this issue (DeCamillo, 2006).Personal experience and discussions among original equipment

manufacturers (OEMs) and users over the past few years indicate

that these low-frequency vibrations have been encountered inturbines, compressors, and gearboxes, using conventional anddirect lube tilt-pad bearing designs. The signature has also beendocumented in separate research investigating stability (Cloud,2007). Accurate stability prediction is a major topic of concern inthe industry. Kocur, et al. (2007), highlight a large spread inpredicted bearing stiffness and damping coefficients amongcomputer codes, and associated ramifications regarding stabilityassessment of critical turbomachinery. At the same time, there arenot many codes available that address direct lube performance anddynamic coefficients (He, 2003; He, et al., 2005).Researchers are presently attempting to sort through some difficult

questions. What is the source of the low-frequency vibrations? Arethey attributable to direct lube bearings, pad flutter, starvation,high speeds, or low loads? Do they affect machine stability, safety,or reliability?This paper presents a chronology of tests, investigations, and

theoretical analyses aimed at providing answers to some of thesequestions. It is desired that the information be of value toresearchers, OEMs, users, and other personnel involved withhydrodynamic bearings and vibration in turbomachinery.

SSV HASH, DEFINITION

Figure 1 is an example radial vibration spectrum from a high-speedcompressor test using direct lube tilt-pad journal bearings. Themain subject of this paper is the low-frequency shaft vibrationsindicated in the figure. Unlike discrete subsynchronous spikes thatoften indicate a serious problem, the vibrations are indiscrete andof low-frequency and amplitude. The plot is paused to capturethe random, broadband frequencies. In order to distinguish thesubsynchronous vibration (SSV) characteristics under consideration,the term SSV hash is defined for use in this paper:

• SSV Hash: A vibration signature characterized by low-frequency,low amplitude, broadband subsynchronous vibrations thatfluctuate randomly.

Figure 1. Example Radial, Low-Frequency, Broadband, Vibration.

Experience with turbomachinery bearing sizes 3.88 to 8.00 inches(100 to 200 mm) in diameter has observed SSV hash frequenciestypically ranging up to 30 Hz with amplitudes on the order of 0.1 to0.2 mils (.0025 to .0050 mm) peak-to-peak.

INITIAL INVESTIGATIONS

Delayed acceptance of a high-speed compressor due to SSV hashin 1999 prompted a research project performed on a high-speed testrig described in detail in a separate reference (Wilkes, et al., 2000).Pertinent information is provided here for convenience. The rig hasa test shaft driven by a variable speed gas turbine through a flexiblecoupling (Figure 2). The test shaft is approximately 5 feet (1.5 m)long and 5 inches (127 mm) in diameter, supported at either end bya pivoted shoe journal bearing. Two orthogonal proximity probesare mounted inboard of each journal bearing, ± 45 degrees offtop-dead-center, to record radial shaft vibration. A spectrumanalyzer was used to acquire fast Fourier transform (FFT) vibration


signatures, presented in paused plots in Figures 3 through 5. They-axis represents peak-to-peak amplitude in mils.

Figure 2. High-Speed Test Rig Journal Bearings and Shaft Schematic.

Figure 3. Evacuated Discharge Configuration, 3.0 gpm (11.4 l/min).

Figure 4. Flooded Discharge Configuration, 6.0 gpm (22.8 l/min).

Figure 5. Evacuated Discharge Configuration, 3.0 gpm (11.4 l/min),SSV Grooves.

Fortunately, it was possible to duplicate the low-frequencyvibration signature in the test rig, which allowed a parametricstudy of many journal bearing designs and configurations over thecourse of this initial investigation. The data presented in Figures 3through 5 are for direct lube, five-pad, leading-edge-groovejournal bearing tests at a shaft speed of 10,000 rpm and a low, 20psi (0.14 MPa) projected load. The bearings have a nominaldiameter of 5 inches (127 mm) and an axial length of 2.25 inches(57 mm). The pads are steel backed with a babbitt surface andhave a 60 degree angle and a 60 percent offset pivot. Theassembled bearing diametric clearance is 0.009 inch (0.23 mm)and the nominal preload is 0.15. Tests were run with ISO VG 32turbine oil supplied at 120�F (49�C).The SSV hash indicated in Figure 3 was recorded for an oil

flow of 3.0 gpm (11.4 l/min) for the direct lube bearing in itsas-designed, evacuated oil discharge configuration. Tests foundthat increasing the oil flow tended to reduce the amplitudes butdid not entirely eliminate the SSV hash signature. Eliminationrequired the installation of floating oil seals as well as an increasein oil flow to 6.0 gpm (22.8 l/min), the results of which are shownin Figure 4. This solution unfortunately required higher oil flowand operated with higher power loss and pad temperatures than theoriginal design.Methods were therefore pursued to address these performance

issues, and results began to suggest that the low frequenciesmay be due to air entering the oil film. Based on this hypothesis,a design was conceived that might eliminate SSV hash whilemaintaining some direct lube benefits. The pads were modifiedwith narrow circumferential SSV grooves, cut in the babbittnear the edges of the pads (Figure 5), to capture and redirectside leakage toward the leading edge of the next pad (Wilkesand DeCamillo, 2002). In this way, additional oil is madeavailable to the oil films without increasing the bulk oil flow tothe bearing.The SSV groove pads were installed and tested in the original

evacuated condition (oil seals removed) with results shown inFigure 5 for 3.0 gpm (11.4 l/min), and tested as low as 2.0 gpm (7.6l/min) with negligible SSV hash indications. Of several methodspursued during the course of this initial investigation, the grooveswere the only solution successful in eliminating SSV hash in anevacuated configuration. The low oil flow and power loss ofthe original design were maintained, with a slight penalty in padtemperature due to the introduction of warm, side leakage oil backinto the oil film.Another observation of this initial investigation was the

tendency for an increase in synchronous amplitudes when SSVhash was eliminated, noticeable by comparing both the floodedsolution (Figure 4) and the evacuated groove solution(Figure 5) with the original open discharge configuration dataof Figure 3.

SUBSEQUENT TESTS

Although solutions were obtained in initial investigations,operating conditions were limited in load and speed. A subsequentseries of tests was initiated in 2005 to further investigate thelow-frequency vibrations. The same test rig was modified toincorporate a radial load system in the bearing housing at the freeend of the shaft, shown schematically in Figure 6. The systemincorporates a hydraulic cylinder that loads through a dowel to asingle journal bearing loader shoe on top of the shaft. This pushesdown on the shaft and loads the test bearing. The applied load ismeasured by a load cell below the hydraulic cylinder. A new dataacquisition system was used to acquire the data, and additionalproximity probes were installed to monitor the vibration of anupper and a lower journal bearing pad. The pad proximity probeswere mounted in the bearing casing and targeted the trailing edgeof the back of the pads at the locations depicted schematicallyin Figure 7.

13JOURNAL BEARINGVIBRATIONAND SSV HASH

Figure 6. Test Rig Radial Load System Schematic, Free End of Shaft.

Figure 7. Bearing Orientation and Pad Numbering.

The test bearings consist of a conventional center pivot bearingwith labyrinth end seals (Figure 8) and evacuated direct lube centerand offset pivot bearings. Three direct lube methods were tested:spray nozzles (Figure 9), leading-edge-grooves (Figure 10), andbetween-pad-grooves (Figure 11). Each bearing was tested forload-on-pad and load-between-pad orientation over a range ofloads, oil flows, and speeds.

Figure 8. Conventional Bearing with Labyrinth End Seals.

Figure 9. Direct Lube Spray Nozzles, Evacuated Configuration.

Figure 10. Direct Lube Leading-Edge-Groove, EvacuatedConfiguration.

Figure 11. Direct Lube Between-Pad-Groove, EvacuatedConfiguration.

Bearing geometry, test oil viscosity, and inlet temperature arethe same as in earlier investigations except that the assembleddiametric clearance is 0.0072 inch (0.18 mm) and the nominalpreload is 0.25. Loads indicated in the following sections andfigures refer to the applied radial load. For data labeled no-load,the journal bearings are supporting the weight of the shaft,which gives a 20 psi (0.14 MPa) projected unit bearing load.Each test bearing is installed around the free end of the shaftwithout disturbing the coupling or the coupling end journalbearing. The same coupling end journal bearing is used in alltests. Vibration data were recorded while varying speed to14,000 rpm, load to 400 psi (2.8 MPa), and oil flow to 30 gpm(114 l/min), holding two of the parameters constant while theother was varied.The speed ramp data are used in this paper, presented in

color-coded FFT waterfall diagrams with a vertical scaledenoting peak-to-peak amplitude in mils. With Figure 12 as anexample, each waterfall diagram contains data for three separaterun-up/run-down speed ramps from 5000 to 14,000 to 5000 rpm(83 to 233 to 83 Hz), one for each of three oil flows. The backplane of the diagram contains a projection of all data. Vibrationsbelow 1 Hz are cut off to delete spurious 0 Hz indications fromobscuring the low frequencies under investigation and the dataare uncompensated, the main focus being the low-frequencyvibrations. As a guide, SSV hash amplitudes above 0.1 mils(0.0025 mm) peak-to-peak are levels that have caused concernin acceptance tests.


Figure 12. Conventional Center Pivot, LBP, No-Load.

Shaft SSV Hash Trends

The volume of data for the different bearing designs allowed for anassessment of trends in shaft SSV hash. This section describes thosethat were fairly common for all test bearings. Conventional bearingspeed ramp data in Figures 12 through 15 are used for reference.

Figure 13. Conventional Center Pivot, LOP, No-Load.

Figure 14. Conventional Center Pivot, LBP, 400 psi (2.8 MPa).

Figure 15. Conventional Center Pivot, LOP, 400 psi (2.8 MPa).

Comparing data among the figures, it can be noticed that mostSSV hash indications occur at low flow and low load. The word“indications” is carefully used in this generalization. Althoughthere were more indications at low flow and low load for mostbearing designs, SSV hash amplitudes were sometimes higher atother operating conditions, as evidenced in results later in this paper.Other common trends derived from the study are that SSV hash

levels for no-load data were similar for load-between-pad (LBP)and load-on-pad (LOP) orientations (e.g., Figures 12 and 13). Acurious result was that while SSV hash decreased with applied loadfor load-between-pad orientation (e.g., Figure 12 versus 14) therewas only a small change with load for load-on-pad orientation(e.g., Figure 13 versus 15). Consequently, load-on-pad orientationhad higher SSV hash levels than load-between-pad orientation athigher loads for all test bearings.It is worth noting that initial investigations and field reports were

for high-speed applications with relatively light rotors, and so SSVhash was formerly associated with high-speed and low-load. Thetest data in these subsequent series of tests indicate that this is notexactly true. Although indications were more pronounced at lowloads, tests indicate there can be undesirable levels of SSV hash inthe case of load-on-pad orientation at higher loads, for exampleFigure 15 at 400 psi (2.8 MPa). There were also many indicationsat lower speeds in the test data, again using Figure 15 as an example.

Conventional Versus Direct Lubrication

Figure 16 displays no-load data for the conventional center pivottest bearing with load-on-pad orientation, the same used in Figure13, but zoomed in on the lower frequencies and amplitudes underinvestigation. Figure 17 contains comparable, center pivot,between-pad-groove direct lube data. A condition mentionedearlier is noticeable in the direct lube data, i.e., there are more SSVhash indications at low flow but with higher amplitudes at anintermediate condition, 8 gpm (30 l/min) and 11,000 rpm in thisparticular test. The figure also supports observations from fieldexperience and initial investigations where SSV hash decreases,but is not necessarily eliminated at higher flows.Comparisons between conventional and direct lube results (Figures

16 and 17) are fairly straightforward. Direct lube data have noticeableSSV hash indications at 8 gpm (30 l/min) whereas conventionalbearing levels are negligible. Conversely, very high subsynchronousvibrations are noticeable in conventional bearing data at 4 gpm (15l/min) where the direct lube design has significantly lower SSVhash indications.These data indicate that flooding and oil flow are not the only

key parameters influencing SSV hash. Conventional bearing testswith labyrinth seals (Figure 16), and initial tests with oil seal rings

15JOURNAL BEARINGVIBRATIONAND SSV HASH

(Figure 4), both required some increase in oil flow to eliminate thesignature, whereas higher oil flow is not as effective in evacuateddesigns (e.g., Figure 17). This suggests cavity pressure may beanother influential parameter that, unfortunately, was not measured.Calculated pressure drops across oil discharge restrictions indicatethat cavity pressures between 0.5 and 2.0 psig (0.04 and 0.14 bar)were present when SSV hash was noticeably reduced. Also fittingthis scenario are the lower SSV hash indications in direct lubedesigns compared to the conventional bearing data at low flow(Figure 17 versus 16), which may be attributable to the more directapplication of oil to the oil films.

Figure 16. Conventional Center Pivot, LOP, No-Load.

Figure 17. Direct Lube Between-Pad-Groove, Center Pivot, LOP,No-Load.

Center Versus Offset Pivot

Center versus offset pivot tests were conducted for several directlube bearing configurations. Comparisons have been difficult toquantify. For example, spray nozzle offset pivot data (Figure 18)have high amplitudes and a broad band of SSV hash midway throughthe 4 gpm (15 l/min) data speed ramps at approximately 12,000rpm, indicated in the figure. At the same conditions, spray nozzlecenter pivot data (Figure 19) have less broadband indications, andamplitudes are actually highest at approximately 6000 rpm. In 8 and12 gpm data (30 and 45 l/min), the offset pivot has slightly less SSVhash. Similar comparisons for between-pad-groove tests (Figure 17versus 20) show lower SSV hash levels in the offset pivot data at4 gpm (15 l/min), and only subtle differences at the higher flows.

Figure 18. Direct Lube Spray, Offset Pivot, LOP, No-Load.

Figure 19. Direct Lube Spray, Center Pivot, LOP, No-Load.

Figure 20. Direct Lube Between-Pad-Groove, Offset Pivot, LOP,No-Load.

The results were unexpected in that offset pivot designs requiremore oil flow to the films, and it was suspected that this would beclearly reflected in SSV hash levels. The data indicate there arefactors other than bulk oil flow that need to be taken into accountregarding SSV hash and pivot offset.


Variations Among Direct Lube Designs

Three offset pivot, direct lube bearings were tested to compareshaft SSV hash characteristics among different methods of directlubrication. All three have an evacuated discharge configuration.The methods include spray nozzles, leading-edge-grooves, andbetween-pad-grooves. Photographs of the direct lube features areprovided in Figures 9 through 11, and corresponding vibration dataare contained in Figures 19 through 21.

Figure 21. Direct Lube Leading-Edge-Groove, Offset Pivot, LOP,No-Load.

In general, all three direct lube methods display noticeable SSVhash indications. Variations in amplitudes and frequencies make itdifficult to generalize the results. The spray nozzles produced thehighest levels of SSV hash. The leading-edge-groove operatedwith lower SSV hash amplitudes but at higher frequencies, andbetween-pad-groove SSV hash levels were lower for all conditionsincomparison with spray nozzle data. Results comparing between-pad-groove and leading-edge-groove data vary with operating conditions.SSV hash levels at 4 gpm (15 l/min) are negligible and lower than theleading-edge-groove. At higher flows, between-pad-groove SSV hashamplitudes are higher but occur at a lower frequency.Distinct differences in shaft SSV hash signatures are obvious

among the designs. Some speed dependence is noticeable in spraynozzle data (Figure 18), and more so in leading-edge-groove data(Figure 21), which appears to be suppressed at higher speeds.Between-pad-groove characteristics (Figure 20) do not showvariations with speed, and seem sensitive to a particular speed banddepending on flow. There are also many conditions throughoutFigures 18 through 21 where the evacuated direct lube bearingsoperate with negligible SSV hash indications.While the reasons for the variations are still under investigation,

the fact that there are differences provides valuable information.Since the pad geometry is the same for the different bearings, thedirect lube feature itself is influencing the shaft SSV hash and,because these are evacuated designs, the influence is most likelyoccurring at the entrance or leading edge of the oil film.

Pad/Shaft Correlation Investigation

A review of 2005 vibration data from the upper and lower padproximity probes (depicted in Figure 7), noted more subsynchronousindications in the upper pad, with the highest amplitudes occurring athigh-load and low-flow conditions in many cases. This is reasonableconsidering that the upper pad clearance increases as the shaft ispushed down under load, away from the upper pad. The upper pad oilfilm consequently generates lower hydrodynamic forces and requiresmore oil flow to fill the gap. Unfortunately, an unexpected result ofthe investigation was that there were only a few random correlationsbetween upper pad and shaft SSV hash for the broad range ofoperating conditions and bearing configurations tested.

This lack of correlation prompted another series of tests in2006 with all pads monitored by proximity probes, installedand positioned as described earlier, and made possible by theacquisition of new high-speed vibration equipment. Figure A-1(refer to APPENDIX A) organizes shaft and pad vibration datafor a leading-edge-groove bearing test at 200 psi (1.38 MPa),for a load-on-pad configuration as an example. The waterfalldiagrams are arranged so that data for the orthogonal shaftprobes and all five pads can be viewed in relation to oneanother. The same scales are used for all waterfall diagrams inthe figure. It is important to note that the shaft was removed andrefurbished between 2005 and 2006 tests and a differentcoupling end journal bearing installed, so data are not comparableto information from earlier tests.Referring to FigureA-1, it can be noticed that the subsynchronous

vibrations in unloaded pad 2 at 4 gpm (15 l/min) do not correlatewith either of the orthogonal shaft probes (labeled +45 degrees and�45 degrees in the figure). A unique observation is that the sidepads 3 and 5, which were not monitored in previous tests, have thestrongest subsynchronous vibrations and show excellent correlationwith the shaft SSV hash indications.

Pad/Shaft Correlation General Trends

The observations from Figure A-1 relate to a specific bearingand set of operating conditions. Comparisons of data for all fivepads, over a broad range of operating conditions for the differentbearing designs allowed for a more precise assessment of pad/shaftinteraction. The following observations and trends were determinedto be fairly common for all test bearings.All shaft SSV hash indications observed in these tests were

confirmed to correlate with vibrations from at least one of the fivepads. The converse is not true. There were many instances wheresubsynchronous vibrations from individual pads did not appear inany shaft data. An important observation is that the shaft SSV hashindications most often correlated with one or more of the side pads,depending on bearing type and orientation. The orientation and padnumber schematic in Figure 7 is helpful in explaining the followingobservations and trends.In the case of load-between-pad orientation under some light

downward loading, side pad 1, side pad 3, or both tended tocorrelate with the shaft SSV hash indications. Upper pad 2 hadmore and higher subsynchronous vibration than the side pads inmany cases, but rarely correlated with the shaft indications. Notedearlier, shaft SSV hash decreased with applied load in the case ofload-between-pad orientation, which is reasonable considering thehigh horizontal stiffness of this orientation with the shaft supportedbetween the two bottom pads. At the same time, pads 1, 2, and 3become less strongly coupled to the shaft and may still experiencesubsynchronous vibrations, but with hydrodynamic forces tooweak to affect the shaft.For a load-on-pad orientation under some light downward

loading, the side pads (1, 2, 3, and 5), individually or in combinations,tended to correlate with the shaft SSV hash indications. Thecorrelation varied with bearing type and operating conditions, andis also likely influenced by any slight skew in applied load vectorand differences in manufacturing heights of the individual pads forthis particular orientation. Noted earlier, shaft SSV hash did notdecrease as much with applied load for load-on-pad orientation,which is reasonable considering the weak horizontal stiffness forthis orientation. Pad 3 and pad 5 tended to correlate more with theshaft SSV hash indications as load was applied. At the same time,pads 1 and 2 become less strongly coupled to the shaft and may stillexperience subsynchronous indications, but with hydrodynamicforces too weak to affect the shaft.

SSV Grooves, Recent Tests

The SSV groove modification (Figure 5), developed duringinitial investigations in 1999, has since been successfully applied in

JOURNAL BEARINGVIBRATIONAND SSV HASH 17

many compressor, turbine, and gearbox applications. At the time,development was based on hypotheses and shaft vibration data.Upgrades in the test rig capability and in high-speed data acquisitionequipment has more recently allowed an investigation of thegroove’s influence on pad vibration.SSV grooves were machined in the leading-edge-groove pads

used for results presented in FigureA-1 and were tested over the fullrange of operating conditions for load-on-pad and load-betweenpad orientation. Test results for the same conditions and orientationas Figure A-1 are presented in Figure A-2 (refer to APPENDIX A)as an example comparison. The tests found that subsynchronouslevels in all pads were significantly reduced or eliminated over thefull range of test operating conditions and configurations, andconfirmed the effectiveness of the method in eliminating shaft SSVhash in evacuated, direct lube bearing designs.

THEORETICAL INVESTIGATIONS

To help shed some light on these SSV hash measurements, thebearing geometry and operating conditions corresponding with thedata in FigureA-1 were modeled using an algorithm developed by He(2003). This algorithm includes the effects of supply flowrate in itsdetermination of a bearing’s steady-state (eccentricity, temperatures,power loss, etc.) and dynamic (stiffness and damping) performancebased on the theories and experiments put forth by Heshmat(1991). Since the measurements presented in these SSV hash testsdemonstrate the importance of supply flow on SSV, any theoreticalinvestigation must account for supply flowrate effects.

Supply Flowrate Effects

The model applies to any bearing configuration where thesupply flowrate is not sufficient to ensure a full film across all thepads’ surfaces. In this case, the model will predict a partialstarvation at the inlet region of some of the pads. Unlike the fullfilm situation where a continuous lubricant film starts at the padleading edge, in a partially starved situation, the continuous film isformed downstream at some point where the clearance is sufficientlyreduced. This physical phenomenon was observed and studied byHeshmat (1991) on a two axial groove, sleeve bearing.Figure 22 illustrates predicted starvation effects for pad number 3,

corresponding with the test data pad numbers in Figure A-1. With asupply flowrate of 8 gpm to the bearing, the film is able to generatepressure across the pad’s entire arc length. Reducing the bearing flowto 4 gpm lowers overall pressure distribution, but the pad stillproduces pressure across almost all of its arc length. The pad isclearly partially starved at 3 gpm where no film pressure is producedat the inlet film region. Decreasing the flow to 2 gpm expands thestarved inlet region. At this flow level, pressure levels are very smalland the pivot is nearly centered within the positive pressure region.Partial starvation has effectively reduced the pad’s pivot offset, whichmay explain unexpected results in center versus offset pivot tests.

Figure 22. Partial Starvation Effects on Pad Film Pressure.

Pad Responsiveness

Since test results confirm that the individual pad motions areinvolved in this SSV hash phenomenon, the tilting-pad bearing’sfull coefficients predicted by the modeling algorithm are ofprimary interest. For a tilting-pad bearing with Npad pads, the fullcoefficients consist of 5Npad+4 stiffness and 5Npad+4 dampingcoefficients. Described in detail by Shapiro and Colsher (1977)and Parsell, et al. (1983), these full coefficients relate pad andshaft motions to forces and moments on the shaft and pads. Forrotordynamic purposes, the magnitude of each individual coefficientis overlooked, focusing instead on their combined or “reduced”coefficients (four stiffnesses and four dampings). Closer scrutinyand understanding of these full coefficients are required when paddynamics become the focus.Table 1 presents the stiffness full coefficients for a 60 percent

offset, LOP, five-pad bearing with 12 gpm supply flowrate where:

�Fx and �Fy = Horizontal and vertical forces on the shaft�x and �y = Horizontal and vertical shaft displacements�M1 … �M5 = Moments on pads 1 through 5��1 … ��5 = Tilt angle changes on pads 1 through 5

Table 1. Stiffness Full Coefficients for the 60 Percent Offset LOPBearing, 11,475 rpm, 200 psi.

Each value in Table 1 represents a stiffness, either �Fx or �Mdivided by �x, �y, or �� depending on the position in the table.Examining a few of these Table 1 coefficients will help explain thephysical meaning behind them. Tilt angle changes of the fourth pad(��4) produce the largest vertical force on the shaft (�Fy), since ithas the highest value of Ky� (�3.261e6 lbf/rad). This is due to thefourth pad being on the bottom, supporting the majority of theload. Conversely, tilt angle changes on the second pad, located inthe upper half of the bearing, produce the smallest vertical force onthe shaft (Ky� = 0.499e6 lbf/rad). The fourth pad is also the mostdifficult to tilt with the highest tilting stiffness K�� value (1.614e6lbf-in/rad), while the second pad is the easiest to tilt (K�� = 0.245e6lbf-in/rad). Tilting motions of the other four pads do not directlycause any moments on the loaded pad. The pads can only effectivelycommunicate to each other through the shaft.Some of the full coefficients along with the pads’ polar inertia (Ip)

form the equation of motion for pad tilting. A simplified version ofthis equation, using a pad’s tilting stiffness K�� and damping C��, isgiven by:

where Mext is an external moment being applied. This equation canbe used to examine the frequency response characteristics of eachpad’s tilting motion. Each pad’s frequency response function (FRF)H(�) is given by:

An FRF’s magnitude indicates how responsive a pad is toexternal moments trying to excite it. Examining the predictedFRFs of each pad as a function of supply flowrate provides for adirect comparison with the pad waterfall diagrams of the test data.


Figure 23 presents the predicted H(�) of pad 3 in Figure A-1 asan example, focusing on the low frequency region below 60 Hz.Regardless of flow, the frequency response remains relatively flatas a function of excitation frequency with no peaks present. This isbecause the pad’s tilting motions remain overdamped for all theflowrates examined. An overdamped system (critical dampinggreater than 100 percent) has no damped natural frequency,resulting in no peaks in its response.

Figure 23. H(�) with Decreasing Supply Flow.

It is unclear as to the exact cause of the 8 Hz peak in FigureA-1’s waterfall diagram for pad 3. One explanation is that theremay be a predominant excitation at this frequency. Heshmat(1991) observed a pulse excitation at similar low frequencylevels in his starvation experiments on a sleeve bearing.Another possibility may be that the pad is actually underdamped,which the model has not been able to predict. In this case, abroadband excitation would result in a definitive peak at thepad’s damped natural frequency. Nothing indicates that thepeak is associated with an unstable, self-excited phenomenon,since the pad’s tilting damping C�� remains greater than orequal to zero. All indications are that the observed vibrationsare forced in nature.Both the measurements in Figure A-1 and the predicted FRF in

Figure 23 show that the third pad’s overall responsiveness doesdramatically increase as supply flow is reduced. According tothe predictions in Figure 23, this pad responds very little toexcitations until the bearing flow is reduced below approximately4 gpm. This increased sensitivity correlates well with the predictedpressure profiles in Figure 22. Below 4 gpm, the predictions inFigure 22 show that the pad becomes partially starved at theleading edge with low overall film pressures. These lower filmpressures allow the pad to respond to external moment excitationsmore easily.According to the H(�) equation, the low frequency responsiveness

of an individual pad is largely dictated by its tilting stiffness, K��.Figure 24 presents the calculated tilting stiffness of all five pads assupply flow is varied. With the exception of the fourth pad,reducing supply flow causes a decrease in all the pads’ tiltingstiffnesses. Below 5 gpm, the fourth pad’s stiffness begins toincrease as it supports more of the entire bearing load. The strengthof this tilting stiffness means the fourth pad does not easily respondto excitations, which correlates well with its relatively low vibrationsin Figure A-1.

Figure 24. Tilting Stiffness of Individual Pads Versus Supply Flow.

When a pad becomes unloaded, it applies no pressure force onthe shaft and its full coefficient stiffness and damping terms go tozero. The pad then moves like a rigid body and its tilting FRF isdictated by the polar inertia. The upper pads (1 and 2) becomeunloaded first as flow is reduced (Figure 24). This is because theirlarger film thicknesses require the most flow to maintain a full filmalong their entire surface. With smaller film thicknesses, the thirdand fifth pads remain loaded until below 2 gpm. These two pads’tilting stiffness values are low in magnitude at reduced flowrates,resulting in the increased responsiveness observed in Figure 23.

CONCLUSIONS

A series of tests and analyses were performed to investigate apeculiar, low-frequency, low amplitude, broadband subsynchronousvibration, termed SSV hash, that has been witnessed in differenttypes of turbomachinery using tilting pad journal bearings.Based on a study of test results for conventional and direct lube

designs over a broad range of speeds, loads, and oil flows, thefollowing shaft SSV hash trends were found to be fairly commonfor all test bearings:

• There were more SSV hash indications at low flow and lowload, although amplitudes were sometimes higher at otheroperating conditions.

• SSV hash levels at light loads were similar for load-between-padand load-on-pad bearing orientations.

• Shaft SSV hash decreased with applied load for load-between-padorientation, but there was only a small change with load forload-on-pad orientation.

• Consequently, load-on-pad operation produced higher SSV hashlevels than load-between-pad orientation at higher loads.

The following general trends for pad/shaft vibration correlationwere found fairly common for all test bearings based on tests ofbearing designs with vibration measurements of all five pads:

• All shaft SSV hash indications were confirmed to correlate withvibrations of at least one of the five pads.

• The converse is not true. There were subsynchronous vibrations inindividual pads that did not appear in measured shaft vibration data.

• The side pads most often correlated with shaft SSV hashindications. The top upper pad for load-between-pad orientationhad the highest subsychronous indications in many cases, butrarely correlated with any shaft data.


The following specific observations were noted in test datacomparisons of bearing types and geometries:

• It is incorrect to associate SSV hash only with high-speed,low-load applications. Indications were noted over a broad range ofoperating conditions.

• Shaft SSV hash has been observed in conventional bearing designsat low oil flow. The data indicate that merely flooding the cavity isinsufficient to suppress shaft SSV hash. At these conditions, directlube designs operate with lower levels of shaft SSV hash, attributedto the more direct application of oil to the film.

• At intermediate test oil flows, shaft SSV hash is more pronouncedin evacuated, direct lube bearings and can be reduced, but notnecessarily eliminated, by increasing oil flow. Amplitudes appearsensitive to certain operating conditions, around which are conditionswhere the evacuated direct lube bearings operated with negligibleSSV hash indications.

• There were mixed results in SSV hash comparisons of centerversus offset pivot direct lube test data. The offset pivot pads inmost cases operated with the same to slightly less SSV hashindications than the center pivot pads.

• Comparisons among three direct lube methods noted distinctdifferences in shaft SSV hash signatures, the results suggesting thatthe differences originate at the entrance or leading edge of theoil film.

Overall, the various tests and comparisons indicate there arefactors other than flooding and bulk oil flow that contribute to SSVhash behavior.There were two solutions determined from test data that

eliminated the SSV hash signature:

• The first required flooding the bearing cavity, using labyrinth orfloating seals and increased oil flow. The solution was effective fordirect lube designs as well, but resulted in higher power loss, flowrequirements, and pad temperatures comparable to a conventional,flooded design.

• The second was a modification consisting of patented SSVgrooves cut in the babbitt near the edges of the pad, to capture andredirect side leakage toward the leading edge of the next pad. Thissolution was successful in eliminating SSV hash in an evacuatedconfiguration. Low oil flow and power loss were maintained, witha slight penalty in pad temperature due to the introduction of warm,side leakage oil back into the oil film.

• In both cases, the elimination of SSV hash increased synchronousamplitudes.

The following conclusions are derived from the theoreticalinvestigation of SSV hash:

• A tilting pad bearing’s full coefficients can be used to assess thedynamics of individual pads.

• Using the full coefficients of one of the test bearings, theoreticalinvestigations suggest that the observed SSV hash is likely a forcedvibration phenomenon.

• Theoretical predictions indicate that a pad’s responsiveness atlow frequencies increases when there is insufficient oil to providefor a full film.

• At lower supply flowrates, predicted partial starvation at theleading edge progresses, which reduces the pad’s tilting stiffness,making it more responsive to excitation.

DISCUSSIONS

Many questions arose over the course of the tests and analysesand during review of the initial drafts of this paper. The followingdiscussions comment on topics that are not necessarily derived

directly from the test results and analyses. As such, the discussionsare subject to debate, which is welcome.

Possible Source of Excitation

Test results and analyses indicate that the shaft SSV hashindications are caused by pad vibration with response characteristicsindicative of a forced vibration. Since there were no means in theSSV hash test apparatus to visualize the flow in the bearing,possible sources of excitation can only be surmised from the testresults presented in this paper. Instead, attention is directed toHeshmat (1991), who did observe a periodic phenomenon in hisvisualization experiments of oil streamlets in an axial groovejournal bearing. The phenomenon is reported as a “pulse” thatperiodically transformed the starved region of the leading edgefilm into a pattern of oil streamlets. This pulse occurred every0.5 to 1.0 seconds and its frequency varied with the degreeof starvation.This phenomenon is consistent with many of the observations

from these SSV hash investigations. Test data and analyses bothshow an increased sensitivity to SSV hash at reduced flowrates.A periodic pulse excitation can explain dominant SSV hashamplitudes at intermediate operating conditions, i.e., peaks noticeablein Figures 17 through 22, as well as difference in SSV hash signaturesamong test bearings, which may be attributable to the influence ofthe different supply methods on the excitation.

Pad Vibration and Damage

Pad flutter, babbitt fatigue, and pivot fretting are often broughtup as topics of concern regarding SSV hash. The babbitt damagestudied by Adams and Payandeh (1982) was a major concern inlarge, conventional, flooded bearings, attributed to self-excited,subsynchronous vibrations of unloaded pads with frequency ratiosapproximately 0.50 that of running speed. The term pad flutter isoften used to describe this motion, and spragging is often usedto describe more violent, full clearance vibrations with forcessufficient to fatigue the babbitt and fret the pivots.The pad vibrations associated with SSV hash do not conform to

these characteristics. Subsynchronous amplitudes are at least anorder of magnitude less than the bearing clearance, and frequencyratios are more on the order of 0.10 on average. Babbitt fatigue isnot considered a concern in regard to the SSV hash characteristicsdescribed in this paper.Fretting is a more difficult phenomenon to assess analytically.

Vibrations certainly contribute to fretting, but there are manysources of vibration in turbomachinery. The only information thatcan be offered from the SSV hash test series in this respect is thatthere were no indications of fretting, or babbitt fatigue, in individualtest pads after 600 hours of operation.

Bearing Clearance, Preload, Pivot Offset

It would seem that preload, via reduced bearing clearance, wouldreduce shaft SSV hash and that increasing the pivot offset shouldmake it worse. It was therefore unexpected when center versusoffset pivot tests showed lower levels for offset pivot operation.Hindsight from the test results and analyses indicate other

factors are involved, e.g., proximity of the individual pads to theshaft, the magnitude of their hydrodynamic forces, sensitivity toexcitation, etc., which make it difficult to generalize cause andeffect. Although preload and clearance tests were not complete atthe time of this paper, scenarios where preload might invoke padvibration and shaft SSV hash can be envisioned when consideringthe other influences.With multiple factors and complex pad/shaft interactions, cause

and effect scenarios for clearance, preload, offset, and other geometryshould be performed by analysis rather than generalizations. Theanalyses presented in the theory section of the paper are usefultools for assessing new designs, as well as evaluating changes toexisting designs.


SSV Hash: Issue, Nonissue, Allowable Levels

The many types of turbomachinery and the wide array of sizes andoperating conditions suggest caution against generalizations. Thereare applications running with SSV hash with no reported problems,and DeCamillo and Clayton (1997) and Edney, et al. (1996), providerotordynamic data showing some level of flow reduction is possiblewithout affecting rotor response. On the other hand, Edney, et al.(1996), do report problems when flow is reduced too much, which islogical and also supported by the theoretical analyses.Personal experience with SSV hash for turbomachinery bearing

sizes 3.88 to 8.00 inches (100 to 200 mm) in diameter generallyfalls within API guidelines. In comparisons with API 617 (2002)nonsynchronous limits, for example, levels are typically less then0.2 mils (.0050 mm) peak-to-peak. Actually, SSV hash more oftenis a consideration in overall vibration limitations, as nonsynchronousfrequencies are typically below 0.25 times the maximum continuousspeed and are indiscrete. Noted earlier, elimination of SSV hashtended to increase synchronous amplitudes in tests. This is broughtup neither as an issue or nonissue, but to provide information that

overall vibration levels may not decrease as much as anticipatedwhen SSV hash is eliminated.Issues more often arise when other specifications further limit

allowable SSV indications or, as stated earlier, because ofuncertainty regarding the cause and nature of the vibration. In thiscase, solutions and test results in this paper can be used to addressthe situation. A more challenging issue is complying with multiplespec limitations in more severe applications. It seems that many ofthe parameters that reduce SSV hash tend to increase power loss,flow requirement, and pad temperature, or produce undesirable rotorresponse. Flooding and pressurized bearing cavities, rotordynamicpreference for center pivot load-on-pad configurations, and tighterclearances are some examples.There is certainly the need for more research and development,

experimental and theoretical, on the subject of SSV hash includingits effects on dynamic coefficients, rotor response, and stability.In the meantime, the authors hope that the tests and analyticalinvestigations presented in this paper will provide a usefulreference for topics related to SSV hash.


APPENDIX A—

Figure A-1. Direct Lube Leading-Edge-Groove, LOP, 200 psi (1.38 MPa), No SSV Grooves.

REFERENCES

Adams, M. L. and Payandeh, S., 1983, “Self-Excited Vibration ofStatically Unloaded Pads in Tilting-Pad Journal Bearings,”ASME Journal of Lubrication Technology, 105, pp. 377-384.

API 617, 2002, “Axial and Centrifugal Compressors andExpander-Compressors for Petroleum, Chemical and GasIndustry,” Seventh Edition, American Petroleum Institute,Washington, D.C.

Booser, E. R., 1990, “Parasitic Power Losses in Turbine Bearings,”STLE Tribology Transactions, 33, pp. 157-162.

Brockwell, K., Dmochowski, W., and DeCamillo, S., 1992,“Performance Evaluation of the LEG Tilting Pad JournalBearing,” IMechE Seminar Plain Bearings—Plain Bearings—Energy Efficiency and Design, MEP, London, UnitedKingdom, pp. 51-58.

Cloud, C. H., 2007, “Stability of Rotors Supported on Tilting PadJournal Bearings,” Ph.D. Dissertation, University of Virginia,Charlottesville, Virginia.

DeCamillo, S. and Clayton, P. J., 1997, “Performance Tests of an18-Inch Diameter, Leading Edge Groove Pivoted Shoe JournalBearing,” Proceedings of the 2nd International Conference onHydrodynamic Bearing—Rotor System Dynamics, Xi’an,China, pp. 409-413.

DeCamillo, S., 2006, “Current Issues Regarding UnusualConditions in High-Speed Turbomachinery,” Keynote

Presentation, 5th EDF & LMS Poitiers Workshop BearingBehavior Under Unusual Operating Conditions Proceedings,pp A.1-A.10.

Edney, S. L., Waite, J. K., and DeCamillo, S. M., 1996, “ProfiledLeading Edge Groove Tilting Pad Journal Bearing forLight Load Operation,” Proceedings of the Twenty-FifthTurbomachinery Symposium, Turbomachinery Laboratory,Texas A&M University, College Station, Texas, pp. 1-16.

Fillon, M., Bligoud, J. C., and Frene, J., 1993, “Influence of theLubricant Feeding Method on the ThermohydrodynamicCharacteristics of Tilting Pad Journal Bearings,” Proceedingsof the 6th International Congress on Tribology, Budapest,Hungary, 4, pp. 7-10.

Harangozo, A. V., Stolarski, T. A., and Gozdawa, R. J., 1991, “TheEffect of Different Lubrication Methods on the Performance ofa Tilting Pad Journal Bearing,” STLE Tribology Transactions,34, pp. 529-536.


He, M., Cloud, C. H., and Byrne, J. M., 2005, “Fundamentals ofFluid Film Journal Bearing Operation and Modeling,”Proceedings of the Thirty-Fourth Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 155-175.


Figure A-2. Direct Lube Leading-Edge-Groove, LOP, 200 psi (1.38 MPa), with SSV Grooves.

Heshmat, H., 1991, “The Mechanism of Cavitation inHydrodynamic Lubrication,” STLETribology Transactions, 34,(2), pp. 177-186.

Kocur, J. A., Nicholas, J. C., and Lee, C. C., 2007, “Surveying TiltingPad Journal Bearing and Gas Labyrinth Seal Coefficients andTheir Effect on Rotor Stability,” Proceedings of the Thirty-SixthTurbomachinery Symposium, Turbomachinery Laboratory, TexasA&M University, College Station, Texas, pp. 1-10.

Parsell, J. K., Allaire, P. E., and Barrett, L. E., 1983, “FrequencyEffects in Tilting-Pad Journal Bearing Dynamic Coefficients,”ASLE Transactions, 26, (2), pp. 222-227.

Shapiro, W. and Colsher, R., 1977, “Dynamic Characteristics ofFluid-Film Bearings,” Proceedings of the Sixth TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas, pp. 39-53.

Tanaka, M., 1991, “Thermohydrodynamic Performance of a TiltingPad Journal Bearing with Spot Lubrication,” ASME Journal ofTribology, 113, pp. 615-619.

Wilkes, J. J., DeCamillo, S. M., Kuzdzal, M. J., and Mordell, J. D.,2000, “Evaluation of a High Speed, Light Load Phenomenonin Tilting-Pad Thrust Bearings,” Proceedings of the Twenty-NinthTurbomachinery Symposium, Turbomachinery Laboratory, TexasA&M University, College Station, Texas, pp. 177-185.

Wilkes, J. and DeCamillo, S., 2002, “Journal Bearing,” UnitedStates Patent No. 6,361,215 B1, Mar. 26, 2002.

ACKNOWLEDGEMENT

The authors would like to thank colleagues at Kingsbury, Inc., fortheir special efforts and help in acquiring and preparing data andinformation for this technical paper. The authors would also like toespecially acknowledge John Kocur of ExxonMobil, Brian Pettinatoof Elliott, and Thomas Soulas of Dresser-Rand, among others, fortheir help and expertise in discussions of the subject matter.


James F. McCraw is a Senior RotatingEquipment Engineer with BPAmerica, Inc.,Operation Efficiency Group, in Houston,Texas. In his 23 years with BP and Amoco,he has worked in machinery engineering,project support, plant maintenance, andmachinery operation. In his position he isresponsible for new equipment specifications,technology applications, research anddevelopment, and failure analysis for a

variety of centrifugal compressors, pumps, gas turbines, andgas engines.Mr. McCraw holds a BSEE from Finlay Engineering/University

of Missouri and is a member of the National Society ofProfessional Engineers.

Vladimir Bakalchuk is Manager, NewProducts, at John Crane Inc., in MortonGrove, Illinois. With more than 25 years ofexperience in various fields of the oil andgas industry, he has spent the betterpart of 15 years specializing in dry gasseal/rotating equipment. Mr. Bakalchukpredicated his career in various technicalcapacities in the Ukrainian Academy ofSciences, the Mechanical Engineering

Department of the University of Calgary, Brown-Root-Braun, NovaGas Transmission, Canadian Fracmaster, and Revolve Technologies.Mr. Bakalchuk holds a degree (Mechanical Engineering) from

L’viv Polytechnical University, passed Ph.D. candidacy at theUniversity of Calgary, and is a registered Professional Engineer inthe Province of Alberta.

Richard Hosanna is Manager of T28 GasSeal Engineering with John Crane U.S.A.,in Morton Grove, Illinois. He joined JohnCrane in 1983 and has held variouspositions within the applications anddesign engineering group, including 20years with dry gas seals.Mr. Hosanna has a B.S. degree (Industrial

Technology) and is the holder of three U.S.Patents associated with the sealing industry.

ABSTRACT

A compressor in a natural gas gathering service experiencedmultiple seal failures on the discharge end. Synthetic oil mistcontained in the sealing gas was identified as a significantsource of contamination that caused the seal failures. Sincesimilar dry gas seals in other compressors at the samelocation were operating satisfactorily, a study of a particularcompressor was undertaken. The seals were instrumented withthermocouples to monitor the temperature distribution acrossthe seals.This paper discusses the findings of the study, resulting

modifications of the dry gas seals, seal controls, and thecompressor itself. This paper also outlines techniques that havebeen developed to mitigate the situation without having to shutthe unit down. Also presented is a design approach to thesealing and seal monitoring in processes where liquid ingressinto the sealing area may occur, including seal materialselection, control system philosophy, seal head, and bearingcavity design.

DESIGNAND MITIGATION TECHNIQUES FORAPPLICATIONSWITHPOSSIBLE LIQUID CONTAMINATION OF THE SEALING GAS DRY GAS SEAL SYSTEM

byJames F. McCraw

Senior Consultant, Rotating Equipment

BPAmerica, Inc.

Houston, Texas

Vladimir BakalchukManager, Type 28 New Products

andRichard Hosanna

Type 28 Engineering Manager

John Crane Inc.

Morton Grove, Illinois

37

INTRODUCTION

An oil and gas production company purchased a centrifugalcompressor to increase the gas sales volume. Their chosenconfiguration consisted of an electrical motor driven centrifugalbooster, and special attention was given to the design of thecompressor’s dry gas seal system (Figure 1). The system controlphilosophy was based on remote diagnostics of all critical andnoncritical seal system operating parameters. The system had alsobeen designed to prevent process gas release into the atmosphere incase of a catastrophic seal failure.

Figure 1. Facility Flow Diagram.

SYSTEM DESIGN

The control system design (refer to APPENDIXA) incorporatedtwo stages of filtration: coalescing and particulate. The particulatefilters (AJ-100A/B) were placed downstream of coalescing(AJ117A/B) to prevent seal contamination by fibrous debris incase of a coalescing element rupture. The sealing gas supply(PDV-120) was based on the differential pressure control principle(PDY-120), with balance piston cavity pressure used as a reference.Since the pressure control principle has been utilized for primaryseal supply and not flow control, each compressor end seal supplyflow was monitored by flow transmitters (FIT-120/125) toguarantee adequate supply volume.Traditionally dry gas seal performance is evaluated based on its

leakage. Since in this application primary vent flow consisted ofroughly the sum of primary seal leakage and secondary seal supplyflows, in order to enable efficient leakage monitoring and preventprimary seal leakage dilution, the secondary seal supply employedflow control (FY-160/166). The primary and secondary leakage flows(FIT-150/151/155/156) and corresponding leakage cavity pressures(PIT-150/151/155/156) were monitored to provide indication of sealhealth. All critical and noncritical operating parameters weremonitored and controlled by a discrete control system.The selected dry gas seal design incorporated a hard silicon

carbide (SiC) rotating face running against a soft carbon stationaryface. The interfaces between the seal and compressor head and theshaft were sealed with fluorocarbon O-rings, while spring energizedpolymers served as the seal’s internal secondary sealing elements. Acircumferential carbon ring pressure bushing was used to separatethe dry gas seal from the bearing cavity. The seals were installed intoa separate seal head, which in turn, was installed into the compressorbody. The supply and leakage porting was cross-drilled and pipingconnections were located on a circle terminating at flanges.Special attention was given to prevention of uncontrolled gas

release into the atmosphere in case of a seal failure. It wasdetermined that, in case of a catastrophic primary seal failure, inorder to depressurize primary seal leakage cavity and redirect gasflow from the primary seal leakage chamber into the flare system,an additional primary seal venting area was required. Toaccommodate the requirement, two vent ports were incorporatedinto the primary seal leakage annulus. Also, to increase the ventingarea, control system secondary seal supply piping was transformedinto vent piping by two sets of pneumatically actuated ball valves.In case of a trip on high high flow, one set of valves (SDV-160/166)would close, isolating the supply system upstream, and the otherset (BDV-161/167) would open, connecting the secondary seal gassupply and, thus, seal the primary leakage cavity to the ventsystem. The secondary seal cavity depressurization scheme was

developed based on similar calculations. By design, the system incase of a catastrophic seal failure of a primary and/or secondaryseal would prevent uncontrolled gas release into atmosphere.

OPERATING BACKGROUND

The compressor was successfully commissioned and started.After a month of operation, a seal failure occurred. The compressor’sdrive end seal failed catastrophically. Debris originating from theseal’s SiC rotating face damaged the seal head cavity during thefailure and scored the lands during seal removal (Figure 2). Prior tothe failure, operating personnel had noticed a decrease in barrierseal supply flow. At factory disassembly, the nondrive-end sealexhibited signs of contamination by both liquid and particulate.Upon disassembly of the barrier seal, it was noticed that sealO-rings were thermally hardened. The degree to which the O-ringshardened was depending on O-ring location: the farther thelocation was from the bearing cavity, the lesser was the degree ofO-ring hardening. Although this fact had been noticed, no relatedaction was taken at that time.

Figure 2. Head Damage at the O-Ring Location Resulting fromRotating SiC Ring Catastrophic Failure.

The compressor head was repaired, a new seal installed, and theunit was restarted. In time, the flow to the barrier seal on the driveend progressively declined and eventually was reduced to a fewliters per minute. At a later date, the unit was shut down to evaluatethe barrier seal condition. The pattern of O-ring hardeningpersisted. During the shutdown, a small volume of liquid wasfound in the seal gas particulate filters located downstream of thecoalescing ones (Figure 3).

Figure 3. Liquid Accumulation in Particulate Filter Housing.

An analysis identified the liquid as a mixture of water andtriethylene glycol (TEG). The gas used for the sealing supply wasprovided to the coalescor at 100�F. To filter the mist contained in thegas, it would have been necessary to increase the gas temperatureby incorporation of a heater into the stream. But the panel valvetemperature rating limited the supply gas temperature increase andprevented additional gas heating. Various modifications to theexisting coalescing filter elements were undertaken. The modificationseliminated liquid accumulations in the filter bowl itself, but liquidcontinued to exit the piping drain valve downstream of the filter.The existing coalescor was replaced by a high efficiency coalescingfilter element. The liquid coalesced in the new filter chamber wasidentified as a mixture of polyalkylene glycol (PAG), TEG, andwater. The PAG was used as lubrication/sealing oil in upstream screwcompressors. The second seal failure occurred in a manner similar tothe first one a year later (Figure 4).


Figure 4. Catastrophic Seal Failure—SiC Fragments that Damagethe Seal Head.

It is widely accepted that in order to maintain reliable operation,a dry gas seal requires clean and liquid-free gas. In reality, a sealis tolerant, to a certain degree, of liquid contamination byvaporizable liquids, while reacting poorly to nonvaporizable liquids.Incompressible fluid ingress onto the seal faces produces twodetrimental effects: sealing film instability and heat generation. Insteady-state compressor operation, if pressure differential across thecompressor does not suddenly change, the rotor usually does notexperience rapid axial movements (Figure 5). Thus, if a nonvaporizable liquid ingress into the seal has occurred at a steady-statecompressor operation, heat generation in liquid shear becomes moreof a concern than sealing film instability. Generated heat, if notdissipated, leads to a thermal expansion of the rotating parts of theseal and generation of high stresses. The typical mode of such afailure would be a fracture of a rotating sealing face followed by itsdisintegration due to consequential impacts in rotation. Interestinglyenough, the same site had other compressors with dry gas seals thathave been successfully operating for years on the same sealing gas,supplied from a common header. Nevertheless, the dry gas seals inthese units appeared to be tolerant of the sealing gas composition,operating up to and in excess of five years.

Figure 5. Mixture of PAG and TEG in the Seal Supply Annulus.

MODIFICATIONS

The issue of the compressor seal reliability can be approachedfrom two directions: elimination of the liquid in the seal gas supplyor creation of an operating environment for the seals that ensuresthat the liquid ingress impact is minimized. The first approach isalways preferred because it eliminates the cause of the seal failures.Unfortunately, in many situations, such a source of clean gas doesnot exist and its creation is economically prohibitive. Since theother compressors at the site were operating satisfactorily and noother source of sealing gas existed at the site, a decision was madeto study and modify the dry gas seal system. The goal was toreduce to a minimum concentration of the nonvaporizable liquid inthe seal supply gas and to eliminate all additional sources of heatgeneration in and around the seal.

Dry Gas Seal Supply Gas Conditioning

A number of proprietary filter elements were tested to determinean optimum choice. While testing one, a two-phase liquid sample

was rejected by a coalescor that consisted of PAG and TEG/water.PAG, viscous synthetic oil, is used in flooded screw compressors inthe facility for sealing and bearing lubrication. The increase influid viscosity adversely affects the coalescing ability of the filters.A filtration element vendor designed a special high-efficiencyelement (99.9998 percent at 0.1 micron aerosol) to enablereduction in the synthetic oil carryover. The element utilized theliquid’s velocity and temperature (viscosity) to maximize itscoalescing ability. The second filter in a duplex arrangement wasrepiped to operate in series with the first one (Figure 6). As a result,a maximum concentration of PAG and TEG equated to 40 ppmwand water concentration to 160 ppmw.

Figure 6. Prefilter Arrangement.

Compressor Seal Cavity

Investigation showed that the design of a cavity between thebalance piston and dry gas seal provided a potential for liquidaccumulation. The same applied to the primary seal supply andleakage annuluses, as the correspondent ports were located in theupper quadrants. In case liquids were introduced into the primaryseal, the accumulations would occur in the primary seal supply andleakage cavities annuluses (Figure 7). To prevent the liquidaccumulation, the cavities were outfitted with the drains (Figure 8).

Figure 7. Seal and Seal Head Prior to Modifications.

Figure 8. Seal and Seal Head after Modifications.

DESIGNAND MITIGATION TECHNIQUES FORAPPLICATIONSWITHPOSSIBLE LIQUID CONTAMINATION OF THE SEALING GAS DRY GAS SEAL SYSTEM 39

Dry Gas Seal Supply Path

Part of the study was to determine what differed with regard tothe compressors operating successfully at the facility from the oneexperiencing seal failures. When a seal gas supply path wasexamined, it was noted that the sealing gas was delivered immediatelyto the sealing faces in the problem unit, while in the other units, itwas first directed onto a solid surface acting as a knock-out plateand than injected between the seal and a process side labyrinth, thusallowing for oil knock-out with further drainage away from thefaces. To modify the existing design, an additional process sidelabyrinth was introduced. Now, the sealing gas entering the cavityis injected on the outer diameter (OD) surface of the new processlabyrinth, which acts as a knock-out (Figure 8). Separated liquid isdrained back to suction through an introduced drain.

Dry Gas Seals

In both seal failures, the compressor seal cavity was damaged byshattered SiC rotating faces. To avoid the lengthy cavity repairprocess resulting from damage by SiC bits, the seals were retrofittedwith ductile (stainless steel with tungsten carbide [WC] coating)rotating faces. Since heat generation was the reason for both sealfailures, the drive-end (DE) dry gas seal and barrier seal wereretrofitted with temperature monitoring devices.

Compressor Bearing Cavity

The issue with oil being found in the secondary vent line startedshortly after the initial startup. The barrier seal’s O-rings were foundto be hardened. To eliminate all possible sources of heat generation,the barrier seal was redesigned to reduce the windage and eliminatethe whipping action of protruding screw heads. A swirl brake wasincorporated into the bearing cavity to facilitate oil drainage.

OBSERVATIONS

After the modifications were completed, thermocouple readingswere taken twice a day and compared to the drain’s content invarious locations. The temperature measurements establishedthat the primary seal temperature was not subject to a high rise.The temperature increase in the secondary seal was the highestand started occurring shortly after the startup. The barrier sealtemperature closely followed the temperature of the secondaryseal, though the rate of increase was not as high.Since the liquid carryover was minute, the correlation between

temperature rise and liquid presence in the filter drains and otherdrains proved to be complicated. Either there was no directcorrelation or it was too difficult to discern the rise in the secondaryseal temperature and the misting of the drains. Half a year after thestartup, the secondary seal temperature was approaching 300�Fwith the barrier seal temperature trending high (Figure 9).

Figure 9. Seal Temperature Trends Prior to Mitigation.

ANALYSIS

The thermocouple readings validated that the secondary sealappears to be the dominant source of heat generation. The liquidingress into the seal cavity was minimized, but some liquidcontinued to be accumulating on the secondary seal faces. Bydesign, in order to conserve nitrogen and allow for an accuratereading of actual seal leakage, the intermediate labyrinth supplywas flow controlled to 1 acfm (velocity of 5 to 7 fps).The original design philosophy was based on an assumption that

no liquids pass through the primary seal and, therefore, gasvelocity of 5 to 7 fps would be sufficient to sweep clean leakagegas into the primary vent. But, this velocity was insufficient toprevent ingress of liquid trapped in the leakage gas into thesecondary seal. Due to the fact that the primary leakage cavitypressure is small (around 1 to 2 psig), the secondary seal leakageis extremely low—a couple of Nl/min. It is not sufficient to providecooling and its flow is not sufficient to remove oil film off thesecondary seal faces. In comparison, the primary seal was beingcooled by more than 100 scfm of cool sealing gas and wasoperating at over 200 psig pressure. As long as the secondary sealoperated at a low pressure with a small leakage, the abovecondition existed and liquids were not removed from the secondaryseal faces, the amount of heat generated in this condition could notbe completely dissipated. Essentially, the compressor’s drive endseal head acted as a heat trap.

MITIGATION

There are very few ways of bringing heat generation undercontrol in such a situation. The most easily attainable solutionsinclude completely stopping the oil mist from depositing on theseal faces or to achieve an N2 flow to the intermediate labyrinth(secondary seal supply) that would be sufficient to remove the heatgenerated by the secondary seal faces. Such flow should removethe heat and prevent the sealing face fracture and/or create enoughleakage flow through the faces to generate a velocity sufficient toremove a majority of the oil film of the faces or separate the faces.To achieve a solution to the failure problem in this application,

the flow to the barrier seal was increased in steps. Temperaturemeasurements indicated no impact on the barrier seal cooling onthe secondary seal temperature, confirming that the secondary sealwas the source of the heat generation. Next, the flow to theintermediate labyrinth (secondary seal supply) was increased insteps, allowing time for temperatures to settle. After a flow of 6scfm was achieved as a result of these stepped changes to thesecondary seal supply, the secondary seal temperature dropped andremained stable at the 210 to 230�F level (Figure 10).

Figure 10. Seal Temperature Trends after Mitigation.


CONCLUSIONS

Although providing clean and liquid free seal supply gasremains a preferred option, many lessons could be learned from thestudy above. In order to enable a reliable operation of a centrifugalcompressor in processes with a possibility of liquid ingression intothe seal, a dry gas seal system and compressor design should incor-porate the following:

• Dry gas seals should be outfitted with temperature monitoringinstruments.

• Dry gas seal design, operating conditions permissive, shouldutilize ductile coated mating rings to prevent cavity damage in caseof the seal failure.

Sealing gas should be diverted from direct entry into the seal. Aknock-out type entry should be provided in the seal gas supply cavity.

• All seal cavities and stagnant flow cavities on the process sideof the seals should be equipped with drains located at thesix o’clock position.

• Control systems should be designed to accommodate possiblefuture increases in supply flow to both primary and secondary seals.

• Bearing cavity design should provide for a proper drainage oflubricating oil preventing pressure build up inside the cavity anduntrained oil accumulation.


41

•

APPENDIX A—

DRY GAS SEAL CONTROL PANEL P&ID

Figure A-1. Primary Seal Supply System.


Figure A-2. Secondary and Barrier Seal Supply System.

Figure A-3. Leakage Monitoring System.


43

Masayuki Kita presently is the Managerof the Compressor Designing Sectionwithin the Turbo Machinery EngineeringDepartment, Mitsubishi Heavy Industries,LTD., in Hiroshima, Japan. He is engagedin the design and development of singleshaft and integrally geared processcentrifugal compressors and expander-compressors for use in petroleum,chemical, and gas industry services that

handle air and gas in accordance with API Standard 617.Mr. Kita received B.S. and M.S. degrees (Mechanical

Engineering) from Shizuoka University.

Shinji Iwamoto presently is an ActingManager of the Compressor DesigningSection within the Turbo MachineryEngineering Department, MitsubishiHeavy Industries, LTD., in Hiroshima,Japan. He is engaged in the design anddevelopment of centrifugal compressors foruse in petroleum, chemical, and gasindustry services that handle air and gas inaccordance with API Standard 617.

Mr. Iwamoto received B.S. and M.S. degrees (MechanicalEngineering) from Kyushu University.

Rinpei Kawashita is a Research Engineerin the Vibration & Noise Control Laboratoryin Takasago Research & DevelopmentCenter, Mitsubishi Heavy Industries, LTD.,in Takasago, Japan. He has been a specialistof rotordynamics and engaged in researchand development of steam turbines, gasturbines, centrifugal compressors, and otherrotating machinery for four years.Mr. Kawashita received B.S. and M.S.

degrees (Mechanical Engineering) from Kyushu University.

ABSTRACT

The prediction of rotating stalls is one of the most important keytechnologies for compressors, especially high-pressure compressors.The operating ranges of compressors are restricted by rotatingstalls because they may cause severe subsynchronous vibration ofthe rotor. As a consequence, there are many papers that havereported on the subject, including the influence of vaneless diffusergeometry on rotating stalls, and predictions of when a rotating stalloccurs through experiment and theory. But, so far, there is no wayto predict accurately the amplitude of subsynchronous rotorvibration caused by a rotating stall.In this paper, the pressure fluctuation caused by rotating stall and

the resulting vibration of the rotor were measured using a test rig.

97

PREDICTION OF SUBSYNCHRONOUSROTORVIBRATIONAMPLITUDE CAUSED BY ROTATING STALL

byMasayuki Kita

Manager, Compressor Designing Section

Hiroshima Turbomachinery Engineering Department

Shinji IwamotoActing Manager, Compressor Designing Section

Hiroshima Turbomachinery Engineering Department

Daisuke KiuchiHiroshima Turbomachinery Engineering Department

andRinpei Kawashita

Research Engineer, Vibration & Noise Control Laboratory

Takasago Research & Development Center


Takasago, Japan

The external force on the impeller during a rotating stall was estimatedfrom the measured pressure fluctuation. The subsynchronous vibrationamplitude was then calculated using rotordynamics analysis by addingthis external force.Comparing the result of this rotordynamics analysis with the

measured subsynchronous vibration, an influence factor of pressurefluctuation to rotor vibration, �, was calculated. The accuracy ofthis factor � was checked by comparison with the result of a shoptest on a compressor. Using this factor, the subsynchronous rotorvibration level caused by a rotating stall can then be predicted at thedesign stage.

INTRODUCTION

Many kinds of studies on rotating stalls have been carried outbecause this phenomenon not only adversely affects the performanceof the compressor, but also causes subsynchronous vibration of thecompressor rotor, especially with high-pressure services.The compressor operating range may be restricted by a rotating

stall, because it usually occurs prior to a surge. It is usuallyinevitable that a rotating stall occurs prior to a surge for all designs.But what must be avoided is severe rotor vibration due to a rotatingstall, not the rotating stall phenomenon itself.The purpose of this paper is to establish a method of predicting

the subsynchronous rotor vibration level caused by a rotating stallat the design stage of centrifugal compressors.

TEST BY SINGLE IMPELLER TEST RIG

Test Rig Arrangement

The tests were performed in a test rig, the rotor of which has oneimpeller and was supported by the two ball bearings as shown inFigure 1. The selected design flow coefficient of the impeller is0.02, because most rotating stall problems occur with low flowcoefficient impellers. The test rotor was driven by a motor througha speed increasing gear.

Figure 1. Test Rig Arrangement.

The two vibration probes (horizontal and vertical orientation)were installed as shown in Figure 1. Six pressure measurement

probes (P1 through P6) were installed to measure pressurefluctuation at the vaneless diffuser.The tests were performed by three parameters, which were

considered to influence the severity of rotating stall and the flowpoint at which a rotating stall occurs. The first parameter is diffuserwidth. From many reports that examined influences of the diffuserwidth on a rotating stall, the reduction in the diffuser width has astrong influence on the point at which a rotating stall occurs. Theratio of the diffuser width to the impeller outlet width was changedfrom 0.83 to 0.4.The second parameter is the clearance of the labyrinth behind

the impeller. A rotating stall occurs when the direction of flowreverses locally at the diffuser due to reduction of flow. If theclearance of the labyrinth behind the impeller increases, the flow atthe diffuser increases with increased flow. It is thought that thismakes the point at which a rotating stall occurs a lower flow level.The clearance of the labyrinth behind the impeller was changedfrom 0 to .039 inches (0 to 1.0 mm) diameter.The third parameter is the machine Mach number. In general, it is

known that if the machine Mach number increases, the pressurefluctuation caused by a rotating stall becomes larger, as does the pointat which the rotating stall occurs. The machine Mach number waschanged from 0.3 to 0.7. Test conditions are summarized in Table 1.

Table 1. Summary of Test Conditions.

Test Results

The six test cases were carried out as shown in Table 1. Figures2 and 3 show the pressure fluctuation and vibration spectra duringthe rotating stall in Case 1. By comparing these figures, it wasfound that there was a correlation between the pressure fluctuationand the rotor vibration at around 15 Hz.

Figure 2. Spectra of Pressure Fluctuation.

Figure 3. Spectra of Rotor Vibration.


Figure 4 shows the relationship between the pressure fluctuationcaused by a rotating stall and the flow coefficient for all cases.Figure 5 shows the relationship between the subsynchronousvibration amplitude and the flow coefficient for all cases. Fromthese results, the followings are found:

Figure 4. Relationship Between Pressure Fluctuation by RotatingStall and Flow Coefficient.

Figure 5. Relationship Between Subsynchronous Vibration Amplitudeby Rotating Stall and Flow Coefficient.

• As the flow reduces, the pressure fluctuation increases. However,as it reaches a certain level it does not increase any further.

• As the flow reduces, the number of cells during a rotating stallis increased from two to three. For many cases, as the number ofcells increases, the pressure fluctuation and subsynchronousvibration amplitude decreases.

• Figures 4 and 5 show similar patterns of behavior, but in Figure5, the effects of changing the test parameters are larger.

Figure 6 shows the ratio of the frequency of a rotating stall,f, to the rotational frequency of the rotor, N. From this figure,it is found that the frequency of a rotating stall increases asthe flow decreases. Furthermore as the number of cellsincreases from two to three, the frequency of a rotating stallalso increases.

Figure 6. Ratio of Frequency of Rotating Stall to RotationalFrequency of Rotor.

Figure 7 shows the relationship of pressure fluctuationmeasured by pressure measurement probes P1 and P2 shown inFigure 1. The amplitude of the pressure fluctuation is larger atP1, because the rotating stall occurs near the impeller exit.However, by reducing the flow, the ratio of P2/P1 increases. Thismeans that the region of the pressure fluctuation cell spreadsradially as flow reduces.

Figure 7. Relationship of Pressure Fluctuation Measured by PressureMeasurement Probe P1 and P2.

PREDICTION OF VIBRATION LEVEL

Prediction by Pressure Fluctuation Level

Assuming the pressure fluctuation magnitude at the diffuser isknown, to establish the effect of a rotating stall on subsynchronousrotor vibration level, it should be examined how much the externalforce from pressure fluctuation affects the rotor vibration. If theinfluence factor of pressure fluctuation as the external force to therotor vibration is found, it will be included in the rotordynamicsanalysis as an external force, and the subsynchronous vibrationamplitude caused by the external force can be estimated.Figure 8 shows the relationship between the pressure fluctuation

and the subsynchronous vibration measured with the test rig for Case1. This figure shows that the vibration amplitude is proportional tothe pressure fluctuation. Furthermore, the influence factor ofpressure fluctuation to the vibration, � defined by the followingequation, can be derived:

where,F = External forceS = Impeller projection areaPAC = Pressure fluctuation at diffuser inlet (impeller outlet)Pdout = Dynamic pressure at diffuser inlet (impeller outlet)

Figure 8. Relationship Between Pressure Fluctuation andSubsynchronous Vibration (Case 1).

Figure 9 summarizes this factor � for all cases. The factor varieswith the test conditions, but the maximum value is approximately 0.6.

Figure 9. Summary of Factor � for all Cases.

99PREDICTION OF SUBSYNCHRONOUSROTORVIBRATIONAMPLITUDE CAUSED BY ROTATING STALL

Prediction by Dynamic Pressure Difference

In order to know the pressure fluctuation level during a rotatingstall, a computational fluid dynamics (CFD) analysis or experimentalstudy is necessary. However, this would require enormous time andcost. Therefore a more efficient way to predict the subsynchronousvibration level was proposed. As an alternative of �, the externalforce was derived from the difference in the dynamic pressuresbetween the impeller inlet and outlet at the design stage.Figure 10 shows the relationship between the difference of the

dynamic pressures and the subsynchronous vibration measuredwith the test rig for Case 1. Applying the same concept as �, byintroducing � defined by the following equation, the influence ofthe pressure fluctuation on the vibration can be estimated.

where,�P = Difference of dynamic pressures between inlet and

outlet of impeller

Figure 10. Relationship Between Difference of Dynamic Pressureand Subsynchronous Vibration (Case 1).

Figure 11 summarizes this factor � for all cases.Although the factorvaries with test conditions, the maximum value is approximately 0.06.

Figure 11. Summary of factor � for all Cases.

COMPARISON WITH SHOP TESTRESULT OF AN ACTUAL MACHINE

In order to confirm the validity of factor �, the pressure fluctuationand subsynchronous vibration were measured using an actualcompressor. By the same procedure, the influence factor of � wasderived. The result was compared with the factor � derived fromthe single impeller test results.

Test Arrangement

The shop tests were performed using a back-to-back typecompressor, which has four impellers in the low-pressure sectionand three impellers in the high-pressure section. The impeller flowcoefficients were approximately 0.008 to 0.024. The test wascarried out under the condition of partial-load around 350 kW.During the test, rotor vibrations were measured using two vibrationprobes for each end (drive and nondrive end). At the same time, thepressure fluctuations were measured at the exit of the dischargescrolls for each section. The test arrangement is shown in Figure 12.

Figure 12. Block Diagram for Measurement.

Test Results

Figure 13 shows the performance curve for the low-pressuresection of the compressor under test conditions. During the test, arotating stall was observed at the measured test points #341 and #351.

Figure 13. Performance Curve of Test Condition.

Figures 14 and 15 show 3D spectra diagrams of the pressurefluctuation and the rotor vibration measured at test point #351. Atthis point, pressure fluctuations were observed at frequencies around11.7 Hz and 23.4 Hz. These frequencies were considered to be causedby a rotating stall. At the same time as this pressure fluctuation, thevibration amplitude of the same frequencies was increased.

Figure 14. 3D Spectra of Rotor Vibration and Pressure Fluctuationfor Low-Pressure Section.


Figure 15. 3D Spectra of Rotor Vibration and Pressure Fluctuationfor High-Pressure Section.

Figures 16 and 17 show coherence between pressure fluctuationand rotor vibration. From these figures, there was a clear correlationbetween pressure fluctuation and rotor vibration at frequencies of11.7 Hz and 23.4 Hz in addition to synchronous frequency.

Figure 16. Coherence Between Pressure Fluctuation and RotorVibration (Vibration-1).

Figure 17. Coherence Between Pressure Fluctuation and RotorVibration (Vibration-2).

In particular, at the frequency of 11.7 Hz, there was a strongcorrelation between rotor vibration-1 and pressure fluctuation-1.And at the frequency of 23.4 Hz, there was a strong correlationbetween rotor vibration-2 and pressure fluctuation-2. Fromthese results, it was concluded that the frequency of a rotatingstall for the low-pressure section was 11.7 Hz, and thefrequency of a rotating stall for the high-pressure sectionwas 23.4 Hz.Figure 18 shows the waveform of the pressure fluctuation

level and the rotor vibration level. In order to compare each levelat the frequency of a rotating stall, a band pass filter for 11.7Hz was used. From these figures, one can see that the pressurefluctuation and the rotor vibration level tend to vary in thesame fashion.

Figure 18. Time Dependence Waveform of Pressure Fluctuationand Rotor Vibration.

VIBRATION ANALYSIS BY ROTORDYNAMICS

Figure 19 shows the model of the compressor rotor forrotordynamics analysis. For the analysis, the maximum,normal, and minimum bearing coefficients were used byconsidering the bearing lube oil temperature of 118.4�F ± 5�F(48�C ± 5�C) and manufacturing tolerance of bearingclearance (Cp).

Figure 19. Compressor Rotor Model.

The rotating stall was expected to occur at the last impeller stageof each section. Therefore, the unit external force from the rotatingstall was added on the impeller of Stage 4 or Stage 7. The resultsof the analysis are shown in Figure 20. From these results, thesubsynchronous vibration level from a unit force of a rotating stallwas estimated.

PREDICTION OF SUBSYNCHRONOUSROTORVIBRATIONAMPLITUDE CAUSED BY ROTATING STALL

101

Figure 20. Result of Rotordynamics Analysis.

The vibration amplitude in Figure 20 was calculated using theunit force. In this way, the expected subsynchronous vibrationamplitude could be calculated using the vibration amplitude at11.7 Hz or 23.4 Hz in Figure 20 multiplied by F. By comparingthe expected subsynchronous vibration amplitude with themeasured vibration amplitude at the shop test, the factor � couldbe obtained.Figure 21 shows the summary of the factor, �, obtained by the

above procedure. The resultant � was approximately 0.15.

Figure 21. Summary of Factor � for Each Bearing Coefficient atPoint #341 and #351.

There was a difference between � of 0.06 obtained from thesingle impeller test and � of 0.15 obtained from the shop test. Thereasons for this difference might be explained as follows:

1. Rotating stall may occur in the stages other than the last impellerand the induced external force may be larger than the assumptionin this study.

2. The estimated increment of dynamic pressure from inlet to outletof the last impellers was smaller than the actual one.

At the design stage, � should be considered as more than 0.15 inorder to judge on the safe side. As the proposed value, � of 0.20should be used. The accuracy of � should be improved by manyfuture shop tests.

CONCLUSION

The effect of the pressure fluctuation from a rotating stall on therotor subsynchronous vibration was investigated. The pressurefluctuation level and the subsynchronous level were measured witha single impeller test rig, and also by using an actual compressor.Using these test results and the rotordynamics analysis, the

influence factor �was estimated as the factor that shows how muchpressure fluctuation works as an external force on the rotor. Theestimated � was approximately 0.15. � of 0.20 was proposed inorder to be on the safe side during the design stage. The accuracyof � should be improved by many future shop tests.

BIBLIOGRAPHY

Ferrara, G., Ferrari, L., and Baldassarre, L, 2006, “ExperimentalCharacterization of Vaneless Diffuser Rotating Stall Part V:Influence of Diffuser Geometry on Stall Inception andPerformance (3rd Impeller Tested),” ASME Turbo Expo 2006:Power for Land, Sea and Air, GT2006-90698.

Fulton, J. W. and Blair, W. G., 1995, “Experience with EmpiricalCriteria for Rotating Stall in Radial Vaneless Diffusers,”Proceedings of the Twenty-Fourth Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 97-106.

Kushner, F., 1996, “Dynamic Data Analysis of Compressor RotatingStall,” Proceedings of the Twenty-Fifth TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&M University,College Station, Texas, pp. 71-81.

Nishida, H., Kobayashi, H., Takagi, T., and Fukushima, Y., 1988,“A Study on the Rotating Stall of Centrifugal Compressors(1st Report, Effect of Vaneless Diffuser Width on RotatingStall),” Transactions Japan Society of Mechanical Engineers(B Edition), 54, (449), pp. 589-594.

Nishida, H., Kobayashi, H., Takagi, T., and Fukushima, Y., 1990,“A Study on the Rotating Stall of Centrifugal Compressors(2nd Report, Effect of Vaneless Diffuser Inlet Shape onRotating Stall),” Transactions Japan Society of MechanicalEngineers (B Edition), 54, (449), pp. 589-594.

Senoo, Y. and Kinoshita, Y., March 1977, “Influence of Inlet FlowConditions and Geometries of Centrifugal Vaneless Diffuserson Critical Flow Angle for Reverse Flow,” ASME Journal ofFluids Engineering, pp. 98-103.

Senoo, Y. and Kinoshita, Y., 1978, “Limits of Rotating Stall andStall in Vaneless Diffuser of Centrifugal Compressors,” ASMEPaper No. 78-GT-23.

ACKNOWLEDGEMENT

The authors gratefully wish to acknowledge the followingindividuals for their contribution and technical assistance in analyzingand reviewing the results, and for their great suggestions and guidancefor practical applications and tests: K. Miyagawa, K. Yamashita, J.Masutani, and M. Ishikawa, of Mitsubishi Heavy Industries.


Phillip Dowson is General Manager,Materials Engineering, with Elliott Company,in Jeannette, Pennsylvania. He has 35years of experience in the turbomachineryindustry. Mr. Dowson is responsible for themetallurgical and welding engineering forthe various Elliott product lines within thecompany. He is the author/coauthor of anumber of technical articles, related to topicssuch as abradable seals, high temperature

corrosion, fracture mechanics, and welding/brazing of impellers.Mr. Dowson graduated from Newcastle Polytechnic in

Metallurgy and did his postgraduate work (M.S. degree) inWeldingEngineering. He is a member of ASM, NACE, ASTM, and TWI.

Derrick Bauer is a Materials Engineer withElliott Company, in Jeannette, Pennsylvania.He joined Elliott Company in 2002, andhas been involved with materials relatedR&D projects, failure analysis, productionand aftermarket support, and remaininglife assessments.Mr. Bauer received his B.S. degree (2002)

from the University of Pittsburgh and iscurrently working toward his M.S. degrees

from the same institution.

Scot Laney is a Materials Engineerwith Elliott Company, in Jeannette,Pennsylvania. He joined Elliott Companyin 2007, and has been involved withmaterials related R&D projects, failureanalysis, and aftermarket support. Healso has experience in the areas ofhigh temperature oxidation/corrosion andprotective coatings.Dr. Laney received his B.S. degree

(2001), M.S. degree (2004), and Ph.D. degree (2007) from theUniversity of Pittsburgh in Materials Science and Engineering. Heis also a member of ASM.

ABSTRACT

In today’s marketplace the selection of materials for the variouscomponents for centrifugal compressors and steam turbines is verycompetitive and an important factor in the overall cost and deliveryof the product. This paper reviews the material selection for majorcomponents for compressors and steam turbines such as shafts,impellers, blading, bolting, seals, etc. This paper is not intended tocover reciprocating, screw compressors, and high temperature hotgas expanders. Various material properties are discussed for themanufacture and service exposure of major components such asimpellers, shafts, bolts, blades, and casings. Due to the variousaggressive corrosive, fouling, and erosive environments in whichboth compressors and steam turbines are operated in, coatingsmust frequently be used to prolong the life of the machine. Theapplication of various coating systems will be reviewed and howeffective the coating system is with respect to withstanding theparticular environment. Also discussed are the repairs for long leadtime components such as rotors and the application of design provenrepair procedures utilizing a design for fitness for service approach.

INTRODUCTION

Over the past 15 years a great deal of progress has been made withrespect to the application of materials and related processes appliedto various components for centrifugal compressors and industrialsteam turbines. This paper will review how materials are specifiedfor the various components and which material properties/factors onemust consider for the application. Both rotating and nonrotatingcomponents of centrifugal compressors and steam turbines will bereviewed for material selection.Due to the very competitive market, material selection has

moved beyond simply finding the material with the most idealproperties. Material cost and delivery can be one of the mostimportant factors in the overall cost and delivery of the product,and therefore have become major drivers when selecting materials.Most original equipment manufacturers (OEMs) are continuouslyreviewing new ways whether by material changes or processes toreduce costs or delivery in order to remain competitive. Thisbecomes even more important with the prevalence of outsourcingand inventory concepts such as “just in time.” OEMs are nowcompeting not only for sales, but also as customers in places likecasting and forging shops. At the same time, the environments thecomponents are exposed to are increasing in severity, leading to the

189

SELECTION OF MATERIALSAND MATERIALRELATED PROCESSES FOR CENTRIFUGAL COMPRESSORS

AND STEAMTURBINES IN THE OILAND PETROCHEMICAL INDUSTRY

byPhillip Dowson

General Manager Materials Engineering

Derrick BauerMaterials Engineer

andScot Laney

Materials Engineer

Elliott Company


need for more specialized materials. The use of these typicallymore costly materials may be avoided by selection of variouscoatings that are resistant to the aggressive environment underconsideration. As stated previously, the paper will review materialsselections for:

• Centrifugal compressors—both rotating and stationary components.

• Industrial steam turbines—rotating and stationary components.

• Application of coatings for protection against corrosion, erosion,and fouling environments.

CENTRIFUGAL COMPRESSORS

The choice of materials for rotating and stationary componentsof centrifugal compressors requires selection based on a number ofdesign and environmental factors. The design factors that requireconsideration are the properties of the materials together with theoperating requirements of the unit. For example, operating amachine intermittently or with variations is usually considered amuch more severe service relative to constant, long-durationservice, and must be designed accordingly. The properties ofmaterials that can be utilized for use in the design are as stated byCameron and Danowski (1973) and listed in Table 1. Long time,high temperature properties such as creep or creep rupture arerarely encountered in centrifugal compressors and, therefore, arenot considered in this selection process. However, in cases wherethe temperature of operation is determined to be significantlydifferent from room temperature, temperature is an importantfactor, since the following properties are temperature dependent:strength of material, corrosion rate, coefficient of thermalexpansion, and fracture toughness.

Table 1. List of Material Properties Used in Design of CentrifugalCompressors.

From an OEM perspective, receiving quality information withrespect to the anticipated operating parameters is vital to producinga compressor that meets the customer’s expectations. API Standard617 (2002) provides some guidelines as to what information shouldbe shared by both the OEM and the purchaser. For instance,paragraph 2.2.1.3 requires the purchaser to specify any corrosiveagents that may be present. As an example of why this is important,there is a substantial difference between the materials that are usedin an air compressor when compared to an application that maycontain wet chlorine.NACE MR0175 (2003) and MR0103 (2005) define which

ferrous and nonferrous alloys can be used in wet hydrogen sulfideservice to resist sulfide stress corrosion cracking. The alloys allowedby these specifications must also be heat treated in accordance withthe specifications and meet a maximum hardness limit. NACEMR0175 (2003) is labeled “Metals for Sulfide Stress CorrosionCracking and Stress Corrosion Cracking Resistance in SourOilfield Environments,” which applies to a component ormachinery used for petroleum production, drilling, flow lines, andfield processing facilities exposed to wet hydrogen sulfide service.NACE MR0103 (2005) is titled “Materials Resistant to SulfideStress Corrosion Cracking in Petroleum Refining Environments,”and is thus more specific to compressor components. Compressorswith wet hydrogen sulfide present in the process gas are usuallyrequired to meet one or both of these NACE specifications to avoidstress corrosion cracking during service.

Rotating Components

The heart of a centrifugal compressor is the rotor, which consistsof a series of impellers and a shaft. The impellers are designed toaccelerate the process gas, which causes it to be compressed in theproceeding diaphragm, while the shaft provides the support androtation to the impellers.

Impellers

Given the importance of the impellers, a great deal of attentionis given to their manufacture. Generally, centrifugal compressorimpellers are fully shrouded, consisting of a solid hub and coverseparated by radial equally spaced blades. The attachment of theblades to the impeller hub and cover can be either done by welding,integral cast, brazing, riveting, electrodischarge machining, integrallymachined to the hub and/or cover, or a combination of these methods.A fully shrouded impeller is shown in Figure 1. Today, for mostimpellers, the blades are integrally machined to the impeller hub orcover and then welded to the nonmachined blade hub or cover.

Figure 1. Photograph of a Fully Shrouded Centrifugal CompressorImpeller.

Throughout the compressor industry selection of impellermaterials can vary depending upon service operating conditions.Since operating conditions govern the material selection requirements,a number of mechanical and chemical properties such as yield andtensile stresses, low temperature properties, corrosion resistance,hydrogen sulphide resistance, weldability, and machinability mustbe considered. For over 50 years, OEMs have utilized impellermaterials such as AISI 4330/4340, 4130/4140, AISI 410/17-4PHand 13Cr4Ni and nickel (Ni) base alloy.Impellers for centrifugal compressors are typically made from

low alloy steels or stainless steels, and both NACE MR0175 (2003)and MR0103 (2005) have identical requirements for commonimpeller materials. Impellers made from AISI 4140 or 4130 steelhave to meet a maximum hardness requirement of 22 HRC in thebase metal, weld metal, and heat-affected zone (HAZ). In order toachieve the necessary hardness requirement, AISI 4140 impellershave to be austenitized, quenched, and tempered after welding toeliminate the HAZ. UNS G43200 and modified versions of G43200with higher carbon contents are acceptable for compressor impellersby both NACE MR0175 (2003) and MR0103 (2005), provided thatthey are heat treated per the NACE standards. These UNS G43200have a maximum yield strength requirement of 90 ksi, but there isno hardness limitation given per the NACE standards. AISI 410stainless steel meets both applicable NACE standards if it isaustenitized and double tempered with the second temper beingperformed at a lower temperature than the first temper. Both NACEMR0103 (2005) and MR0175 (2003) allow the use of low carbonstainless steel CA6NM (13%Cr-4%Ni) for impellers at a maximumhardness of 23 HRC after the material has been austenitized anddouble tempered at the temperature ranges defined in the NACEspecifications. Precipitation hardenable stainless steels UNS17400 (17-4PH) and UNS15500 (15-5PH) also meet both NACEspecifications at a maximum hardness of 33 HRC after it has beenthrough a solution annealing followed by a double aging treatment.


Since the maximum hardness requirement is 33 HRC for thesesteels, they are often used when impellers with higher yieldstrengths are necessary in hydrogen sulfide service environments.Although other manufacturing techniques have been utilized in the

past 20 years, most impellers are welded using various weld processes.Other manufacturing techniques that have been utilized are:

• Cast impellers.• Electrodischarge machining to shape the gas passage ofimpellers.

• Riveted impellers.• Brazing impellers.• Electron beam welding.

• Combination of electron beam welding and brazing.

• One piece machine impellers.

Early on, shielded metal arc was the welding process utilized forjoining the sections of impellers, whether it would be milled bladesto cover/hub to the unmilled section or joining preformed blades toa machined hub and cover. This welding process is still used today.In the 1980s several manufacturers started to utilize automationwith other welding processes such as gas tungsten arc welding(GTAW), gas metal arc welding (GMAW), and submerged arcwelding (SAW). Generally, these processes were applied to simpletwo-dimensional (2D) configuration impellers with 2D curvedblades. Special torches were designed for both open bladed andclosed configuration welding.Due to technological advances in five axis milling, two-piece

three-dimensional (3D) geometric compressor impellers with varioustypes of three-dimensional curved blades were being designed andmanufactured. These impellers were designed to increase performanceof the wheel. Since the impellers are two-piece designed, that is,either the blades are milled to the hub or cover, material selectionbecomes more important for consideration of manufacturability.Brazing of impellers has been used since the 1950s but in the

recent years its popularity has increased. Early brazed impellers weredone in a dry hydrogen atmosphere. More recently, however,impellers are done in a vacuum furnace, which produces moreacceptable results and consistency. The inspection of the braze jointshas also improved over the years with the application of C-scanimmersion ultrasonic testing. The author’s company utilizes a specialcalibration block to represent the impeller braze joint. This specialblock was utilized for the application of C-scan ultrasonic immersionequipment. The Figure 2 shown is a printout of an acceptable C-scanultrasonic immersion test. Brazed impellers are also fluorescentpenetrant inspected to validate the soundness of the braze joint.

Figure 2. Immersion C-Scan Ultrasonic Results from a BrazedImpeller Joint.

Materials such as low alloy steels have limited hardenability.Hardenability also becomes an issue as the diameter of the

impeller increases. This is illustrated in Figure 3, which shows thehardenability curve for various section diameters of AISI 4140. Tounderstand why material mechanical properties such as AISI 4140,4340, and 4330 diminish from the surface to the center of thematerial, one must review the continuous cooling transformationdiagram (CCT) for that particular alloy and an elementaryunderstanding of heat transfer. Figure 4 shows the CCT for AISI4140 (Atkins, 1980). Depending on the rate of cooling, whether byair, oil, polymer, or water, for various thicknesses of material,certain microstructure phases such as ferrite, pearlite, bainite, andmartensite are formed. Each one of these phases has an effect onthe strength, ductility, and toughness of the material. Table 2(Cameron, 1989) shows mechanical properties of impellermaterials. Consequently, special consideration must be appliedwhen selecting these low alloy steels in heavy forgings from whichthe blades are milled. For example, during austenitizing treatmentto a fabricated impeller followed by a rapid quench, specialattention must be made to changes in section thickness, such as thatwhich occurs where the blades are welded to either the hub or coverin order to minimize the risk of quench cracking at the blade tohub/cover location. In some cases, quenching by forced air-coolingor salt bath may be required to avoid the quench cracking phenomenon.Other high alloy materials such as AISI 410, 17-4PH, and 13Cr4Nihave good hardenability with less drastic cooling rates; requiring onlyair or forced air cooling to achieve the desired mechanical properties.Figures 5 (Atkins, 1980) and 6 (Arlt, et al., 1988) show the CCTdiagrams for AISI 410 and 13Cr4Ni, respectively.

Figure 3. Hardenability Curves for Standard and Controlled YieldStrength AISI 4140.

Figure 4. CCT Curves for AISI 4140.



191

Table 2. Mechanical Properties of Impeller Materials.

Figure 5. CCT Diagram for AISI 410.

Figure 6. CCT Diagram for 13Cr4Ni.

The material 13Cr4Ni (UNS S42400) has been utilized formanufacturing impellers to meet a wide range of requirementssuch as high strength, controlled hardness H2S, low temperature,and good corrosion resistance. For low temperature applicationsusing 13Cr4Ni, controlling the volume fraction of austenite in thestructure increases the material toughness. This increase intoughness also can lead to a reduction in hardness. Dowson (2002)also substantiated the application of a two-stage temperingtreatment to obtain the maximum requirement of HRC 23maximum. For welding, matching welding consumables are alwaysapplied to higher alloy steels. The author’s company has formulateda patented chemistry for the welding consumables of 13Cr4Ni tobe applied for impellers. This patent chemistry enabled HRC 23maximum hardness to be achieved consistently for H2S controlledhardness environments. Procedures were developed using 13Cr4Nifor the various operating conditions utilizing the various welding

processes and brazing application. These procedures weredeveloped for welding and brazing applications requiring highstrength, qualification to meet NACE MR0175-200 (2003)(metallic sulfide stress cracking resistant materials), and lowtemperature (down to �166�F (�110�C) applications. The Figures7 and 8 show the properties that can be achieved using the 13Cr4Nimaterial (Dowson, 2002).

Figure 7. Plot of Strength and HardnessVersus TemperingTemperature(4 Hour Hold) for 13Cr4Ni.

Figure 8. Plot of Charpy Impact EnergyVersusTemperingTemperature(4 Hour Hold) for 13Cr4Ni.

The precipitation grades such asArmco 17-4PH,Armco 15-5PH,and Carpenter Custom 450 obtain their mechanical properties bysolution treatment and precipitation treatments. By varying theprecipitation treatments, one can obtain a range of tensile strengths.In comparing the Armco 17-4PH with Armco 15-5, whosechemistries overlap, 17-4PH is more prone to form delta ferrite than15-5PH, which may result in a loss of ductility/toughness. Theprecipitation treatments for these grades are usually in the range of1050�F to 1175�F. For application in hydrogen and hydrogen sulfideenvironments, 17-4 or 15-5PH material is given a double precipitationtreatment at approximately 1150�F to obtain the necessarymaximum hardness of HRC 33 maximum for H2S environmentsand 120 ksi maximum yield strength for hydrogen environments.

Special Impeller Materials

There are other special materials that are utilized in certainspecialized applications. In moist halogen machines, such as achlorine machine, Monel® K500 has been successfully used forimpellers. Monel® K500 was also used in oxygen machines,because of its resistance to sparking. Special welding procedureswere developed to obtain good quality fabricated Monel® K500impellers. Titanium and titanium alloys have been used for wetchlorine and in special cases where low density makes the materialattractive. Precipitation hardenable nickel base alloys such as UNS


N0 7716 and UNS NO 7725 have been applied to aggressivecoal gasification applications with a great deal of success. Fortemperatures down to �320�F, the 9%Ni steel has widely beenaccepted for impellers in compressors for boil off gas from liquidmethane due to its high fracture toughness at these low temperatures.Special processing of these grades is required to obtain therequired mechanical properties. For some applications, aluminumalloys have also been used successfully especially in connectionwith liquefied natural gas projects.

Rotor Shafts

Generally, shafts are manufactured from either rolled bars orforgings. In the size range where both are acceptable by API617 (2002) (currently inches finished machined), the choicebetween rolled bar or a forging should be dependent on costand availability, as there is little difference in the end result, ifmanufactured to relevant American Society for Testing andMaterials (ASTM) procedures. Tolerances on rolled bar can bemore tightly controlled, which results in less machining requiredthan for forgings and have a slight advantage in terms of thermalstability. Conversely, depending on availability of the particularmaterial and size, selecting rolled bar may require the purchase ofan entire length of bar or an entire heat, while forged shafts are“made to order.” Shafts with a finished diameter greater than 8inches are generally made from forgings.For shaft materials the standard AISI 4330 and 4340 can be heat

treated to give the required mechanical strength and toughnessvalues. Due to these materials having higher strength values, thelateral expansion parameter is utilized for ASME Boiler and PressureVessel Code Section VIII Division I (2007). The requirement is 15mils, which was proposed by Gross and Stout (1957). Thesematerials have been used successfully down to �150�F with thetensile strength at 110 ksi minimum and yield strength of 90 ksiminimum. For temperatures lower than �150�F, ASTM A470Class 7 has been used with a great deal of success.Shafts that are used in H2S environment cannot be limited to the

NACE requirements due to the need for the higher strengthrequired at the drive end of the shaft. Consequently, the appliedstress in the main body of the shaft, which comes in contact withthe gas, is low (<10 ksi). This value is below the threshold stressvalue for sulfide cracking as a function of hardness (Figure 9)(Warren and Beckman 1957).

Figure 9. Threshold Stress for Sulfide Cracking as a Function ofHardness.

The selection of shaft materials is also important to prevent certaintypes of failures, in particular, wire wooling. Wire wooling is aninfrequent, but devastating, failure mechanism that occurs duringstartup (not necessarily the initial startup) in the bearing and seal areashafts. Areas where oil films are thin and loads are high, such as at

the thrust bearing, are most susceptible to this type of failure. Theprocess of wire wooling begins when a small particle of foreignmaterial enters the bearing or seal. Through a series of localizedtemperature increases, due to high coefficient of friction betweenparticle and rotor, material transfers from the rotor to the particle,and by hardening mechanisms, the particle becomes a hard, blackscab that is able to cut and spin material from the shaft. Shaftmaterial also continues to transfer to the scab as the cutting isoccurring, allowing the scab to grow and propagate the failure. Theresult is a deeply grooved shaft. The material cut from the shaft oftenhas the appearance of wire wool. Figure 10 shows damage to thejournal and thrust collar due to wire wooling and the correspondingblack scab on the journal bearing. For a more detailed description ofthe mechanism, the reader is directed to Fidler (1971).

Figure 10. Photographs ShowingWireWooling Failure of a Journaland Thrust Collar and the Corresponding Black Scab on theJournal Bearing.

For reasons that are not clearly understood, shaft materials withhigher Cr contents are more susceptible to wire wooling. Thecritical Cr content proposed by Fidler (1971) is 1.8 percent.Because of this, shafts should be made of 4140 or 4340 whereverpossible. If a higher Cr content material must be used, steps mustbe taken to reduce the risk of wire wooling. Altering the surface ofthe shaft or increasing clearances are ways that can be used toreduce the risk. One way to alter the surface of the shaft is toharden it by nitriding or hardfacing with hard chrome or carbidecoatings. Sleeving with a wire wooling resistant material, such as4140, can also work. Increasing the clearances reduces risk byincreasing the size of particle able to initiate the failure. All ofthese have their pros and cons, which must be evaluated whendeciding which method yields the best overall results.

Stationary Parts

The bulk of the centrifugal compressor consists of stationary parts.The most obvious is the casing, which is the pressure containing shellthat surrounds the rotor. Diaphragms are responsible for slowingdown the gas after it is accelerated by the impellers, causing it to becompressed. Bolts are used to fasten the parts together. Split lineseals fit between the casing pieces to maintain the overall pressure.Abradable and rub tolerant seals maintain the pressure between eachindividual stage. Piping is used for gas path connections.

Casings

Nearly all multistage centrifugal compressor casings are madefrom carbon or a low alloy material that is either cast or fabricatedfrom castings, forgings, plate, or a combination of these.Occasionally higher alloy steels are required for aggressivecorrosion conditions or to achieve the desired toughness for extremelow temperature (�320�F) operating conditions, such as those forboil off gas compressors in liquified natural gas service. For lowtemperature application down to �175�F, low alloy steels such asshown in Table 3 are applied in propane and ethylene compressors.The table must be used with some caution and understanding of theeffect chemical makeup will have on the ability to achieve thedesired toughness.



193

Table 3. MinimumTemperature for LowTemperature Application ofLow Alloy Steels.

Compressor casings are typically manufactured from carbonsteel castings or carbon steel plate that is formed and welded.NACE MR0103 (2005) limits the hardness of carbon steelcastings and plate to 200 HBW maximum and NACE MR0175(2003) limits the hardness to 22 HRC (237 HBW). Underboth specifications, carbon steel castings are acceptable in thenormalized and normalized and tempered conditions, andcarbon steel plate is acceptable in the hot-rolled condition aslong as the material meets the maximum hardness requirement.The hardness of the carbon steel is largely controlled by thecarbon content, so keeping the carbon content low is essential tomeeting the maximum hardness requirement. Welding of thecarbon steel plates under NACE MR0103 (2005) is controlledby NACE MR0472 (2005) which also limits the maximumhardness in the base metal and the weld metal to 200 HBWwhile limiting the hardness of the HAZ to 248 HV 10. Themaximum hardness of 22 HRC is required in the base metal,weld metal, and HAZ under MR0175 (2003). The chemicalcomposition of the filler metal used during welding must besimilar or identical to the base metal composition. The hardnessmay be verified by the welding procedure qualification when allof the welding parameters and filler metal composition definedby the procedure qualification are controlled and followed. Apost weld heat treatment (PWHT) may be performed to ensurethat the hardness values meet the required specifications. NACEMR0103 (2005) has a lower maximum hardness limit thanMR0175 (2003) to compensate for nonhomogeneity of someweld deposits and normal variations in production hardnesstesting using a portable Brinell tester.During the last 15 years, compressor casings have become

larger with increases in pressure rating. This has also beenapplied to applications where the compressor casings aresubjected to low temperatures. Design engineers now requirethicker sections due to the higher pressure ratings to prevent gasleakage at the various joints connections. For the thickersections, special controlled chemistry and heat treatments arerequired to achieve the desired toughness for low temperatureapplication. For low alloy nickel steels, the elements C, S, and Pare controlled to ensure good toughness is attained. Table 4shows results of chemistries and their corresponding toughnessresults. These chemistries not only apply for steel plate but alsofor forgings and castings. Since the application of argon oxygendecarburization (AOD) and calcium-argon injection, cleanersteels, with sulfur as low as 0.005 percent, are attainable.Because of the addition of calcium compounds, the inclusionsthat remain resist elongation during rolling; remainingspherical. Consequently, inclusion shape control steel platesshow improved ultrasonic quality, macrocleanliness, andmechanical properties as can be observed in Figures 11, 12, and13 (Lukens Fineline Steels).

Table 4. Results of Chemistries and Their CorrespondingToughness Results.

Figure 11. Comparison of Sulfur Contents of 50 Heats of ConventionallyProcessed Steel and 50 Heats of Lukens Fineline Double-O Five.

Figure 12. Comparison of Charpy Impact Energy for Conventionaland Fineline ASTM A633C Plate, 4 Inches Thick, Normalized.

Figure 13. Comparison of Charpy Impact Energy for ASTMA633CPlates, 1 to 2 Inches Thick, Normalized, Transverse Orientation.


The introduction of inclusion shape control steel plates has alsominimized the susceptibility of lamellar tearing during the fabricationof casings. Generally, this phenomenon can occur in thicknessesgreater than 3 to 4 inches. By achieving lower carbon, sulfur,phosphorus, and inclusion shape control structures, excellenttoughness properties can be achieved. However, for the largerthicknesses, accelerated quenching during heat treatment using wateras the media from the austenitizing temperature is required. Tomaximize the quench affect of the water, the following is required:control of water temperature 80�F maximum, good agitation of thewater, adequate amount of water for the weight of component (2gallons per pound of metal) and time from furnace to water less thanone minute. In some cases movement of the component in the watersideways or up and down is required (Figures 14, 15, 16, and 17)(Metals Handbook, 1981).

Figure 14. Surface Cooling Power of Moderately Agitated WaterVersus Water Temperature.

Figure 15. Effect of Concentration on Cooling Rate.

Figure 16. Effect of Temperature on Cooling Rate.

Figure 17. Effect of Agitation on Cooling Rate.

In Cameron’s (1989) paper, the changes in the 1987 ASMEBoiler and PressureVessel Code Section 8 Division I as required byAPI 617 (2002) specified the required Charpy V-notch energyabsorption values had been made a function of the plate thickness.Refer to ASME Boiler & PressureVessel Code - SectionVIII (2007)to show the values required. Since most of the compressor casingsare fabricated, the welding consumables have to satisfy the impacttest requirement of the base material. Actual test results of theweldments and surrounding heat-affected zone depend strongly onthe preheat, interpass temperature, thickness of individual weldbead, heat input rate, and post weld heat treatment. Manufacturershave developed specific welding consumables, processes, andprocedures to obtain the required mechanical properties for thecasings. Since the soaking time of post weld heat treatments canaffect the toughness of the materials as well as the weldments,simulated PWHT is performed on the materials that are used forlow temperature application. Figure 18 shows effect of PWHT timeon toughness (Charpy V-notch impact value).

Figure 18. Impact Energy Versus Temperature for SA516 Grade 60after Various PWHT.

Diaphragms

In the past, diaphragms were generally constructed from greycast-iron materials. When higher strength is required, ductileirons or fabricated mild steel is applied. In all cases, the casting orfabrications are heat treated to produce low levels of internal stressand difficulty with instability in service is virtually unknown. Inthe last 10 years, the choice of diaphragm material has been mildsteel with the blades either milled integral or welded and thetwo-piece construction bolted together. One of the main reasons forthe change is that the mild steel materials can be repaired bywelding more readily as compared to grey cast-iron. The use ofductile iron that has improved weldability over grey cast-iron canalso be selected as a material choice for diaphragms.

Sulphide Stress Cracking

There is extensive literature on sulfide cracking in oil well casingmaterials going back to about 1950. Incidents involving centrifugalcompressors, however, have been rare as reported by Kohut andMcGuire (1968) and Moller (1968). However, a failure due to sulphidestress cracking (SSC) can be serious; leading to loss of service of avital link in the production chain of a chemical or petrochemical plant.For SSC to occur, the following conditions must be fulfilled:

• Hydrogen sulfide must be present.• Water must be present in the liquid state.• The pH must be acidic.

• A tensile stress must be present.

• Material must be in a susceptible metallurgical condition.

195SELECTION OF MATERIALSAND MATERIAL

RELATED PROCESSES FOR CENTRIFUGAL COMPRESSORSAND STEAMTURBINES IN THE OILAND PETROCHEMICAL INDUSTRY

When all the above conditions are present, sulfide cracking canoccur with the passage of time.As stated previously, all manufacturers follow the requirements

specified by NACE. In reviewing the work done Treseder andSwanson (1968) showed the effect of pH on sulphide stresscracking, Figure 19 where Sc is an experimental stress value. ThisSc value increased substantially as the pH increased from 2 to 5.Other work done by Keller and Cameron (1974) has shown theimportance of pH. In their work, quenched and tempered weldedAISI 4140 material stressed to 80 percent of yield strength failedat pH 2.5, while at pH 4.2 no specimens failed. The yield strengthof the base material was reported to be 126 ksi. From the samepaper the material AISI 4140 quenched and tempered with a yieldstrength of 83 ksi was welded and given a temper treatment of1100�F. In this case, at pH 2.5 all specimens failed. At pH 4.2, alarge percentage of specimens failed, while at 6.5 pH there wereno failures in those tests. It is important to note that these arecontrolled laboratory experiments. In practice, it is difficult tospecify a threshold concentration of H2S in a gas below whichsulfide cracking will not occur, because other components of thegas can cause the pH to vary. Chlorides, for example, can lower thepH well below 4.3, while other components may cause an increase.

Figure 19. Critical Stress for Sulfide Cracking as a Function of pH.

Bolting

Bolting for use in hydrogen sulfide service environments is alsodefined by NACE MR0175 (2003) and MR0103 (2005). Bothspecifications allow the use of AISI 4140 steel and AISI 410stainless steel at a maximum hardness of 22 HRC provided thematerial is quenched and tempered. Both AISI 4140 and AISI 410bolting is commonly available from bolting suppliers; however, anextra tempering treatment is required to meet 22 HRC maximumhardness requirement and conform to ASTMA193 Grade B7M forthe AISI 4140 steel or ASTM A193 Grade B6M for AISI 410stainless steel. For higher strength bolting materials, UNS 17400and UNS 15500 stainless steel may be used. NACE MR0175(2003) allows these alloys to be used at a hardness of 33 HRCwhile NACE MR0103 (2005) limits the hardness to 29 HRCmaximum for pressure-retaining bolting.

Piping

The piping of a compressor unit is typically manufactured fromcarbon steel piping. NACE MR0103 (2005) limits the hardness ofcarbon steel piping to 200 HBW while NACE MR0175 (2003)allows carbon steel piping up to 22 HRC (237 HBW). The carboncontent of the carbon steel pipe has a large influence on thehardness, so the carbon content must be limited to comply with themaximum hardness limit. Under both NACE specifications, thepiping must also be thermally stress relieved following any colddeforming by rolling, cold forging, or any other manufacturingprocess that results in a permanent outer fiber deformation greaterthan 5 percent. For carbon steel piping welds, NACE MR0103

(2005) defers to NACE RP0472 (2005) which allows a maximumhardness of 248 HV10 in the HAZ and a maximum hardness of200 HBW in the base metal and weld metal. NACE MR0175(2003) requires a maximum hardness of 22 HRC at all carbon steelpiping weld locations. When matching filler metals are used, a postweld heat treatment is not required if all hardness values meet themaximum hardness requirement. The hardness values may beverified by the welding procedure qualification if all weldingparameters and the filler metal composition are defined by thequalification and are followed during production per NACEMR0175 (2003). Under NACE MR0103 (2005), which defers toNACE RP0472 (2005), 5 percent of all production piping buttwelds must be hardness tested.

Splitline Seals/O-Rings

Turbomachinery casings usually consist of two or more pieces,which are bolted together. With respect to centrifugal compressors,the seam created is sealed using elastomeric materials in the formof O-rings. The choice of material used for this seal is vital to theoperation of the compressor, as it is typically the weakest link inthe casing in terms of chemical resistance, temperature resistance,mechanical properties, longevity, etc. Due to the wide range ofoperating parameters encountered in a centrifugal compressorand wide range of reactions to those parameters by the variouselastomers, there is not a single, one stop choice. Table 5 is acompatibility chart for some common gases and O-ring materials.In fact, choosing the proper material frequently consists of anevaluation and acceptance of several compromises, particularlyin cases with complex process gases. As a simplified example,fluoroelastomers will work well in hydrogen and nitrogen separatelyup to 300�F. If they are mixed at 250�F, such that a small amountof ammonia can form, the situation is no longer clear cut. Thefluoroelastomer is not compatible with ammonia, nitrile has amaximum temperature of 212�F, and silicone is not compatiblewith hydrogen. Clearly a compromise must be made to accept whatshortcoming will occur when the chosen material is exposed to theoffending environmental parameter.

Table 5. Compatibility of Split Line Seal Materials in VariousEnvironments.

Abradable and Rub Tolerant Seal Materials

Selection of abradable material for application in centrifugalcompressors is very important. There are a number of propertyfactors that need to be considered when selecting the abradablematerial:

• Abradability and erosion resistance• Compatibility of the abradable material with the gas• Temperature limits of the abradable material• Coefficient of thermal expansion of the material

Abradability and erosion resistance—In the paper presented byDowson, et al., (1991), abradable material mica-filled trifluoroethanol(TFE), nickel graphite, and silicon rubber were found to have goodto very good abradability. The abradable silicon aluminum polyester,although not as good as those abradables stated previously, didperform well at lower rates of interaction and higher rubbing


velocities. The author’s company has applied the silicon aluminumpolyester abradable material in various centrifugal compressorapplications with great success. Also since the 1991 paper, nickelgraphite abradable seals have been used extensively in variousapplications for centrifugal compressors. However, to achieveoptimum abradability and erosion resistance, special control of thehardness needs to be achieved. In the early years of manufacture ofthese seals, it was found that high heat input spray processes suchas plasma could reduce the percent graphite in the coating andthereby reduce the abradability. During testing of an abradablenickel graphite seal the labyrinth rotating teeth were found to gallwith the seal causing excessive damage to the impeller seal eyelocation (Figure 20).

Figure 20. Damage to Nickel Graphite Impeller Seal.

Compatibility of the abradable material with the gas—Themica-filled TFE is generally impervious to corrosive attackfrom most gaseous mixtures and would be acceptable with allhydrocarbon gases, sour natural gas, and chlorine gaseous conditions.However, for wet gaseous conditions (water > 2 percent) certaindesign considerations must be addressed to account for waterabsorption leading to possible swelling of the seal during service.The aluminum silicone abradables would operate under the samegaseous conditions that are generally applied to conventionalaluminum seals. Exposed to extreme sour natural gases and chlorinegaseous conditions, these materials would be attacked severely. Thenickel graphite coating could be used for all hydrocarbon gases andmost sour natural gases. However, for hydrocarbon gases thatcontain large quantities of carbon monoxide (CO), a reaction canoccur between the base coat of nickel and CO leading to delaminationof the nickel graphite abradable seal as shown in Figure 21.

Figure 21. Delamination of Nickel Graphite Seal Due to Reactionof CO with Ni.

Temperature limits of the abradable material—The temperaturelimits for the various abradables are shown in Table 6.

Table 6. Temperature Limits.

Coefficient of thermal expansion of the material—Whendesigning seals, the coefficient of thermal expansion of the materialmust be taken into account at the design. The mica-filled TFEmaterial has a coefficient well above that of steel and, therefore,dimensional changes due to temperature must be calculated in theoverall design of the compressor. However, since the sprayedabradables are only 0.1 inch thick bonded to a metallic substrate,the coefficient of the substrate would apply for design purposes.For silicon rubber abradable material, the material is flexible/softand will deform elastically to accommodate any thermal strainscaused by the substrate.

Rub Tolerant Polymer Seals

Rub tolerant labyrinth polymer seals are seals with reducedclearances and, if contact is made between the stationary seal andsmooth rotating member, the stationary teeth will deflect duringcontact without wear or damage to the rotor or seals.The rub tolerant plastic seals that are used in centrifugal

compressors today are the new thermoplastics, which havebetter resistance to elevated temperatures. The thermoplasticsmatrix materials are tougher and offer the potential of improvedhot/wet resistance.Because of their high strains to failure, they are the only matrixes

that offer the new intermediate modulus, high strength (and strain)carbon fiber to use their full strain potential in the composite. Thesematerials include such resins as polyetheretherketone (PEEK),which is intended to maintain thermoplastic character in thefinal composite. Others, such as polyamideimide (PAI), which isoriginally molded as a thermoplastic and is then postcured in thefinal composite to produce partial thermosetting characteristics.The partial thermosetting characteristic of the PAI enables animproved subsequent temperature resistance (Engineered MaterialsHandbook: Composites, 1987).When considering thermoplastic materials in rotating equipment,

one has to understand their thermal properties (Table 7). The twothermoplastic materials used exclusively as labyrinth seals incentrifugal compressors are PEEK with additives and PAI. For thePEEK materials as labyrinth seals the temperature limit is dependanton the glass transition temperatures Tg of the material. Addition ofadditives such as chopped or continuous wound carbon fibers orgraphite powder or PTFE will not increase the Tg of the material. TheTg is the temperature at which the crystalline polymer changes to aviscous or rubbery condition. In other words, the material has adramatic change in properties. Generally, for labyrinth seals fromPEEK material, the operation temperature is limited to 290�F.



197

Table 7. Properties of Thermoplastics.

For PAI materials, since the material is partial crystalline incharacter with some amorphous features, the Tg of the material ishigher. However, when the crystalline character of the polymer isdecreased, its resistance to solvents and water decreases also. Thehigher the degree of crystallinity, the higher the modulus and thehigher the resistance to solvents and water. In the case of PAI carefulconsideration must be given to attack from amines, ammonia,oxidizing acids, and stray bases. Due to its partial crystallinity incharacter, PAI is prone to moisture absorption. Due to the moistureabsorption, dimensional changes can occur that need to be addressedat the design stage and in manufacturing/storage. Generally,thermoplastic resin suppliers provide some information on waterabsorption after a 24 hour immersion in water. A more severe test isthe 24 hour boiling water test.Due to the high thermal expansion coefficient of both the PEEK

and PAI materials, accurate calculations of the growth of the sealswith respect to the diaphragms and rotating components need to bedone to enable one to calculate the clearances.Also, in the manufacture processing of PAI or PEEK into a tubular

form, there exists a size limitation for polymer labyrinth seals.

• PAI—In tubular bulk form the size limitation is 35 inches.However, larger sizes can be constructed up to 45 inches bysegmenting the seals.

• PEEK with chopped carbon fibers—Generally supplied inmolded bulk form with a size limitation of 30 inches.

• PEEK with 68 percent continuous wound carbon fiber—Generally supplied in a tubular form and there is no size limitation.

Abradable seals have been used successfully in centrifugalcompressors for reducing clearances and improving the efficiency ofcompressors. The author’s company has applied abradable materialsuch as Ni graphite, aluminum silicon polyesters, and fluorosint tonumerous labyrinth seal applications with great success.Careful consideration must be applied to ensure that rub tolerant

polymer seals can be utilized in centrifugal compressor labyrinths.The tests that were done at the author’s company indicated that thePAI material for labyrinth seals may not be suitable for temperaturesgreater than 100�F using similar clearances to that of abradables.The PAI material at a temperature of 150�F was found to weardramatically when it came into contact with the rotating member.Other polymer materials, such as PEEK or carbon wound PEEK,may be more suitable (Dowson, et al., 2004).

STEAM TURBINES

Steam turbines present some different problems relative tocentrifugal compressors. Temperatures are usually higher, whichaffects the various temperature dependent processes like corrosionrate and degradation mechanisms, such as creep, must now beconsidered. The key components of a steam turbine are the rotor,blades (buckets), casing, and bolting.With the exception of 12%Cr blading alloys, low alloy materials

have been conventionally used for the major components of steamturbines for many years. For example, 1CrMoV has been used forhigh temperature rotor forgings and 3½NiCrMoV for low temperatureforgings. High temperature pipe work and castings for valve chestsand turbine casings utilize 2¼CrMo or CrMoV. These materials were

available in the 1950s and 1960s and over the past 30 to 40 yearsmaterials development concentrated on optimizing the properties ofthese materials for their application to larger components, which inturn, increases the reliability of components and their fabrications.

Rotors

Mechanical drive steam turbine rotors are manufactured fromlow alloy steel forgings. Rotors consist of a shaft and a series ofdisks, which may be monoblock, in which the disks are integralparts of the shaft (A type), disks shrunk fit onto the shaft (B type),or sections of the rotor may be welded together (C type).Mechanical drive turbines are in general “A” or “B” type rotors,with the majority being “A” type.For the higher temperature application up to 1025�F, the

1Cr1Mo¼V steels are currently in wide use but are limited to1025�F. For higher temperatures, the 12%Cr steel forgings havegood experience at 1050�F, where they have an excellent combinationof creep strength, ductility, and toughness. However, there is adrive to use the most creep resistant version of 12%Cr up totemperatures of 1100�F. In order to provide an additional margin ofsafety, the allowable creep strength at 1100�F needs to be equivalentto that currently used with 12%Cr steel at 1050�F (Gold and Jaffee,1984; Armor, et al., 1984). For developing improved alloys, atentative goal for rupture strength has been established at 14.5 ksifor a rupture life of 100,000 hours at 1100�F. Figure 22 shows acomparison of 100,000 hour rupture strengths for various low alloysteels, 12%Cr and other development rotor steels. The Figure 23shows the Larson-Miller rupture curves for commercial anddevelopmental 12 percent rotor steel (Newhouse, 1987). Othercritical material properties for rotor integrity are toughness,resistance to crack, initiation under creep, thermal fatigue andresistance to subcritical crack propagation in creep and fatigue.Toughness is discussed in more detail below.

Figure 22. Plot Showing 100,000 Hour Rupture Strength of VariousRotor Steels.

Figure 23. Larson-Miller Rupture Curves for Commercial andDevelopmental 12 Percent Rotor Steel.


Toughness

In CrMoV rotors, ASTM A470 Class 8 specification limits thefracture appearance transition temperature (FATT) to 250�F. TheFATT values for the base material normally range from 185 to260�F. Since the advancement of fracture mechanics technology, ithas now become possible to characterize toughness in terms of acritical crack size ac. The cold start sequence of a rotor is mostcritical and the critical flaw size ac based on analysis of temperatureand stress can lead to catastrophic failure (Viswanathan and Jaffee,1983). For the Gallatin rotor, the temperature at the time of thepeak stress 74 ksi was 270�F. The combined crack size reached itslowest value 0.27 inches at this temperature and stress (Figure 24).

Figure 24. Illustration of Cold-Start Sequence and AssociatedVariations in Stress (σ), Temperature (T), and Critical Flaw Size(ac) as Functions of Time from Start.

Variations in temperature, stress and material in homogeneitythroughout the rotor dictate the critical flaw size ac. By using lowerscatterband values of KIC, the critical flaw size can be converted(Schwant and Timo, 1985). Another method of estimating thelower-bound values of KIC for CrMoV steel is by using theexpression (Jones, 1972):

where:KIC is expressed in MPaT is in �CAlthough equipment manufacturers have records of the FATT

value of the rotor material prior to service, the effect of temperembrittlement during service increases the FATT value and decreasesthe KIC value (Figure 25) (Viswanathan and Jaffee, 1983).

Figure 25. Effect of Temper Embrittlement on Fracture Toughnessof CrMoV Rotor Steel.

The maximum temper embrittlement occurring at location of therotor exposed to temperatures from 700�F to 800�F. Consequently,evaluation of FATT (or KIC at the concerned location) in theservice exposed condition is critical for damage assessment(Dowson, et al., 2005).

An ASTM special task force on large turbine generator rotors ofsubcommittee VI of ASTM Committee A-1 on steel has conducteda systematic study of the isothermal embrittlement at 750�F ofvacuum carbon deoxidized (VCD) NiCrMoV rotor steels. Elements,such as P, Sn, As, Sb, and Mo were varied in a controlled fashion andthe shifts in FATT, (�) FATT were measured after 10,000 hours ofexposure. From the results, the following correlations were observedin Equation (2):

where:�FATT is expressed in degrees Fahrenheit and the correlation of allthe elements are expressed in weight percent. According to thiscorrelation, the elements P, Sn, andAs increase temper embrittlementof steels, while Mo, P, and Sn interaction decrease the temperembrittlement susceptibility.All available 10,000 hour embrittlement data are plotted in

Figure 26 as a function of calculated �FATT using Equation (2)(Newhouse, et al., 1972). A good correlation is observed betweencalculated and experimental �FATT. The scatter for these data isapproximately ± 30�F for 750�F exposure and ±15�F for the650�F exposure.

Figure 26. Correlation Between Compositional Parameter “N”and the Shift in FATT of NiCrMoV Steels Following Exposure at650�F and 750�F for 8800 Hours.

Other correlations for determining the temper embrittlementsusceptibility of steel, such as the J factor proposed by Watanabeand Murakami (1981) and factor proposed by Bruscato (1970),are widely used. These factors are given by:

The Figures 27 and 28 show relationship between increaseof FATT and J factor and factor at 750.2�F (399�C) for a3.5%NiCrMoV steel.



199

m

Figure 27. Correlation Between Compositional Factor “J” and theShift in FATT of NiCrMoV Steels Following Exposure at 650�F and750�F for 8800 Hours.

Figure 28. Relationship Between Increase of FATT and .

In the high temperature regions, creep and creep rupture are aconcern. The traditional approach is to use a Larson-Miller plot asshown in Figure 29. The design stresses are generally based onthe 105 hours smooth bar creep rupture stress divided by someappropriate safety factors.

Figure 29. Larson-Miller Stress Rupture Curve for 1Cr 1Mo ¼VRotor Steel.

Steam Turbine Blades (Buckets)

There are basically three groups of steam turbine blade materialused by turbine manufacturers. These are various grades of 12 to 13percent chromium (cr) steels with additions of Mo, W, Cb, and V,higher chromium precipitation hardening steels such as 17-4PH andtitanium alloys. Table 8 gives a listing of the commercial availablematerials used in blading (ASTM A1028, 2003). Additional dataregarding the heat treatment and mechanical properties can befound in Aerospace Structural Metals Handbook: Volume 2 (1988),and Briggs and Parker (1965) (Figures 30 and 31).

Table 8. Composition and Mechanical Properties of CommercialBlading Materials.

Figure 30. Rotational Bending Fatigue Behavior of Type 403 atVarious Temperatures.

Figure 31. Stress-Range Diagrams at Various Temperatures forType 403 Heat Treated to 26-32 HRC.


In designing blades one must take into account the materialproperties, the operating conditions, and the quality/purity of thesteam. The principal failure mechanisms that occur in blades arehigh cyclic failure and creep rupture. Blades are designed toprevent creep rupture failures; therefore, it is rare that steamturbine blades fail by this mechanism. In general, most bladefailures are due to high cycle fatigue caused by a number offactors. Some of the factors are listed below:

• Dynamic stresses caused by nonsteady steam forces, nozzlewakes, thermal transient per revolution diaphragm harmonics, andflow instabilities

• Strong exciting harmonics of rotational speed such as the nozzlepassing frequency

• Steam purity and its effect on corrosion in fatigue and pittingcorrosion

Speidel (1981) highlighted the satisfactory corrosion fatigueresistance of X21CrMoV121 steel in a good purity steam(Figure 32). However, the growth of fatigue cracks in 12percent chromium steels is greatly enhanced by the presence ofchloride solutions at low cyclic stress intensities and high meanloads (Figure 33). Speidel (1981) also illustrated how the �Kth(threshold stress intensity) is greatly reduced by the presence ofcertain environments (Figure 34). Other work done by Batteand Murphy (1981) showed similar results on the fatiguestrength reduction in various aqueous solutions (Figure 35).With these data, it can be concluded that the most concerningconditions for fatigue crack growth is a high mean stress withsuperimposed cyclic stresses in the presence of aqueoussolutions. These conditions are the service conditions thatsteam turbine materials are subjected to. It is well known thatshot peening enhances the fatigue resistance of metallicmaterials by introducing compressive stresses at the surfacelayer of the component. However, in a corrosive environment,pits may penetrate the compressive stress layer, therebynegating the benefit gained by peening (Figure 36). Theimportance of steam purity and corrosion resistance of theblade material are clear when considering that it is not feasiblein most cases to avoid the stress state and exposure to aqueoussolutions present in the turbine and that treatments such aspeening are not reliable.

Figure 32. Fatigue and Corrosion Fatigue of X21CrMoV 121 inAir, Deaerated Water, and Aerated Hot Chloride Solutions.

Figure 33. Effect of Chloride Solutions on the Fatigue CrackGrowth Rate of 12%Cr Steels.

Figure 34. Effect of Environment on the Threshold Stress IntensityKth of 12%Cr Steels.

Figure 35. Effect ofVariousAqueous Environments on Fatigue Strength.



Figure 36. Effect of Environment on Fatigue Resistance of ShotPeened 12%Cr Steel.

Casings

The various composition of steels used for turbine casings areshown in Table 9. During the years the drive toward improving creepstrength to accommodate the steadily increasing temperatures led toprogressive changes in material from the C-½ Mo and 1Cr½Mo tothe 1Cr1Mo¼V and 2¼Cr1Mo. In the late 1960s and early 1970s,numerous instances of reheat cracking (stress relief cracking) inthe weld heat affected zones occurred in the 1Cr1Mo¼V and theimportance of rupture ductility was realized. The material½Cr½Mo¼V was standardized by some manufacturers whoimplemented stringent specifications relating to control of residualelements (particularly phosphorus, antimony, tin, copper, aluminum,and sulfur), deoxidation practices, and welding procedures. Casingdesigns were modified to eliminate manufacturing and in-servicereheat cracking. Other manufacturers utilized the 2¼Cr1Mo steeldue to its higher creep ductility, higher low cycle fatigue resistance,and better weldability. The current designs use either ½Cr½Mo¼Vor the 2¼Cr1Mo steel material. For steam temperatures of 1050�Fand above, and up to 1100�F, the 9 to 12%Cr casting grade steelswill be utilized.

Table 9. Various Composition of Steels Used for Turbine Casings.

The optimization of low alloy materials has taken place over aperiod of 50 years or more. The development and optimization of the9 to 12%Cr have only just begun. A lot of work remains to be doneto characterize these materials for design lives of up to 250,000hours and to cast variation in their properties. The development ofjoints between the modified 9%CrMo and low alloy material hasbeen done in order to optimize selection of welding consumables,welding parameters, and post weld heat treatment. Creep tests onsuch joints have indicated that in long-term tests, their rupturestrength falls near the lower bound of the low alloy parent material’sstrength. Due to composition gradients, carbon depleted, ferriticzones can form on the low alloy side of the interface between thehigh and low alloy materials. Consequently, joint locations aredesigned such that the temperature and stress are lower so that creepand carbon diffusion in service will be very limited.

Nozzles and Diaphragms

In steam turbines, stationary nozzles and diaphragms are selectedbased upon stress/temperature, oxidation, and corrosion. The bladesof nozzles and diaphragms are made from wrought 13%Cr seriesstainless material. The blade holders are manufactured from mildsteel, ductile iron, or low alloy steels. The choice of materialdepends on the method of construction and the operating designstress at temperature.During service, erosion due to wet steam can occur on the latter

stages of the steam turbine diaphragms. Normally repairs in thoseareas are done by depositing weld material that has improvedwater/steam erosion resistance, i.e., AWS E309/ER309 orInconel® weld consumable. In some instances, a mechanical fixmay be performed using an austenitic stainless or nickel basewrought material.

High Temperature Bolting

In a steam turbine, bolts in flange joints operating in the creeprange at temperatures up to 1050�F must be able to withstandsteam pressures up to 2 ksi or in some cases even higher. The mainrequirement of bolts in the creep range is to maintain joints withoutrelaxing below the required design stress limit, which may allowleakage. Consequently, the important property for bolts is the stressrelaxation characteristics of the material. For a given joint, the loadrequired to exceed the steam load is applied to the flange area bytensile loading of the bolts. During service, creep in bolt causesrelaxation of this initial load. The elastic strains produced by initialtightening of the bolts are progressively converted to creep strains;elongating the bolt and thereby, reducing the effective load in thejoint. For design, the final relaxation stresses must be in excess ofthe design stress to keep the joint tight. Depending on the materialand duty requirements, bolts are usually tightened to a predefinedcold strain, e.g., 0.15 percent from which the initial stress on thebolt can be calculated. Consequently, the bolts need to retain theirdesign stress between each overhaul when the bolts are removedfor maintenance of the machine and retightened on reassembly.Typical composition of bolt steel, their properties, and various

national standards are summarized in Tables 10 and 11 (Everson,et al., 1988). For convenience, the Central Electricity QuenchingBoard of the United Kingdom has classified these compositionsinto grouping numbered one through eight. Several of thesegroups of materials show stress relaxation behavior after 30,000hours as a function of temperature when subjected to a coldprestrain of 0.15 percent (Figure 37) (Branch, et al., 1973). Inselection of bolt material, the compatibility of the thermalexpansion coefficient of the bolts with respect to the jointmaterials, as well as its susceptibility to various fracturemechanisms, must be taken into account. When sufficient stressrelaxation data are not available, one can apply the usefulcorrelations that exist between the creep rupture data andrelaxation stress at 0.15 percent strain (Figure 38).

Table 10. Various National Bolting Material Specifications.

Table 11. United Kingdom Bolting Material Specifications.


Figure 37. Comparison of Stress-Relaxation Behavior after 30,000Hours as a Function of Temperature for Various Bolt MaterialsSubjected to a Cold Prestrain of 0.15 Percent.

Figure 38. Relationship Between Relaxed Stress and RuptureStrength at the Same Duration for Times from 1000 to 30,000Hours and Temperature from 885 to 1110�F (475 to 600�C).

Fracture of a bolt can occur if the local creep strain reaches thecreep ductility of the material from which the bolt was made.Consequently, low rupture ductilities lead to notch sensitive boltmaterial. Numerous failures of 1Cr-1Mo-V steel bolts were attributedto notch sensitive failures. By reducing the impurity elements of thebolt steel, appreciable improvements have been made to ruptureductility without compromising the creep strength (Figure 39). Theresidual element content can be reduced by careful scrap selection,avoidance of air melt, and use of double vacuum melting (Figure 40).

Figure 39. Relationship Between Residual Element Content andDuctility.

Figure 40. Effect of Double Vacuum Melting on Stress-RuptureProperties of 1Cr-Mo-V Steels at 1020�F.

Other improvements in notch ductility were in a new class ofCrMoV containing titanium and boron with subsequent grainrefinement. Further improvements in the rupture ductility of1CrMoVTiB steel have been made by reducing the major embrittlingelements. The application of very clean steel practices have shownbeneficial effects on ductility and notch strength. This material willcontinue to be used as a very cost-effective option for temperaturesup to 1049�F (565�C). Also nickel-based super alloys such asNimonic 80A and IN901 have been used for some of the hightemperature bolts where stress relaxation is an issue.

REPAIRS

In the rotating equipment industry, service of components requireseither replacement with new or refurbishment of the components.The refurbishment can mean restore the component to its originaldimension either by welding/mechanical fix or other sprayed/platingprocesses. All OEMs have developed repair procedures for most ofthe long lead delivery components, e.g., components such asimpeller, shaft, casing, and diaphragms. API RP 687 (2001) specifiesthe minimum requirements for performing the repairs to rotors.Throughout the last 20 years hundreds of rotors and shafts for

both compressors and steam turbines have been successfully weldrepaired by various OEMs. Welded rotor restoration, like any othercritical repair technology, requires a highly analytical approach toassure component and machine reliability. Most OEMs developproperty data for weldments and heat affected zones for the variousmaterials. This would include materials for both steam turbines andcentrifugal compressors. The data that are generated, but are notlimited, are as follows:

• Room and elevated temperature tensile properties

• Impact and fracture appearance transition temperature data• Creep/stress rupture data• Fatigue properties• Stress corrosion cracking (SCC) threshold limits

Several papers (LaFave, 1991; Dowson, 1995; Dowson andWiegand, 1996) outline some of the material property data that canbe generated. Figures 41, 42, and 43 show some generic examplesof data that are typically used. These data allow for the weld metalproperties to be compared against mechanical analysis results andcompany design criteria to assure that the proposed repair will bereliable. Shorter term testing such as weldability, tensile hardness,and toughness is routinely done on the base metal, weld metal, andHAZ for each engineered rotor repair.



Figure 41. Typical Larson-Miller Curve Generated from VariousWeldments.

Figure 42. Typical S-N Curve from Fatigue Testing of Weldments,Fully Reversed (R = �1), Endurance Limit Set to 108.

Figure 43. Stress Corrosion Crack Growth Rate as a Function ofStress Intensity of Weld Metal, HAZ, and Base Metal.

When repairs to impellers are required, it is generally due tosome form of mechanical damage or corrosion/erosion attack. Ifthe impeller has failed by fatigue, the component is generallyreplaced. When corrosion attack has occurred, repairs will bemade by applying corrosive resistant welding consumables. Forerosion attack, various coatings can be applied to extend the lifeof the impeller.

COATINGS

Despite best efforts in choosing a suitable alloy, it is frequentlyimpossible to find a single material that is ideal for the particularapplication. In this case, composite systems, i.e., coatings, may beused to exploit the properties of two or more materials. There areseveral reasons why coatings may be used, ranging from enhancinga particular property for a specific application to reducing cost byallowing for the use of less expensive substrate materials. Coatingsare also often designed to be multifunctional, addressing multipleproblem areas simultaneously. The petrochemical industry presentsseveral opportunities for the usage of coatings due to the frequentlyharsh conditions encountered in this service.

Compressors

Corrosion

Corrosion is a common problem in several industries, includingturbomachinery. It is an electrochemical attack that can occur as aneven attack of the surface (general corrosion) or uneven attack(pitting), as well as lead to stress corrosion cracking. Coatings forcorrosion protection work by acting as a barrier that separates thesubstrate material and the environment. Typically, the coatingmaterial is viewed as a sacrificial layer and has slower reactionkinetics than the base metal in the particular environment.For selection of materials for sour gas service, API 617 (2002)

refers to NACE MR0103 (2005) and MR0175 (2003), whichallow the use of coatings for general corrosion resistance, but notfor protection against stress corrosion-cracking. This is due to thefact that localized defects in the coating can lead to a stresscorrosion crack. With any coating, there is always a risk that thecoating will be removed at a small location, either by erosion,localized pitting, or foreign object damage, which will expose thesubstrate to the corrosive environment. Diffusion coatings are toothin to provide a complete barrier between the substrate and theenvironment. Metallic, electrolytic coatings contain microscopiccracks that will allow the H2S to contact the substrate material.Hydrogen sulfide can diffuse through polymer coatings to reachthe substrate while thermal spray coatings are porous and will notprovide adequate protection.There appears to be a limited number of coatings advertised by

the various OEMs that are applied solely for corrosion protection.Part of this is due to the fact that changing the base metal to a morecorrosion resistant material, such as a stainless steel, often givessatisfactory results. A second reason for the lack of corrosioncoatings is that corrosion is often accompanied by fouling incentrifugal compressors and therefore the antifoulant coatings,discussed below in more detail, typically are designed to preventcorrosion as well. One coating that the authors’ company has usedwith some success is a baked phenolic coating. This coating is usedin certain applications where it performs better than stainless steelor where the increased price of using stainless steel cannot bejustified by the end user.

Fouling

Fouling is a common problem in compressors and to someextent, steam turbines. Fouling refers to the build of solids,usually polymeric materials, on the internal aerodynamic surfacesof the machine. While it does not usually lead to catastrophicfailure, it does gradually reduce the efficiency of the machine byincreasing the mass of the rotor, altering the aerodynamics, andblocking flow paths. If left unchecked, fouling can block the flowpath to the extent that production is stopped or cause imbalancesthat can damage the machine. Depending on the service, foulingsubstances may come from outside of the machine or begenerated internally. External foulants may come from airbornesalt, submicron dirt, and organic or inorganic pollutants in theprocess gas. A well-maintained filtration system usually helps tominimize this type of fouling (Meher-Homji, et al., 1989; Guineeand Lamza, 1995). In petrochemical compressors, the situationis much more complicated, as the foulants can be generatedinternally. For example, in ethylene cracked gas compression,fouling results from the polymerization reactions intrinsic to thecompression process. Fouling has imposed significant cost onpetrochemical production.A material that is required to resist fouling must have excellent

release properties. Materials with a combination of low coefficient offriction and chemical inertness are usually used for this application.A common and widely known coating material for centrifugalcompressors is polytetrafluoroethylene (PTFE [Teflon®]). These aremulticomponent, sprayed coatings designed for fouling andcorrosion resistance.Wang, et al. (2003), showed a dramatic decrease


in time required to release an applied foulant on samples coated withtwo PTFE type coatings offered by the author’s company versus baresteel samples (compare E and E+ versus steel in Figure 44). Figure45 shows an example of a compressor rotor with PTFE coatedimpellers. Unfortunately, PTFE coatings are removed by erosiveliquids (example water washing) or solids. Electroless nickel (EN)has also been shown by Dowson (2007) to exhibit excellent releaseproperties (523 in Figure 44), while remaining adherent in erosiveconditions. In fact, the release properties are as good as or better thanresults from PTFE. EN is applied by submerging the part into a Niand P containing solution where an autocatalytic process plates thepart with a well-bonded, amorphous Ni-P alloy. The P in the alloy isbelieved to be responsible for the release properties, whileamorphous nature aids in corrosion resistance.

Figure 44. Comparison of Fouling-Release Performance of BareSteel and Coated Panels.

Figure 45. Centrifugal Compressor Rotor Coated with a PTFEType Coating.

Case Study of Electroless Nickel Coating

The application of electroless nickel as an antifoulant coatingis a relatively recent advance. A brief survey of antifoulantcoating offerings by OEMs shows that the majority do not offer anelectroless nickel coating. Due to the current lack of offerings, anupdate of a case study initially introduced by Wang, et al. (2003),is presented below in order to justify the use of electroless nickelas a corrosion and foulant resistant coating option.

Background

A major chemical plant in Corunna has a centrifugal compressorthat used to be coated with the coating 3P for antifouling. The coatinghad suffered severe deterioration two times during the four yearoperation since 1997. Analysis indicated that the deterioration wasrelated to the heavy washing injection, which contained aggressivechemical additives, and steam-cleaning operation. However, thewashing and cleaning were essential to the plant because of theextensive fouling and efficiency drop. In order to withstand theinjections, the compressor rotor and diaphragms were recoated withEN during a plant turnaround in September 2001. The compressorhad been in service until 2006 with satisfactory performance.

Operating Status of the Electroless Nickel Coating

The compressor with EN coating had been successfully operatedsince the 2001 turnaround until 2006. The rotor vibration has beenmonitored at a much lower level than the previous run periods,as shown in Figure 46, which indicates that the fouling has beeneffectively controlled by the EN coating. This improvement isattributed to the ability of EN to withstand the heavy washingoperation as compared to the 3P coating. The oil washing andchemical injections have been kept in the same manner as in theprevious operating periods. The Corunna plant feels that the EN hassuccessfully functioned as an antifouling coating. When the unit wasreplaced in 2006 with a new machine, the customer requested thesame successful EN coating to be applied to the rotor and internals.

Figure 46. Vibration Data Before and After Application ofElectroless Nickel to the Rotor.

After the original unit was pulled out of service, examination of therotor shows that the EN coating remained in tact after the five yearoperational period. Pictures of a coated impeller are given in Figures47 and 48. The coated impeller is free of corrosion and foulantbuildup, and the impeller serial number remains readable. This isproof of the remarkable antifoulant properties of the EN coating.

Figure 47. Picture of First Stage Impeller with Electroless NickelCoating after 5 Years of Service. The Coating Remains Intact andthe Serial Number Is Clearly Visible.



205

Figure 48. Picture of First Stage Impeller with Electroless NickelCoating after 5 Years of Service. The Gas Path of the Impeller HasNo Foulant Buildup.

Erosion

Erosion can be a significant issue in some applications ofcentrifugal compressors. It can occur as solid particle or liquiddroplet erosion. Solid particle erosion is usually caused by externalcontaminants, such as those mentioned above. Liquid dropleterosion can be due to the condensation during compression orintentional injection of liquid for cleaning. The particle or dropletis accelerated by the carrier gas. The resultant impact upon themetallic compressor components removes small amounts ofmetal; creating pits and microcracks at the surface. Typically, thecompressor design is robust and the material removal rate is suchthat erosion on its own is not a major problem. Problems arisewhen erosion is combined with other factors such as corrosion andcyclic stresses. In corrosive environments, erosion can compromiseprotection schemes and cause an accelerated corrosion attack. Anexample of this was mentioned above, where erosion can removesome antifoulant coatings. When cyclic stresses are present, themicrocracks created can easily serve as initiation sites for fatiguecracking. Figure 49 shows the leading edge and fracture surface ofa compressor impeller blade that failed by fatigue initiated bydamage from liquid droplet erosion.

Figure 49. Stereomicrograph Showing a Centrifugal CompressorImpeller Blade that Failed Due to Liquid Droplet Erosion.

Erosion can often be thought of as a type of wear; therefore,similar coatings are used to prevent both. Coatings used for erosionprotection are usually hard coatings, such as tungsten or chromiumcarbide. Any surface exposed to the gas path should be coated inapplications where erosion is a problem. One problem that is anissue with all coatings for compressors, but is particularly true for

these hard, erosion coatings, is the method of application. Carbidecoatings are typically applied using techniques such as highvelocity oxyfuel (HVOF), plasma spraying, or detonation gun.These processes have certain requirements, such as spray distanceor line-of-sight, that may not be possible for the gas paths ofsmaller, closed impellers or other parts with poor accessibility.

Steam Turbines

Corrosion

Under ideal conditions, corrosion is not a major issue in steamturbines. Stainless steel materials perform extremely well withoutcoatings in the elevated temperature, steam environment. Thecorrosion product formed on the stainless steel is usually a thin,uniform layer that grows slowly with time. Corrosion becomes aproblem when there is an underlying problem with the process.Steam purity is the most common problem. Impure steam carrieswith it elements that increase the corrosion process. Typically, theamount of impurities in the steam is not large enough to change thegeneral corrosion rate; rather they cause localized attack that leadsto pitting. These pits provide easy initiation sites for fatigue cracks,as shown in Figure 50. Figure 51 is a plot of stress amplitude versuscycles to failure by fatigue. The drastic reduction in stressamplitude required for a given fatigue lifetime illustrates the easein which fatigue cracks form in corrosion pits.

Figure 50. Micrograph Showing a Crack Initiating from aCorrosion Pit.

Figure 51. Plot Showing the Effect of Corrosion and Pitting onFatigue Life.

Similar to compressors, there are not many coatings applied tosteam turbine components for corrosion only. If there are problemswith corrosion, usually other problems are occurring, such aserosion, which also must be addressed by a coating. The authors’company has used chromium diffusion coatings in the past wherepitting corrosion was a problem.

Erosion

Erosion is a serious problem in steam turbines. The gas path ofa steam turbine is much more closed than that found in a centrifugalcompressor, providing more area for erosive media to impact.


Erosion-corrosion mechanisms are more prevalent due to thehigher temperatures. The chances of cyclic stresses are also veryhigh due to the complicated stress state found in turbine blades.


The erosion occurs on the blade-vane leading edges caused by theexfoliation of scales from the boiler tubes mainly during transientconditions. For rotating blades, sprayed Cr3C2 coatings are appliedusing processes such as detonation gun, plasma, and high velocityoxygen fuel processes to protect against the scales. Other processessuch as diffused boride coatings have successfully been applied tostationary nozzles. Figure 52 shows a photomicrograph of boridecoating on AISI 422. The boride diffusion coatings applied by packcementation and the Cr3C2 coatings applied by the spray processesmentioned above will continue to be the industry best choices.

Figure 52. Cross-Section of a Boride Diffusion Coating on AISI 422.

Liquid Droplet Erosion

The damage of rotor blades in the latter low pressure stages ofsteam turbines by condensed water droplets can be a problem thathas troubled designers and operators for many years. The damageconsists of removal of material from the leading edge and adjacentconvex surfaces of the moving blade. It is related to the wetnessconditions in the low pressure regions and the velocity with whichthe surface of the blade strikes the water droplets. Stresses producedby impact of drops have been calculated by existing theories to besufficiently high to initiate damage in the latter stages of blades ina steam turbine. Turbine experience and laboratory testing haveshown that erosion rate is time dependent with three successivezones: a primary zone in which damage is initiated at slip planeswith little or no weight loss, a secondary zone where the rate risesto a maximum, and finally a tertiary zone where the rate diminishesto a steady-state value. Moreover, it is this tertiary region that isimportant to designer and operator alike rather than the initial andsecondary zones since it is this region that the turbine erosionshields operate for most of their lives (Figures 53 and 54).

Figure 53. Erosion Rate of 630 DPH, 18W-6Cr-0.7C Tool SteelComparator Specimens.

Figure 54. Images Showing Erosion of Blades from the SameTurbine after 400 Hours and 70,000 Hours.

While not strictly coatings, various countermeasures have beenutilized by OEMs, including water drainage devices in a cylinderwall, flame or induction hardening of the leading edge of the blade,and applying stellite or tool steel strips to the leading edge of ablade by welding or brazing. All of the above methods show somedegree of success, with the stellite material providing the betterperformance in the more stringent water droplet environment.

Fouling

Fouling and corrosion can also be a problem in steam turbinesnot only causing material damage but also can gradually reduce theefficiency of the turbine. Industrial turbines, whether condensingor noncondensing, can encounter problems with deposits buildingup on the turbine airfoils. In a turbine, hydroscopic salts, such assodium hydroxide, can absorb moisture when superheated steambecomes saturated and condenses in the latter stages of theturbine/Wilson line. Wet sodium hydroxide has a tendency toadhere to turbine metal surfaces and can entrap other impuritiessuch as silica, metal oxides, and phosphates. Once these depositshave formed they can be difficult to remove. Build up of thesedeposits may be a cause of decrease in efficiency and possibly anincrease in vibration. A smooth clean steam path will not collectdeposits so easily as a dirty, previously contaminated surface.Consequently, a previously contaminated turbine will accumulatedeposits more rapidly than a clean one. Therefore, it is desirableto prevent further deposit buildup and to remove the problemsassociated with the presence of the deposits by cleaning theturbine. The authors’ company has provided support to end userturbines for water washing of steam turbines (Watson, et al., 1995).The effectiveness of the water removal procedures mainly dependson the adherence of the deposits to the substrate.A second route is to coat the surface with a material that has

superior antifouling or antistick/corrosion characteristics. This inturn is beneficial to the turbine blades by reducing the tendency forcontaminants to stick to the blades and increase the effectiveness ofthe water washing. Titanium nitride coatings with a chromiumundercoat (Cr-TiN) have also been used by steam turbine OEMs tocoat turbine blades. This coating provides corrosion protection andcan be used on all stages of a steam turbine rotor, however, theCr-TiN only provides limited antifoulant benefits. The authors’company has recently developed a proprietary coating, which isa corrosion resistant antifoulant coating designed for the laterstages of the turbine rotor (where the deposit buildup is mostsevere). Testing has shown that this coating provides significantimprovement in foulant release ability (Figure 55), excellentcorrosion protection (passes over 1000 hours of ASTM B117corrosion testing under a 5 percent salt solution), and erosionprotection (Figure 56), and, while having little effect on thefatigue properties (Figure 57).



Figure 55. Comparison of Foulant Release Performance of BareSteel Against Proprietary Coating and Cr-TiN Coated Samples.

Figure 56. Comparison of Bare AISI 403 Stainless Steel AgainstProprietary and Cr-TiN Coated Samples after 10 Hours ofModified ASTM G32 Testing.

Figure 57. Results of R.R. Moore Fatigue Testing.

SUMMARY

An overview of materials and material related processes hasbeen presented for centrifugal compressors and steam turbines.Special attention has been given to address some of the problemsassociated with material selection for the various components andthe steps taken to prevent and/or minimize reoccurrence.What will the future bring to materials/or materials related

processes? The application of composite materials for rotatingcomponents in compressors may be seen in the not so distantfuture. The application of refined existing processes to manufacturecomponents to near net shape such as P/M net shape, HIP process,or metal rapid prototyping based on laser microwelding of metallicpowders may also be seen.

ACKNOWLEDGEMENT

The authors are grateful for the support from the MaterialsEngineering Department at Elliott Company and recognize ElliottCompany for permission to publish this paper.

REFERENCES

API RP 687, 2001, “Rotor Repair,” American Petroleum Institute,Washington, D.C.


Aerospace Structural Metals Handbook: Volume 2, 1988, Departmentof Defense—Materials and Ceramics Information Center.

Armor, A. F., Jaffee, R. I., and Hottenstine, R. D., 1984, “AdvancedSupercritical Power Plants—The EPRI Development Program,”Proceedings of the American Power Conference, 46, p. 70.

Arlt, N., Gumpel, P., and Sahni, P., 1988, “Uber Einige Eigenschaftenvon Nichtrostenden Nickelmartensitischen Chromstahlen,”Thyssen Edeist, Techn. Ber. 14, Band Heft 1, p. 71.

ASME Boiler & Pressure Vessel Code - Section VIII, 2007, AmericanSociety of Mechanical Engineers, NewYork, NewYork, pp. 66-67.

Atkins, M., 1980, Atlas of Continuous Cooling TransformationDiagrams for Engineering Steels, Revised U.S. Edition,Copyright by the American Society for Metals, pp. 168 and 188.

Batte, A. D. and Murphy, M. C., September 1981, “The CorrosionFatigue of 12%Cr Blade Steels in Low Pressure Steam TurbineEnvironments,” Corrosion Fatigue of Steam Turbine BladeMaterials,Workshop Proceedings, Palo Alto, California.

Branch, G. D., et al., 1973, “High Temperature Bolts for SteamPower Plant,” International Conference on Creep and Fatiguein Elevated Temperature Applications, Sheffield andPhiladelphia, Institute of Mechanical Engineers, London,Conference Publication 13, pp. 192.1-192.9.

Briggs, J. Z. and Parker, T. D., 1965, “The Super 12% Cr Steel,”Climax Molybdenum Co., NewYork, NewYork.

Bruscato, R. M., 1970, “Temper Embrittlement and CreepEmbrittlement of 2� Cr-1 Mo Shielded Metal-Arc WeldDeposits,” Welding Journal, 35, p. 148s.

Cameron, J. A., 1989, “Materials for Centrifugal Compressors—AProgress Report,” Proceedings of the Eighteenth TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&M University,College Station, Texas, pp. 9-22.

Cameron, J. A., and Danowski, F. M., Jr., 1973, “SomeMetallurgical Considerations in Centrifugal Compressors,”Proceedings of the Second Turbomachinery Symposium, GasTurbine Laboratories, Texas A&M University, College Station,Texas, pp. 116-128.

Dowson, P., 1995, “Fracture Mechanics Methodology Applied toRotating Components of Steam Turbine and CentrifugalCompressor Rotors,” Third International Charles Parson’sConference, Tyne, United Kingdom.

Dowson, P., 2002, “Development of 13Cr4Ni Material forBrazed/Welded Impellers for H2S and LowTemperature Service,”Supermartensitic Stainless Steels 2002, Brussels, Belgium.

Dowson, P., 2007, “Antifouling and Corrosion Resistant Coatingsfor Cracked Gas Compressors and Steam Turbines,” AICHESpring National Meeting, Houston, Texas.

Dowson, P., andWiegand, R., May 1996, “Welded Rotor Restoration—Technical Justification & Fit for Service Assurance,” SecondInternational EPRI Conference and Vendor Exposition.


Dowson, P., Ross, S. L., and Schuster, C., 1991, “The Investigationof Suitability of Abradable Seal Materials for Application inCentrifugal Compressors and Steam Turbines,” Proceedings ofthe Twentieth Turbomachinery Symposium, TurbomachineryLaboratory, Texas A&M University, College Station, Texas,pp. 77-90.

Dowson, P., Walker, M., and Watson, A., 2004, “Development ofAbradable and Rub Tolerant Seal Materials for Application inCentrifugal Compressors and Steam Turbines,” Proceedings ofthe Thirty-Third Turbomachinery Symposium, TurbomachineryLaboratory, Texas A&M University, College Station, Texas,pp. 95-102.

Dowson, P., Wang, W., and Alija, A., 2005, “Remaining LifeAssessment of Steam Turbine and Hot Gas ExpanderComponents,” Proceedings of the Thirty-Fourth TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas, pp. 77-92.

Engineered Materials Handbook: Composites, Volume 1, 1987,American Society of Metals, Materials Park, Ohio.

Everson, H., Orr, J., and Dulieu, D., 1988, “Low Alloy FerriticBolting Steels for Steam Turbine Applications: The Evolutionof Durehete Steels,” in Advances in Materials Technology forFossil Power Plants, R. Viswanathan and R. I. Jaffee, Editors,American Society for Metals, Metals Park, Ohio.

Fidler, F., 1971, “Metallurgical Considerations in Wire Wool TypeWear Bearing Phenomena,” Wear, 17, pp. 1-20.

Gold, M. and Jaffee, R. I., 1984, “Materials for Advanced SteamCycles,” ASM Journal of Materials for Energy Systems, 6 (2),pp. 130-145.

Gross, J. H. and Stout, R. D., April 1957, “Ductility and EnergyRelations in Charpy Tests of Structural Steels,” WeldingResearch Supplement, 22, pp. 151s-159s.

Guinee, M. J. and Lamza, E. W., 1995, “Cost Effective Methods toMaintain Gas Production by the Reduction of Fouling inCentrifugal Compressors,” SPE30400, pp. 341-350.

Jones, G. T., 1972, Proceedings Institute of Mechanical Engineers,186, p. 31-32.

Keller, H. F. and Cameron, J. A., 1974, “Laboratory Evaluation ofSusceptibility to Sulfide Cracking,” Carrier Corporation,NACE Paper #99.

Kohut, G. B. and McGuire, W. J., July 1968, “Sulfide StressCracking Causes Failure of Compressor Components inRefinery Service,” Materials Protection, 7, pp. 17-22.

LaFave, R. A., 1991, “Submerged Arc Weld Restoration of SteamTurbine Rotors Using Specialized Welding Techniques,”Proceedings of the Twentieth Turbomachinery Symposium,Turbomachinery Laboratory, TexasA&MUniversity, pp. 19-34.

Lukens Fineline Steels, Form No. 675 4-85., Lukens Steel Company,Coatesville, Pennsylvania.

Meher-Homji, C. B., Focke, A. B., and Wooldridge, M. B., 1989,“Fouling of Axial Flow Compressors—Causes, Effects,Detection, and Control,” Proceedings of the EighteenthTurbomachinery Symposium, Turbomachinery Laboratory,Texas A&M University, College Station, Texas, pp. 55-76.

Metals Handbook, Ninth Edition: Volume 4, 1981, AmericanSociety of Metals, Materials Park, Ohio, p. 35 and p. 43.

Moller, G. E., 1968, “Corrosion, Metallurgical and MechanicalExperiences of Petroleum Refinery Compressors,” NACETaskGroup T-8-1 Interim Report.

NACE Standard MR0103, 2005, “Materials Resistant to SulfideStress Cracking in Corrosive Petroleum Refining Environments,”NACE International, Houston, Texas.

NACE Standard MR0175, 2003, “Petroleum and Natural GasIndustries—Materials for Use in H2S-Containing Environmentsin Oil and Gas,” NACE International, Houston, Texas.

NACE Standard RP0472, 2005, “Methods and Controls to PreventIn-Service Environmental Cracking of Carbon SteelWeldments in Corrosive Petroleum Refining Environments,”NACE International, Houston, Texas.

Newhouse, D. L., et al., 1972, “Temper Embrittlement of AlloySteels,” ASTM STP 499, pp. 3-36.

Newhouse, D. L., July 1987, “Guide to 12Cr Steels for High-andIntermediate-Pressure Turbine Rotors for the AdvancedCoal-Fired Steam Plant,” Report CS-5277, Electric PowerResearch Institute, Palo Alto, California.

Schwant, R. C. and Timo, D. P., 1985, Life Assessment of GeneralElectric Large Steam Turbine Rotors, Life Assessment andImprovement of Turbogenerator Rotors for Fossil Plants,Viswanathan, R., Editor, New York, New York: PergamonPress, p. 3-25-3-40.

Speidel, M. O., September 1981, “Corrosion-Fatigue of SteamTurbine Blade Materials,” Corrosion Fatigue of Steam TurbineBlade Materials, Workshop Proceedings, Palo Alto, California.

Treseder, R. S. and Swanson, T. M., 1968, “Factors in SulfideCorrosion Cracking of High Strength Steels,” Corrosion, 24,pp. 31-37.

Viswanathan, R. and Jaffee, R. I., October 1983, “Toughness of Cr-Mo-V Steels for Steam Turbine Rotors,” ASME Journal ofEngineering Material Techniques, 105, pp. 286-294.

Wang, W., Dowson, P., and Baha, A., 2003, “Development of Anti-fouling and Corrosion Resistant Coatings for PetrochemicalCompressors,” Proceedings of the Thirty-Second TurbomachinerySymposium, Turbomachinery Laboratory, Texas A&MUniversity, College Station, Texas, pp. 91-97.

Warren, D. and Beckman, G. W., October 1957, “SulphideCorrosion Cracking of High Strength Bolting Material,”Corrosion, 13, pp. 631t-646t.

Watanabe, J. and Murakami, Y., 1981, “Prevention of TemperEmbrittlement of CrMo Steel Vessels by the Use of Low SiForged Steels,” American Petroleum Institute, Chicago,Illinois, p. 216.

Watson, A. P., Carter, D. R., and Alleyne, C. D., 1995, “CleaningTurbomachinery without Disassembly, Online and Offline,”Proceedings of the Twenty-Fourth Turbomachinery Symposium,Turbomachinery Laboratory, Texas A&M University, CollegeStation, Texas, pp. 117-128.



209

Otakar Jonas is a Consultant with Jonas, Inc., in Wilmington,Delaware. He works in the field of industrial and utility steam cyclecorrosion, water and steam chemistry, reliability, and failure analysis.After periods of R&D at Lehigh University and engineeringpractice atWestinghouse SteamTurbine Division, Dr. Jonas startedhis company in 1983. The company is involved in troubleshooting,R&D (EPRI, GE, Alstom), failure analysis, and in the productionof special instruments and sampling systems.Dr. Jonas has a Ph.D. degree (Power Engineering) from the

Czech Technical University. He is a registered ProfessionalEngineer in the States of Delaware and California.

Lee Machemer is a Senior Engineer at Jonas, Inc., inWilmington,Delaware. He has 13 years of experience with industrial and utilitysteam cycle corrosion, steam cycle and water chemistry, andfailure analysis.Mr. Machemer received a B.S. degree (Chemical Engineering)

from the University of Delaware and is a registered ProfessionalEngineer in the State of Delaware.

ABSTRACT

This tutorial paper discusses the basics of corrosion, steamand deposit chemistry, and turbine and steam cycle design andoperation—as they relate to steam turbine problems andproblem solutions.Major steam turbine problems, such as stress corrosion

cracking of rotors and discs, corrosion fatigue of blades, pitting,and flow accelerated corrosion are analyzed, and their rootcauses and solutions discussed. Also covered are: life prediction,inspection, and turbine monitoring. Case histories are describedfor utility and industrial turbines, with descriptions of root causesand engineering solutions.

INTRODUCTION

This tutorial paper discusses steam turbine corrosion and depo-sition problems, their root causes, and solutions. It also reviewsdesign and operation, materials, and steam and deposit chemistry.References are provided at the end of the paper.With an increase of generating capacity and pressure of individ-

ual utility units in the 1960s and 70s, the importance of large steamturbine reliability and efficiency increased. The associated turbinesize increase and design changes (i.e., larger rotors and discs and

longer blades) resulted in increased stresses and vibrationproblems and in the use of higher strength materials (Scegljajev,1983; McCloskey, 2002; Sanders, 2001). Unacceptable failurerates of mostly blades and discs resulted in initiation of numerousprojects to investigate the root causes of the problems (McCloskey,2002; Sanders, 2001; Cotton, 1993; Jonas, 1977, 1985a, 1985c,1987; EPRI, 1981, 1983, 1995, 1997d, 1998a, 2000a, 2000b, 2001,2002b, 2002c; Jonas and Dooley, 1996, 1997; ASME, 1982, 1989;Speidel and Atrens, 1984; Atrens, et al., 1984). Some of theseproblems persist today. Cost of corrosion studies (EPRI, 2001a,Syrett, et al., 2002; Syrett and Gorman, 2003) and statistics (EPRI,1985b, 1997d; NERC, 2002) determined that amelioration ofturbine corrosion is urgently needed. Same problems exist insmaller industrial turbines and the same solutions apply(Scegljajev, 1983; McCloskey, 2002; Sanders, 2001; Cotton, 1993;Jonas, 1985a, 1987; EPRI, 1987a, 1998a; Jonas and Dooley, 1997).The corrosion mechanisms active in turbines (stress corrosioncracking, corrosion fatigue, pitting, flow-accelerated corrosion)are shown in Figure 1.

Figure 1. Corrosion Mechanisms Active in Steam Turbines.

Purpose, Design, and Operation of Steam Turbines

The steam turbine is the simplest and most efficient engine forconverting large amounts of heat energy into mechanical work. Asthe steam expands, it acquires high velocity and exerts force on theturbine blades. Turbines range in size from a few kilowatts for onestage units to 1300 MW for multiple-stage multiple-componentunits comprising high-pressure, intermediate-pressure, and up tothree low-pressure turbines. For mechanical drives, single- anddouble-stage turbines are generally used. Most larger modernturbines are multiple-stage axial flow units. Figure 2 shows atypical tandem-compound turbine with a combined high pressure(HP), intermediate pressure (IP) turbine, and a two-flow low-pressure(LP) turbine. Table 1 (EPRI, 1998a) provides alternate terminologyfor several turbine components.

211

STEAMTURBINE CORROSIONAND DEPOSITS PROBLEMSAND SOLUTIONS

byOtakar Jonas

Consultant

andLee MachemerSenior Engineer

Jonas, Inc.

Wilmington, Delaware

Figure 2. Typical Tandem Compound, Single Reheat, CondensingTurbine. (Courtesy of EPRI, 1998a)

Table 1. Alternate Terminology for Turbine Components.

Steam enters from the main steam lines through stop and controlvalves into the HP section. The first (control) stage is spaced slightlyapart from subsequent stages to allow for stabilization of the flow.After passing through the HP turbine, cold reheat piping carries thesteam to the reheater (if present) and returns in the hot reheat pipingto the integrated HP and IP cylinder to pass through the IP turbinesection. The flow exits the IP turbine through the IP exhaust hoodand then passes through crossover piping to the LP turbine and exitsto the condenser through the LP exhaust. The typical modern steamturbine has a number of extraction points throughout all sectionswhere the steam is used to supply heat to the feedwater heaters.During its expansion through the LP turbine, the steam crosses

the saturation line. The region where condensation begins, termedthe phase transition zone (PTZ) or Wilson line (Cotton, 1993;EPRI, 1997c, 1998a, 2001b), is the location where corrosiondamage has been observed. In single reheat turbines at full load,this zone is usually at the L-1 stage, which is also in the transonicflow region where, at the sonic velocity (Mach = 1), sonic shockwaves can be a source of blade excitation and cyclic stressescausing fatigue or corrosion fatigue (EPRI, 1997c; Jonas, 1994,1997; Stastny, et al., 1997; Petr, et al., 1997).

Design

Because of their long design life, steam turbines go throughlimited prototype testing where the long-term effects of materialdegradation, such as corrosion, creep, and low-cycle fatigue,cannot be fully simulated. In the past, when development was slow,relatively long-term experience was transferred into new products.With new turbine types, larger sizes, new power cycles, and water

treatment practices coming fast in the last 25 years, experience wasshort and limited, and problems developed, which need to becorrected and considered in new designs and redesigns. While theturbine seems to be a simple machine, its design, including designagainst corrosion, is complex.There are five areas of design that affect turbine corrosion:

• Mechanical design (stresses, vibration, stress concentrations,stress intensity factor, frictional damping, benefits of overspeedand heater box testing)

• Physical shape (stress concentration, crevices, obstacles to flow,surface finish, crevices)

• Material selection (maximum yield strength, corrosion properties,material damping, galvanic effects, etc.)

• Flow and thermodynamics (flow excitation of blades, incidenceangle, boundary layer, condensation and moisture, velocity,location of the salt zone, stagnation temperature, interaction ofshock wave with condensation)

• Heat transfer (surface temperature, evaporation of moisture,expansion versus stress, heated crevices)

Recognition of these effects led to a formulation of rudimentarydesign rules. While the mechanical design is well advanced and thematerial behavior is understood, the flow excitation of blades andthe effects of flow and heat transfer on chemical impurities atsurfaces are not fully included in design practices.Selection of some combinations of these design parameters can

lead to undesirable stresses and impurity concentrations thatstimulate corrosion. In addition, some combinations of dissimilarmaterials in contact can produce galvanic corrosion.Design and material improvements and considerations that

reduce turbine corrosion include:

• Welded rotors, large integral rotors, and discs withoutkeyways—eliminates high stresses in disc keyways.

• Replacement of higher strength NiCrMoV discs with lower(yield strength < 130 ksi (896 MPa) strength discs.

• Repair welding of discs and rotors; also with 12%Cr stainlesssteel weld metal.

• Mixed tuned blade rows to reduce random excitation.

• Freestanding and integrally shrouded LP blades without tenoncrevices and with lower stresses.

• Titanium LP blades—corrosion resistant in turbine environmentsexcept for NaOH.

• Lower stress and stress concentrations—increasing resistance toSCC and CF.

• Flow path design using computerized flow dynamics and viscousflow—lower flow induced vibration, which reduces susceptibilityto CF.

• Curved (banana) stationary blades that reduce nozzle passingexcitation.

• New materials for blade pins and bolting—resistant against SCC.

• Flow guides and double-ply expansion bellows—reducesimpurity concentration, better SCC resistance.

• Moisture extraction to improve efficiency and reduce flow-accelerated corrosion (FAC) and water droplet erosion and use ofalloy steels to reduce FAC.

LP Rotor and Discs

There are three types of construction in use for LP rotors:

• Built-up (shrunk-on design) with forged shaft onto which discsare shrunk and keyed,


Machined from one solid piece (most common), or

• The discs welded together to form the rotor (Figure 3).

The rotor and disc construction is governed by the practices ofindividual manufacturers, capabilities of steel mills, cost, and,during the last few decades, by their resistance to SCC. The solidand welded rotors do not have a problem with disc bore SCC. Thethree types, shown in Figure 3, have little effect on the SCC and CFsusceptibilities of the blade attachments.

Figure 3.ThreeTypes of Rotor Construction. (Courtesy of EPRI, 1998a)

LP Blades

Blade and blade path design and material selection influenceblade CF, SCC, pitting, and other forms of damage in many ways(Sanders, 2001; EPRI, 1981, 1997c, 1998a, 2001b; BLADE-ST™,2000). The main effects of the blade design on corrosion, corrosionfatigue strength, stress corrosion cracking susceptibility, and pittingresistance include:

• Vibratory stresses and their frequencies.• Maximum service steady stresses and stress concentrations.

• Flow induced vibration and deposition.

• Mechanical, frictional, and aerodynamic damping.

Rotating LP turbine blades may be “free standing” (notconnected to each other), connected in groups, or all blades in therow may be “continuously” connected by a shroud. Connectionsmade at the blade tip are termed shrouds or shrouding. Shroudsmay be inserted over tenons protruding above the blade tips andthese tenons then riveted down to secure the shrouds, or they mayconsist of integrally forged or machined stubs, which, duringoperation, provide frictional damping of vibration because theytouch (Figure 4). This design also eliminates the tenon-shroudcrevices where corrosive impurities could concentrate. In somecases, long 180 degree shrouds are used or smaller shroudsegments are welded together.

Figure 4. Typical Turbine Shrouds. (Courtesy of EPRI, 1998a)

Blades are connected at the root to the rotor discs by severalconfigurations (Figure 5). There are several types of serratedattachments: the fir tree configuration, which is inserted intoindividual axial slots in the disc; and the T-shape, which is insertedinto a continuous circumferential slot in the disc. The “finger” typeattachment is fitted into circumferential slots in the disc andsecured by axially inserted pins. All of the blade root designs havegeometries that result in higher local stresses at radii and stressconcentrations that promote SCC and CF. The goal of the designshould be to minimize these local tensile stresses.

Figure 5. Types of Blade RootAttachments. (Courtesy of EPRI, 1998a)

The airfoil of blades may be of constant cross-section for shortblades, and twisted for longer ones. The longest blades for the lastfew rows of the LP are twisted to match the aerodynamics atdifferent radii and improve aerodynamic efficiency. The longerblades are usually connected at the point of highest vibrationamplitude to each other by a tie or lashing wire, which reducesvibration of the airfoil. To reduce random excitation, mixed tuningof long rotating blades has been used in which adjacent blades havedifferent resonant frequencies (EPRI, 1998a).Stationary blades in LP stages are typically arranged in

diaphragms of cast or welded construction. In wet stages,diaphragms may be made with hollow blade vanes or other designfeatures as a means of drawing off moisture that would otherwiselead to liquid droplet erosion (EPRI, 2001b). Recently, somestationary blade designs have also been leaned or bowed,improving flow and efficiency and lowering the excitation forceson the downstream rotating blades.

Casings

Turbine casings must contain the steam pressure and maintainsupport and alignment for the internal stationary components.They are designed to withstand temperatures and pressures up tothe maximum steam conditions. Their design has evolved over theyears and casings are now multiple pressure vessels (for example,an inner and outer casing in the HP and IP cylinder, or a triplecasing) allowing smaller pressure drops and thinner wall thickness.These thinner cross-sections allow for a lower temperature gradientacross the casing section and thus lower thermal stresses. The LPcasing may also be a multiple part design with the inner casingcontaining the supports for the diaphragms and the outer casingdirecting the exhaust to the condensers.

Design Recommendations for Corrosion Control

New designs, redesigns, and failed components should bechecked to determine if they meet allowable corrosion-relateddesign specifications and other corrosion related requirements(Jonas, 1985c).

213

•


In new designs and redesigns of turbine components, use shouldbe made of the new design tools, including 3D finite elementstress and vibration analysis and 3D viscous flow analysis, andconsideration of condensation and impurity behavior. Bladeresonance frequencies should be verified by telemetry. To ensure acorrosion-free design, a corrosion engineer and a chemist shouldbe consulted during the design activity.The following should be considered in design of steam turbines:

• Stresses (Jonas, 1985c)—The mechanical design concepts foravoidance of SCC and CF should include an evaluation of the materialcorrosion properties and defects that influence susceptibility to SCCand CF, i.e., threshold stress (σSCC), threshold stress intensity(KISCC and �KthCF), crack growth rate ((da/dt)SCC and (da/dN)CF),corrosion fatigue limit, pitting rate, and pit depth limit. Trueresidual stresses (micro and macro) should also be considered.Because SCC and CF initiate at surfaces, the maximum surfacestresses must be controlled, usually by control of the elastic stressconcentration factor, kt. The stresses should be the lowest in the“salt zone” region where corrosion is most likely.

• Vibratory stresses are rarely accurately known, except whentelemetry on operating turbines is performed. The design approachshould be to minimize flow excitation, tune the blades, providemaximum damping, and perform laboratory and shop stationaryfrequency testing. Heater box overspeed and overspeed testingduring operation are generally beneficial in reducing operatingstresses by local plastic deformation.

• Heat transfer and flow (EPRI, 1997c)—Surface temperatureresulting from heat transfer and flow stagnation should be consideredalong with its effect on thermodynamic conditions of the impuritiesand water film at surfaces (i.e., evaporation of moisture). Floweffects on blade vibration and deposit formation are complex, andthere are over 15 flow blade excitation mechanisms to be considered.

• Flow of moisture—To avoid flow-accelerated corrosion (EPRI,1996; Kleitz, 1994; Jonas, 1985b; Svoboda and Faber, 1984) andwater droplet erosion (Ryzenkov, 2000; Pryakhin, et al., 1984;Rezinskikh, et al., 1993; Sakamoto, et al., 1992; Povarov, et al.,1985; Heyman, 1970, 1979, 1992), the flow velocity of wet steamshould not exceed the allowable velocity specific to the materialsand moisture chemistry. Regions of high turbulence should beavoided or higher chromium steels should be used. Blade pathmoisture can be extracted.

• Crevices—Crevices can act as impurity traps and concentrators,facilitate formation of oxygen concentration cells, and maygenerate high stresses by the oxide growth mechanism. The worstcrevices are those with corrosive impurities and metal temperaturewithin the “salt zone.” Some disc bore and keyway and bladetenon-shroud crevices fall into this category.

• Galvanic effects—When dissimilar materials are coupled together,corrosion of both materials can be affected by the associated shift incorrosion potentials into the stress corrosion cracking (SCC) orpitting regions. The more active of the two materials may suffergalvanic corrosion. In addition, in some environments, thepotential shift could be into the region where one of the coupledmaterials is susceptible to stress corrosion cracking or pitting.

• Inspectability—In designing turbine components, the questionof inspectability should be addressed. In particular, creviceand high stress regions should be reachable using availableinspection techniques.

• Chemical compounds used during machining, cleaning,nondestructive testing (NDT), and other activities—Many differentchemical compounds are used during manufacture, storage, erection,and inspection of turbine components. Some of them containchlorine and sulfur as impurities or as a part of the organic matrix.During thermal decomposition of the residues of these compounds,

hydrochloric, sulfuric, carbonic, and organic acids can form (Jonas,1982). It is recommended that their composition and use be carefullycontrolled to minimize the risk of residual contamination andsubsequent corrosion. For the compounds that can remain on turbinesurfaces during operation, chlorine and sulfur levels should berestricted to low ppm levels in that compound. Of specific concernare: MoS2 (Molylube™) (Turner, 1974; Newman, 1974), Loctite™,thread compounds (Cu, Ni, graphite), and chlorinated solvents.

Materials and Corrosion Data

There is little variation in the materials used for blades, discs, rotors,and turbine cylinders, and only a few major changes have beenintroduced in the last decade. Titanium alloy blades are being slowlyintroduced for the last LP stages, and better melting practices toprovide control of inclusions and trace elements are being evaluatedfor discs and rotors. Table 2 (Jonas, 1985a) lists common materials andthe typical corrosion mechanisms for the various turbine components.

Table 2. LP Turbine Components, Materials, and RelatedCorrosion Mechanisms.

LP rotors are typically constructed of forgings conformingto ASTM A293 (Class 2 to 5) or ASTM A470 (Class 2 to 7),particularly 3.5NiCrMoV. Shrunk-on discs, when used, are madefrom forgings of similar NiCrMoV materials conforming toASTMA294 (Grade B or C), or ASTM 471 (Classes 1 to 3). The strengthand hardness of turbine components must be limited because thestronger and harder materials become very susceptible to SCC andCF (EPRI, 1998a); particularly turbine rotors, discs, and bladescannot be made from high strength materials.The crack propagation rate increases exponentially with yield

strength at high yield strength values and SCC starts being influencedby hydrogen embrittlement. Because of this sensitivity to high yieldstrength, practically all turbine discs, fossil and nuclear, with yieldstrength higher than ~140 ksi (965 MPa) have been replaced withlower strength materials. Figure 6 is a correlation of crack propagationrates versus yield strength for several operating temperatures (Clark,et al., 1981). This type of data has been used to predict the remaininglife and safe inspection interval. There is also an upper temperaturelimit for LP rotor and disc steels, ~650�F (345�C) aimed at avoidingtemper embrittlement (EPRI, 1998a).

Figure 6. Average Crack Growth Rates Versus Yield Strengthfor Several Operating Temperatures for NiCrMoV Disc Steel.(Courtesy of Clark, et al., 1981)


Since the 1930s, most LP turbine blades have been manufacturedfrom a 12%Cr stainless steel—typically types AISI 403, 403-Cb, 410,410-Cb, and 422, depending on the strength required. Types 403 and410 have better SCC and CF resistance than Type 422, an importantcharacteristic for use in the wet stages of the LP turbine. There arenumerous specifically customized versions of these generic materials(Carpenter H-46, Jethete M152 (modified 403), etc.).The precipitation hardened stainless steels designated 17-4 PH,

15-5PH, and PH13-8Mo have been used for some fossil andnuclear LP turbine blades. The composition of 17-4PH is 17percent Cr, 4 percent Ni, and 4 percent Cu. These steels may bedifficult to weld and require post-weld heat treatment. The copperrich zones in the copper bearing stainless steels are often subject toselective dissolution, forming pits filled with corrosion products.These pits can be crack initiation sites.Titanium alloys, primarily Ti-6Al-4V, have been used for turbine

blades since the early 1960s (EPRI, 1984d, 1984e, 1985c). Thereare numerous benefits to using titanium alloy blades, including theability to use longer lasting stage blades, favorable mechanicalproperties in applications involving high stresses at low temperatures,excellent corrosion resistance, and resistance to impact and waterdroplet erosion damage. Drawbacks to titanium include highercost, difficult machining, and low material damping.LP turbine casings are typically constructed of welded and cast

components. Materials acceptable for lower temperatures, such ascarbon steel plate, are used.Considering the typical steam turbine design life of 25 to 40

years and the relatively high stresses, these materials have beenperforming remarkably well. Turbine steels are susceptible to SCCand CF in environments such as caustic, chlorides, acids,hydrogen, carbonate-bicarbonate, carbonate-CO2, and, at higherstresses and strength levels, in pure water and steam.

Corrosion Data

Corrosion data should provide allowable steady and vibratorystresses and stress intensities for defined design life or inspectionintervals. It is suggested that SCC data include threshold stress(σSCC), threshold stress intensity (KISCC), crack growth rate(da/dt), and crack incubation and initiation times. Corrosionfatigue data should include fatigue limits for smooth and notchedsurfaces and proper stress ratios, crack growth data, and corrosionfatigue threshold stress intensities.Examples of the type of data needed are shown in Figures 7 to 9

for the NiCrMoV disc material and in Figure 10 for 12%Cr bladesteel. Properly heat treated 12%Cr blade steel (yield strength 85ksi, 600 MPa) is not susceptible to stress corrosion cracking andstress corrosion data are not needed. The data shown in Figures 7to 10 can be used in turbine disc and blade design in which theallowable stresses and stress intensities should be below thethreshold values for SCC and CF. The use of these data is outlinedin (Jonas, 1985c).

Figure 7. Stress Corrosion Behavior of NiCrMoV Disc Steel VersusYield Strength for “Good” Water and Steam (Compiled fromPublished Data); KISCC—Threshold Stress Intensity, σSCC—Threshold Stress, and da/dtq—Stage 2 Crack Growth Rate.(Courtesy of Jonas, 1985a)

Figure 8. Corrosion Fatigue (Goodman) Diagram for NiCrMoVDisc Steel; Tested to 108 Cycles. (Courtesy of Haas, 1977)

Figure 9. Air Fatigue Strength Reduction of NiCrMoV Disc SteelCaused by Pitting (Courtesy of McIntyre, 1979)—Effects of PitDensity Were Not Investigated.

Figure 10. Corrosion Fatigue (Goodman) Diagram for ThreeStainless Steel Blading Alloys. (Courtesy of Atrens, et al., 1984)


215

The needed data are difficult to obtain because of the long timeneeded for testing and because of the great range of possible serviceenvironments and temperatures. There does not seem to be anyuniversally accepted accelerated test or environment. The KISCC testin hydrogen sulfide gives a reasonable approximation of KISCCfor low alloy steels, and ultrasonic fatigue tests give usable data toa high number of cycles that may be usable for turbine design.As a rule of thumb, the elasticity limit at temperature (usually

between 0.4 and 0.6 of the 0.2 percent yield strength) can beused as a good estimate of the SCC threshold stress for low andmedium strength materials in mildly corrosive environments. Thisis consistent with the oxide film rupture or strain induced crackingtheory of stress corrosion cracking.Shot peening (EPRI, 2001b)—has been used as a means of

reducing high surface tensile stresses. At a sufficiently high shotpeening intensity, a surface layer of residual compressive stressesis produced. One turbine vendor uses shot peening and othersurface treatments extensively and they have almost no SCC ofdiscs and corrosion fatigue (CF) of turbine blades. There is aconcern that in corrosive environments, pits can grow through thecompressive stress layer into the subsurface region with muchhigher tensile stresses.Other surface treatments for protection against corrosion, such

as coatings and electroplating, have been evaluated (EPRI, 1987a,1993a, 2001b; Jonas, 1989) and sometimes used. There are nowseveral suppliers of steam turbine coatings (EPRI, 1987a, 1993a;Jonas, 1989).

Environment—Stress—Material

Turbine stress corrosion cracking and high- and low-cycle corrosionfatigue mechanisms are typically governed by a combination ofenvironmental effects (steam chemistry, temperature, etc.), steadyand vibratory stresses, and material strength, composition, anddefects (Figure 11). It should be noted that even pure water and wetsteam can cause cracking of turbine materials, particularly in thelow alloy rotor and disc steels, and that medium and high strengthmaterials are very susceptible to environmentally induced crackingin any environment, including pure water and steam.

Figure 11. Three Components of Turbine Stress Corrosion andCorrosion Fatigue Cracking in Turbines.

Turbine environment plays a major role in corrosion duringoperation and layup. The uniqueness of this environment is causedby the phase changes of the working fluid and the impurities carriedby the steam (steam, moisture, liquid films, and deposits). Withinthe steam flow path and on the turbine component surfaces, theparameters controlling corrosion, such as pH, concentration of saltsand hydroxides, and temperature, can change within a broad range.Even though steam impurity concentrations are controlled in thelow parts per billion (ppb) range, these impurities can concentrateby precipitation, deposition, and by evaporation of moisture topercent concentrations, becoming very corrosive (Figures 12 and13) (EPRI, 1994a, 1997b, 1997c, 1999; Jonas, et al., 1993).

Figure 12. Physical-Chemical Processes in LP Turbines.

Figure 13. Cross Section of an LPTurbine with the Locations of theProcesses Listed in Figure 15-11.

Steam Chemistry

Steam chemistry or purity, together with the thermodynamicsand flow design, determines corrosiveness of the deposits andliquid films on turbine component surfaces (Jonas, 1982, 1985a,1985d; Jonas and Dooley, 1996, 1997; EPRI, 1984b, 1994a, 1997b,1997c, 1999; Jonas, et al., 1993; Jonas and Syrett, 1987;Schleithoff, 1984). In fossil units, the LP turbine requires the lowestconcentration of impurities in the cycle, that is, low parts per billionconcentrations (1 ppb is 1 µg/liter). Steam purity is controlled bythe purity of makeup, condensate, and feedwater and in drumboilers by boiler water chemistry, boiler pressure, and carryover. Asa minimum, steam purity should be monitored by isokineticsampling and by analysis of sodium and cation conductivity (EPRI,1986, 1994c, 1998b, 1998c, 2002a; Jonas, 2000).The corrosiveness of the steam turbine environment is caused by

one or more of the following:

• Concentration of impurities from low ppb levels in steam topercent levels in steam condensates (and other deposits) resultingin the formation of concentrated aqueous solutions

• Insufficient pH control and buffering of impurities by watertreatment additives such as ammonia

• High velocity and high turbulence flow of low-pH moisturedroplets (FAC)

The situation is generalized in Figure 14, which is a Mollier diagramshowing the LP turbine steam expansion line and thermodynamicregions of impurity concentrations (NaOH, salts, etc.) andresulting corrosion. Low volatility impurities in the “salt zone” arepresent as concentrated aqueous solutions. The NaCl concentrationcan be as high as 28 percent. Note that the conditions at the hotturbine surfaces (in relation to the steam saturation temperature)


can shift from the wet steam region into the salt zone and above.This can be the reason why disc stress corrosion cracking oftenoccurs in the wet steam regions. The surfaces may be hot becauseof heat transfer through the metal or because of the stagnationtemperature effect (zero flow velocity at the surface and change ofkinetic energy of steam into heat).

Figure 14. Mollier Diagram with LP Steam Expansion Line andThermodynamic Regions for Impurity Concentration andCorrosion Mechanisms.

The impurity concentration mechanisms include:

• Precipitation from superheated steam and deposition.

• Evaporation and drying of moisture on hot surfaces.• Concentration on oxides by sorption.• Nonhomogeneous nucleation of concentrated droplets and crystalson surfaces.

Dissolved impurities deposit from superheated steam when theirconcentration exceeds their solubilities, which sharply decreases asthe steam expands. Depending on their vapor pressure, they can bepresent as a dry salt or an aqueous solution. In the wet steamregion, they are either diluted by moisture or could concentrate byevaporation on hot surfaces.The region of passivity for iron and low alloy and carbon steels

is narrow, falling within the pH range of 6 to 10. Since pH controlin a power plant is mostly for the protection of the preboiler cycleand the boiler, it often does not match the needs of the turbinesurfaces. The pH in the turbine depends on temperature, mechanicaland vaporous carryover of impurities, and water treatmentchemicals from the boiler and their volatility (distribution betweenthe vapor and the surface film).When hydrochloric acid forms in cycles with ammonia all-volatile

treatment (by decomposition of chlorinated organics, Cl� leakagefrom polishers, or seawater or other cooling water leakage in thecondenser), ammonium chloride forms in the water, and the acidmay be neutralized. However, because of its volatility, ammoniumchloride is transported with steam into the turbine where ithydrolyzes, forming NH3 gas and HCl.Deposits on turbine surfaces in units with sodium phosphate

boiler water treatment (most drum boilers) are less corrosive (Jonasand Syrett, 1987; EPRI, 1984b; Jonas, 1985d). Sodium phosphate isa better neutralizing agent through the cycle; fewer acids aretransported into the turbine and phosphate frequently codepositswith harmful impurities, providing in-situ neutralization andpassivation. This is most likely the reason for lower frequency ofturbine corrosion in systems with phosphate water treatments.Measurements of pitting potential of disc and blading alloys confirmthe beneficial effects of sodium phosphate in the presence of NaCl.Besides the corrosion during operation, turbines can corrode

during manufacture (corrosive products from machining fluids,exposure to tool tip temperatures), storage (airborne corrosiveimpurities, preservatives containing Cl and S), erection (airborne

impurities, preservatives, cleaning fluids), chemical cleaning(storage of acid in hotwells), nondestructive testing (chlorinatedcleaning and NDT fluids), and layup (deposits plus humid air).Many of these corrosive substances may contain high concentrationsof sulfur and chloride that could form acids upon decomposition.Decomposition of typical organics, such as carbon tetrachlorideoccurs at about 300�F (150�C). The composition of all of the abovecompounds should be controlled (maximum of 50 to 100 ppm Sand Cl each has been recommended), and most of them should beremoved before operation.Molybdenum disulfide, MoS2, has been implicated as a

corrodent in power system applications (Turner, 1974; Newman,1974). It can cause stress corrosion cracking of superalloys andsteels by producing an acidic environment. Its oxidation productsform low pH solutions of molybdic acid and even ammoniummolybdate, which can form during operation, causing rapid attackof turbine steels. MoS2 has been used as a thread lubricant and inthe process of disc-rotor assembling when the discs are preheatedand shrunk on the rotors. Analysis of disc bore and keywaysurfaces often reveals the presence of molybdenum and sulfur. Insteam, MoS2 reduces the notch strength of disc steel, to about 30percent of its strength in air. It has also been implicated in bolt androtor shaft failures.Layup corrosion of unprotected turbines increases rapidly when the

relative humidity of air reaches about 60 percent. When salt depositsare present, pitting during unprotected layup is rapid. Pit growth inturbine blade and rotor alloys in chloride-metal oxide mixtures in wetair is about as fast as in a boiling deaerated 28 percent NaCl solution.Turbine layup protection by clean dry air is recommended.Progress in controlling turbine corrosion through better control

of the steam chemistry includes (Jonas, 1982, 1985d, 1994; EPRI,1984b, 1986, 1994a, 1994c, 1997b, 1997c, 1998b, 1998c, 1999,2002a; Jonas, et al., 1993, 2000; Jonas and Syrett, 1987;Schleithoff, 1984, “Progress in...,” 1981):

• Decreasing concentration of corrosive impurities in makeup andfeedwater, lower air inleakage and condenser leakage, etc.

• Oxygenated water treatment for once-through fossil units forexcellent feedwater chemistry and clean boilers.

• Layup protection.• Turbine washing after chemical upsets to remove depositedimpurities.

• Reduction or elimination of copper and its oxides and theirsynergistic corrosion effects by reducing oxygen concentration,operating with a reducing (negative oxidation-reduction potential[ORP]) environment and a low ammonia concentration, or byreplacing copper alloys with steel or titanium.

PROBLEMS, THEIRROOT CAUSES, AND SOLUTIONS

Steam turbine corrosion damage, particularly of blades anddiscs, has long been recognized as a leading cause of reducedavailability (Scegljajev, 1983; McCloskey, 2002; Sanders, 2001;Cotton, 1993; Jonas, 1985a, 1987; EPRI, 1981, 1998a, 2001b; Jonasand Dooley, 1996, 1997; NERC, 2002). It has been estimated thatturbine corrosion problems cost the U.S. utility industry as much asone billion dollars per year (EPRI, 1985b, 2001a; Syrett, et al.,2002; Syrett and Gorman, 2003; Jonas, 1986) and that the cost forindustrial turbines, which suffer similar problems, is even higher.In this section, the main corrosion problems found in LP turbines

and their root causes are summarized, and solutions to reduce oreliminate each problem are discussed. The field monitoring equipmentshown in Figure 15 can be used to diagnose and prevent many commonLP turbine corrosion and deposition problems (EPRI, 1997c, 2001b;Jonas, 1994; Jonas, et al., 2007). In addition, there are also monitorsavailable to detect vibration, blade and rotor cracking, steam leaks, airinleakage, rotor position, and wear of bearings (Jonas, et al., 2007).


217

Figure 15. LP Turbine Troubleshooting Instrumentation CanIdentify Specific Corrosive Conditions. (Courtesy of EPRI, 1997c)

An overview of the low-pressure turbine corrosion problemstogether with erosion and other problems is given in Table 3. Theproblems are listed according to their priority and impact with discrim blade attachment stress corrosion cracking being the highestimpact problem today because of the long time required for weldrepair or procurement of a new disc or new rotor (up to six months).Cracking of discs, corrosion fatigue of the rotor shaft, and fatigue orcorrosion fatigue of long blades can become a safety issue becausethey can lead to perforation of the casing and other destructiveevents (EPRI, 1981, 1982a, 1998a; Jonas, 1977; Turner, 1974). It isestimated that inadequate mechanical design (high steady andvibratory stresses, stress concentration, and vibration) is responsiblefor about 50 percent of the problems, inadequate steam chemistryfor about 20 percent, and nonoptimum flow and thermodynamicdesign for about 20 percent. Poor manufacturing and maintenancepractices account for the remaining 10 percent of the problems.

Table 3. LP Turbine Corrosion, Erosion, and Deposition Problems.

When a corrosion problem is discovered during inspection or byequipment malfunction, the failure mechanism and the root causesare not always known. Even when the damage fits a description ofa well-known problem (disc or blade cracking), replacement partsmay not be readily available and the decision for what to do has tobe made quickly. The main objectives in handling identified andpotential problems are maintaining safety and avoiding forcedoutages. The questions should be asked: can we operate safely until

the next planned inspection or overhaul? If not, what is the safeinspection interval? If repeated failures are likely and repair wouldtake a long time and lead to a large loss of production, spare rotorsmay be a good economical solution.Data collected by the North American Electric Reliability

Council (NERC) for 1476 fossil units between 1996 and 2000shows that LP turbines were responsible for 818 forced andscheduled outages and deratings, causing the utilities a 39,574GWh production loss. The outages are often characterized as“low frequency high impact events.” (NERC, 2002) shows thecomponents that were responsible for the failures as well as theMWh losses associated with the failures. Many of these outagesare caused by corrosion (except for bearings).

Table 4. Forced and Schedule Outages and Deratings Caused byLP Turbine Components for the Years 1996 to 2000 (1476 Units,168 Utilities).

Life Prediction and Inspection Interval

Experience shows that pits and ground-out stress corrosioncracks can remain in-service for several years, depending on stressand environment. However, components containing high-cyclecorrosion fatigue cracks should not be left in-service. Proceduresfor prediction of residual life and determination of a safe inspectioninterval have been developed for all major failure mechanismsincluding SCC, CF, fatigue, FAC, and creep. The procedures forSCC of turbine discs (Clark, et al., 1981; EPRI, 1989; Rosario, etal., 2002), low cycle corrosion fatigue, and FAC (EPRI, 1996) havebeen successfully applied because all variables influencing thesemechanisms can be reasonably predicted or measured. However,life prediction for high cycle corrosion fatigue and fatigue hasnot been so successful because the vibratory stresses and thecorrosiveness of the environment are usually not accurately known.Life prediction is based on results of inspection, fracture

mechanics analysis of components with defects, and application ofSCC and CF crack growth data. Time or number of load cycles toreach ductile or brittle fracture is predicted and a safety factor isapplied to determine the time for the next inspection. In theprocedure used by OEMs and Nuclear Regulatory Commission(NRC) for nuclear turbines for determining the inspection intervalfor turbine discs under SCC conditions, the safety factor of twowas applied to the predicted time-to-failure.

Stress Corrosion Cracking of Discs

Stress corrosion cracking of LP turbine disc keyways and bladeattachments have been the two most expensive generic problems inlarge steam turbines (Cheruvu and Seth, 1993; EPRI, 1982a, 1982b,1984a, 1984c, 1985a, 1985d, 1987b, 1989, 1991a, 1997a, 1998d;Jonas, 1978; Speidel and Bertilsson, 1984; Clark, et al., 1981;Rosario, et al., 2002; Nowak, 1997; Kilroy, et al., 1997; Amos, et al.,1997; Turner, 1974; Newman, 1974; Parkins, 1972; Holdsworth,2002). The keyway cracking problem has been resolved byredesigns of the shrunk-on or bolted-on discs, material replacementwith lower strength material, and by elimination of the corrosive discbore lubricants, based on molybdenum disulfide (MoS2), used inassembling the rotor. SCC of blade attachments of various designs isstill a problem (EPRI, 1997d, 1998a, 2002c; Nowak, 1997).


There are several corrosion damage mechanisms and manyfactors affecting discs (Figure 16). Typical locations and orientationsof SCC cracks in LP discs are shown in Figures 17 and 18. Therehas also been SCC in pressure balance holes. Most incidents ofdisc rim attachment cracking have been found in nuclear units,however, there have also been problems in fossil units. In anindependent survey, 13 of 38 (35 percent) boiling water reactors(BWRs) and 28 of 72 (39 percent) pressurized water reactors(PWRs) reported disc rim attachment cracking while 29 of 110 (26percent) supercritical fossil units and only 20 of 647 (3 percent)subcritical units reported cracking (EPRI, 1997d). Disc rim bladeattachment cracking occurred in multiple-hook (steeple), fir treeattachment designs (typically occurs in the corners of the hooks),in straddle mount dovetail and pinned-finger attachments, and inT-roots. It is always associated with stress concentrations.

Figure 16. Crack Locations in Turbine Discs with Probable ImpurityConcentration and Corrosion Mechanisms and Corrodents. (Courtesyof Jonas, 1985a)

Figure 17. Typical Locations and Orientations of SCC Found in LPTurbine Discs. (Courtesy of EPRI, 1982a)

Figure 18. Typical Locations of Disc Rim Cracking. (Courtesy ofEPRI, 1982a)

Materials—All low alloy steels used for LP turbine rotorsand discs are susceptible to SCC and CF in numerous turbineenvironments including pure water and wet steam. The strongestmaterial factor influencing SCC is yield strength. At higher yieldstrength, SCC crack growth rate can be several orders of magnitudehigher than for lower yield strength materials. The purity of thematerial and the steel melting practices mostly influence thefracture toughness, which determines the maximum tolerable cracksize before a disc brittle fracture and burst.Fractography—SCC cracks of low alloy steels often initiate

from pits and propagate intergranularly with branching. The initialpart of the crack can be corroded and filled with magnetite. Therecould be beach marks or stretch marks, caused by overloading thecrack during overspeed testing. Another type of beach mark can becaused by changes of the environment or by fatigue. Depending onthe ratio of the steady and cyclic stresses, the disc cracks can be amixture of intergranular and transgranular cracking. Figure 19shows an SCC crack initiating at the radius of the upper serrationof L-1 blade steeple. The intergranular crack initiated from a pit.

Figure 19. SCC Crack in the Upper Steeple Serration-L-1 BladeAttachment.

The factors that determine the SCC crack initiation time andpropagation rate include material yield strength, surface stress,temperature, and the local chemical environment. Some of theserelationships are shown in Figures 6 and 7. At yield strengthsabove ~135 ksi (930 MPa), these low-alloy steels show high SCCgrowth rates.Pitting often initiates SCC.When corrosive deposits are present,

pitting during unprotected layup can be faster than pitting duringoperation. This is because during the layup, there can be 100percent relative humidity and there is oxygen present. At highstresses, above the elasticity limit of the material, pitting isenhanced through the mechanism identified as stress inducedpitting (Parkins, 1972). In some cases, blade attachment crackingis a combination of stress corrosion cracking and corrosion fatiguebecause of the effects of blade vibration.

Root Causes

SCC of discs (at keyways, bores, and blade attachments) iscaused by a combination of high surface stresses, a susceptiblematerial, and operational and shutdown environments. Design-relatedroot causes are the most important and prevalent. They includehigh surface tensile stresses and stress concentrations, and use ofhigh strength materials.Sources of stresses that contribute to SCC of discs include:

• Basic centrifugal load caused by rotor rotation. Locally highconcentration of centrifugal loads caused by variation in the gaps(gauging) between blade and disc rim attachment.

• Residual machining stresses.• Vibratory stresses—interaction of SCC and corrosion fatigue. Also,vibratory stresses reduce the life of the cracked disc when the flawsreach a sufficient size that fatigue becomes a dominant mechanism.

Steam chemistry root causes of SCC and CF cracking include:

• Operating outside of recommended steam purity limits forlong periods of time; sometimes caused by organic acids fromdecomposition of organic water treatment chemicals.


219

Condenser leaks—minor but occurring over a long period of time.

• Condenser leaks—major ingress, generally one serious event,and the system and turbine not subsequently cleaned.

• Water treatment plant or condensate polisher regenerationchemicals (NaOH or H2SO4) leak downstream.

• Improperly operated condensate polisher (operating beyondammonia breakthrough, poor rinse, etc.).

• Shutdown environment: poor layup practices plus corrosive deposits.

Sodium hydroxide is the most severe SCC environmentencountered in steam turbines. The sources of NaOH includemalfunctioning condensate polishers and makeup systems andimproper control of phosphate boiler water chemistry combinedwith high carryover. Many other chemicals can also cause SCC oflow alloy steels. The chemicals used in turbine assembly andtesting, such as molybdenum disulfide (lubricant) and Loctite™(sealant containing high sulfur), can accelerate SCC initiation(Turner, 1974; Newman, 1974).

Solutions

In most cases where material yield strength is <130 ksi (895MPa), the solution to disc SCC is a design change to reducestresses at critical locations. This has been achieved by eliminatingkeyways or even disc bores (welded rotors) and by larger radii inthe blade attachments. Higher yield strength (>130 ksi, 895 MPa)low alloy steel discs should be replaced with lower strengthmaterials. The goal is to keep the ratio of the local operating stressto yield stress as low as possible, ideally aiming for the ratios to beless than 0.6. Minimizing applied stresses in this manner is mostbeneficial in preventing initiation of stress corrosion cracks. Oncecracks begin to propagate, a reduction in stress may be onlymarginally effective unless the stress intensity can be kept below~10 to 20 ksi-in1/2 (11 to 22 MPa-m1/2). This is because of therelative independence of the crack growth rate over a broad rangeof stress intensities. For many rim attachment designs, such levelsof applied stress intensity are impossible to achieve once an initialpit or stress concentration has formed. An emerging solution todisc rim stress corrosion cracking is a weld repair with 12%Crstainless steel. Another solution has been to shot peen the bladeattachments to place the hook fit region into compression.Good control of the steam purity of the environment can help to

prevent or delay the SCC. Maintaining the recommended levels ofimpurities during operation and providing adequate protectionduring shutdown can help minimize the formation of deposits andcorrosive liquid films, and lengthen the period before stresscorrosion cracks initiate. The operating period(s), events, ortransients that are causing excursions in water and steamchemistry should be identified using the monitoring locations andinstrumentation recommended in the independent water chemistryguidelines (EPRI, 1986, 1994c, 1998b, 1998c, 2002a; Jonas, et al.,2000) and special monitoring as shown in Figure 15 and elsewhere(EPRI, 1997c, 2001b; Jonas, et al., 2007).

Corrosion Fatigue andStress Corrosion Cracking of Blades

LP turbine blades are subject to CF, SCC, and pitting of theairfoils, roots, tenons and shrouds, and tie wires (Holdsworth, 2002;EPRI, 1984c, 1984d, 1985c, 1985d, 1987c, 1991b, 1993b, 1994b,1998d; Jaffe, 1983; Evans, 1993; Singh, et al.; BLADE-ST™,2000). Figure 20 depicts the typical locations on an LP turbinerotating blade that are affected by localized corrosion and cracking.In addition, the blade surfaces are also subject to fatigue, deposition,water droplet erosion, and foreign object damage. CF is the leadingmechanism of damage. It is a result of the combination of cyclicstresses and environmental effects. There are always environmentaleffects in fatigue cracking in LP turbines (EPRI, 1984d, 1984e) and

all fatigue cracking should be considered corrosion fatigue. Thefatigue limit of all turbine materials in turbine environmentsincluding pure steam is lower than the air fatigue limit. Corrosionfatigue cracks often originate from pits.

Figure 20. Typical Locations of Cracking and Localized Corrosionon LP Turbine Rotating Blades. There Has Also Been SCC and CFCracking in the Tiewire Holes. (Courtesy of EPRI, 1998a)

Figures 21 and 22 show corrosion fatigue cracks of L-1 bladeroot and airfoil, respectively. Figure 23 illustrates pitting in theblade tenon-shroud area, which sometimes initiates corrosionfatigue cracking. Fractography shows that CF blade cracking ofteninitiates from a pit, continuing for 50 to 150 mils (1.3 to 3.8 mm)by intergranular cracking and then proceeding as a flat fatiguefracture with beach marks and striations.

Figure 21. Corrosion Fatigue of L-1 Blade Attachment. (Courtesyof EPRI, 1998a)

Figure 22. Corrosion Fatigue of L-1 Blade Airfoil. (Courtesy ofEPRI, 1981)


•

Figure 23. Pitting in the L-1 Blade Tenon-Shroud Area. (Courtesyof EPRI, 1981)

Damage by corrosion fatigue occurs in the last few rows of LPturbines, mostly in the phase transition zone (PTZ) (salt zoneshown in Figure 14). The PTZ moves according to load changes,but is typically near the L-1 row in most LP fossil turbines (Figure24). Units that increase cycling duty may be subject to worsenedcorrosion fatigue problems. As the unit is ramped up and shutdown, the blades pass through resonance more frequently and thephase transition zone shifts, potentially affecting more stages duringthe transients. In addition, the steam purity can be significantlyworse during transients than during steady-state operation, and ifthe unit is shut down as part of the cycling operation, significantdegradation of the local environment can occur (deposits andhumid air).

Figure 24. Distribution of Blade Failures in U.S. Fossil Turbines byRow. (Courtesy of Power, 1981)

Causes of blade and blade attachment failures are listed in Table 5(Jonas, 1985a). To find the true causes of corrosion, it is essential toanalyze the local temperature, pressure, chemistry, moisture dropletflow, and stress conditions. These analyses are often neglected.

Table 5. Causes of Blade Failures in LP, IP, and HP SteamTurbines.

Root Causes

High stresses and marginal steam chemistry acting together are themost frequent root causes. Corrosion fatigue cracks are driven bycyclic stresses with high mean stress playing a large role. Figure 10shows how fatigue strength is reduced at high mean stresses. It hasbeen estimated that the corrosion fatigue limit for long LP turbineblades in the area of high mean stress is as low as 1 ksi (7 MPa). Roughsurface finish and pitting often shorten the time to crack initiation.Cyclic stresses are caused by turbine startups and shutdowns

(low number of cycles, high cyclic stresses), by the turbine andblades ramping through critical speeds at which some componentsare in resonance (high amplitude, high frequency), and by over 15causes of flow induced blade excitation during normal operationthat include:

• Synchronous resonance of the blades at a harmonic of the unitrunning speed.

• Nonuniform flows.

• Blade vibration induced from a vibrating rotor or disc.

• Self-excitation such as flutter.• Random excitation-resonance with adjacent blades.

• Shock waves in the transonic flow region and shock wave-condensation interaction.

• Bad blade design with wrong incidence flow angle and flowseparation.

Elevated concentrations of steam impurities (particularly ofchloride, sodium, and sulfate) and the resulting deposition andconcentration by evaporation of moisture are underlying causes ofcorrosion fatigue. When feedwater, boiler water, and steamimpurity levels exceed recommended limits, cycle chemistry canbe a contributor or even the root cause. Poor shutdown and layupprocedures are primary contributors to aggressive environmentsthat can lead to pitting and corrosion fatigue. High steam sodiumor cation conductivity may indicate conditions that can lead torapid accumulation of deposits and concentrated liquid films onblade surfaces.

Solutions

The solutions to blade corrosion fatigue problems include:

• Design of blades that are not in resonance with running speedand its harmonic frequencies or with any of the excitation sourceslisted above.

• Design with friction damping.• Elimination of the sources of excitation.• Reduction of mean and alternating stresses by design (lower stressconcentrations, etc.).

• Better materials, such as by avoiding high strength alloys, usingmaterials with high material damping, or using titanium alloys.

• Improvement of steam chemistry.

Long-term actions for dealing with corrosion fatigue begin witheconomic and remaining life assessments. Depending upon theseverity of the problem and the costs to eliminate it, some solutionsmay not be practical for all circumstances. The available preventionstrategies fall into four main categories: redesigning the blade toreduce resonance, redesigning the blade or attachment to reducestress levels, improving steam purity, and/or changing the materialor surface (better surface finish, shot peening) of the blade.Stress reduction options include:

• Changing the vibration resonance response of the blade bydesign modification (adding or reducing weight of the blade,


221

changing to free standing, grouped or fully connected shroudedblades, moving tiewires or tenons, shroud segment integrallymachined with the blade—no tenons, etc.).

• Changing the response of the blades by mixed tuning.• Increasing the damping of the blades.• Changing operating procedures; for example, avoiding off-designoperation such as very low load, high backpressure operation(which can cause stall flutter), and changing rotational speedduring startups.

Options to improve the turbine environment include:

• Controlling impurity ingress (reduce air inleakage, plug leakingcondenser tubes, etc.).

• Changing unit operating procedures, particularly for shutdownsand startups.

• Control of boiler carryover by drum design and water chemistry.

• Optimizing or changing feedwater and boiler water treatment toreduce concentration of impurities in steam.

Another option to improve the environment at surfaces is toimprove the surface finish of blading. Deposition and subsequentconcentration of impurities are a function of blade surface finish(EPRI, 2001b), and this improvement may help slow the accumulationof impurities. Improving the continuous monitoring of steamchemistry (sodium, cation conductivity) will help to verifyimprovements in the environment.If such changes are not sufficient, then changing to a more

corrosion resistant material, such as a material with a higherchromium content or a titanium alloy, is generally recommended.It should be noted that higher strength materials are often more

susceptible to stress corrosion cracking. It has been shown that403SS, with yield strength above ~90 ksi (620 MPa), becomessusceptible to SCC.

Pitting

Pitting can be a precursor to more extensive damage from CF andSCC, although extensive pitting of blades can also cause significantloss of stage efficiency by deteriorating the surface finish (EPRI,2001b). Pitting is found in a wide range of components. It occursmost prevalently during shutdown when moisture condenses onequipment surfaces and as a result, it can be found in stages of theturbine that are dry during operation. However, it can also occurduring operation, particularly in crevices (crevice corrosion).In LP turbines, pitting is primarily found on the turbine blades

and blade-disc attachments, particularly in the “salt zone” (Figure14). There is little pitting on wet stages because corrosion impuritiesare washed away by the steam moisture. Pitting is frequently foundin the blade tenon-shroud crevices because, once corrosive impuritiesenter, they cannot be removed. Titanium alloys are the mostresistant to pitting corrosion, followed by duplex ferritic-austeniticstainless steels (Fe-26Cr-2Mo), precipitation hardened stainlesssteels (Fe-14Cr-1.6Mo) and 12%Cr steels.

Root Causes

Steam and deposit chemistries are the main causes of pitting,with chlorides and sulfates being the main corrodents. Copper andiron oxides accelerate pitting by providing the matrix that retainssalts. Copper oxides also transport oxygen to the corrosion sites.Dissolved oxygen does not concentrate in the liquid films formingon blade surfaces during operation but does, however, accumulatein the liquid films and wet deposits that can form during unitshutdown if proper layup practices are not used. There have beencases of severe blade pitting requiring blade replacement on brandnew rotors from which preservatives were stripped leaving themexposed to sea salt during prolonged erection periods.

Solutions

The first line of defense against pitting is controlling steampurity. This will improve the local environment produced duringoperation and decrease the amount of deposition. The chemistryguidelines established by an independent research and developmentfirm (EPRI, 1986, 1994c, 1998b, 1998c, 2002a; Jonas, et al., 2000)particularly for cation conductivity, chloride, and sulfate, should befollowed. There should be a layup protection of the LP turbines bydehumidified air, vapor phase inhibitors, or nitrogen.After a steam chemistry upset, such as a large condenser leak,

the turbine should be washed online or after a disassembly. Doingnothing may result in multimillion dollar corrosion damagerequiring rotor replacement.

Flow-Accelerated Corrosion

Flow-accelerated corrosion of carbon and low-alloy steels in thesteam path two-phase flow has been less widespread in fossilplants than in nuclear plants; however, it has occurred at somelocations (EPRI, 1996; 1998a; Kleitz, 1994; Jonas, 1985b;Svoboda and Faber, 1984) such as:

• Wet steam extraction pipes and extraction slots.

• Exhaust hood and condenser neck structure.• Casings.• Rotor gland and other seal areas.• Disc pressure balance holes.• Rim and steeples of last row disc.

• Rotor shaft—last disc transition.

• Leaking horizontal joint.• Transition between the stationary blades and the blade ring.

While most cases of flow-accelerated corrosion damage are slowto develop and are found during scheduled inspections, FAC ofpiping and turbine casing horizontal joints can lead to leaks andFAC of rotors and discs can initiate cracking.

Root Causes

Root causes of FAC in the turbine (EPRI, 1996, 1998a; Jonas,1985b) include:

• Susceptible material: carbon steel or low-chromium steel.

• Locally high flow velocities and turbulence.

• High moisture content of the steam.• Low levels of dissolved oxygen, excess oxygen scavenger.

• Low pH of moisture droplets.

• Water/steam impurities.

Solutions

It is recommended that a comprehensive FAC control programbe implemented, including an evaluation of the most susceptiblepiping and other components (EPRI, 1996). Areas of local thinningneed to be periodically inspected and repaired. Approaches topiping repair include replacement with low alloy steels, weldoverlay, and plasma arc and flame spraying to protect susceptiblesurfaces. The material applied should have high chromium content.If a component is replaced, the material of the new componentshould contain some chromium. For example, carbon steel pipeshould be replaced with 1.25 percent or higher chromium steel.Little can be done about changing the moisture concentration infossil turbines. Steam chemistry improvements through bettercontrol of feedwater and boiler water chemistry, such as reductionof organic acids, could result in increase of pH of the earlycondensate and less FAC (EPRI, 1997b, 1997c, 1999).


Treatment of feedwater, such as maintaining high pH levels(above 9.6) and elevated oxygen concentrations, can also reduceFAC in the turbine.

Other Phenomena (Noncorrosion)

Although not specifically corrosion related, there are otherproblems that occur in steam turbines including: deposition onblade surfaces; water droplet erosion of wet stage blades; low cyclethermal fatigue of heavy high temperature sections of rotor, casing,and pipes; solid particle erosion of turbine inlets and valves; andwater induction and water hammer.

Deposition on Blade Surfaces

Deposits are the result of impurities in the feedwater, boilerwater, and attemperating water being carried over into the turbine(Jonas and Dooley, 1997; EPRI, 1997b, 2001b; Jonas, et al., 1993;Jonas, 1985d). All impurities are soluble in superheated and wetsteam and their solubility depends on pressure and temperature.The steam leaving the steam generator is at the highest steampressure and temperature in the cycle. As it passes through theturbine, the pressure and temperature decrease, the steam losesits ability to hold the impurities in solution, and the impuritiesprecipitate and deposit on the turbine blades and elsewhere.The main impurities found in turbine deposits are magnetite,

sodium chloride, and silica. It takes only a few hours of a chemicalupset, such as a major condenser leak or a boiler carryover event,to build up deposits, but it takes thousands of hours of operationwith pure steam to remove them.The impact of deposits on turbine performance is the most

pronounced in the HP section. Performance loss depends ondeposit thickness, their location (steam pressure), and the resultingsurface roughness (EPRI 2001b). Deposits will change the basicprofile of the nozzle partitions resulting in losses caused bychanges in flow, energy distribution, and aerodynamic profiles, aswell as by surface roughness effects. These changes can result inlarge megawatt and efficiency losses. With replacement powertypically over $100/MWh and costing as much as $7000 per MWhin the summer of 1998, the savings from reducing this depositioncan be very high.In the LP turbine, deposits are often corrosive, they can change

the resonant frequency of blades, increase the centrifugal load onblade shrouds and tenons and, in the transonic stages, they caninfluence the generation of shock waves.Solutions—Optimization of cycle chemistry (Jonas, 1982,

“Progress in...,” 1981; EPRI, 1986, 1994c, 1998b, 1998c; 1985d,2002a; Jonas, et al., 2000, 2007;ASME, 2002) is the easiest methodfor reducing impurity transport and deposition. The optimal cyclechemistry will result in reduced corrosion and minimized impuritytransport. This is especially important if copper alloys are present inthe system because the optimal feedwater pH for copper alloys andferrous materials are not the same and the incorrect pH can result inhigh levels of iron or copper transport. Other methods for managingdeposition on blade surfaces include:

• Specify good surface finishes (polished blades) on all new andreplacement blades.

• Determine the effect of erosion and deposition on maximumMW and efficiency with a valves wide open (VWO) test. The datafrom several VWO tests can be compared to determine the rate ofMW loss and MW versus chemistry and operation. This can beused to optimize the system.

• Turbine washing to reduce deposition and MW losses andimprove efficiency. Both, a turbine wash of an assembled turbineand a wash of a disassembled rotor can be used to remove solubledeposits. To remove corrosive salt deposits, several days ofwashing may be needed. A wash is usually completed when theconcentration of corrosive impurities, such as sodium and chloride,in the wash water is less than 50 ppb.

Water Droplet Erosion

In the last stages of the LP turbine, the steam expands to wellbelow saturation conditions and a portion of the vapor condensesinto liquid (EPRI, 2001b; Ryzenkov, 2000; Oryakhin, et al., 1984;Rezinskikh, et al., 1993; Sakamoto, et al., 1992; Povarov, et al.,1985; Heyman, 1970, 1979, 1992). Although the condenseddroplets are very small (0.05 to 1 µm, 2 to 40 µin diameter), someof them are deposited onto surfaces of the stationary blades wherethey coalesce into films and migrate to the trailing edge. Here theyare torn off by the steam flow in the form of large droplets (5 to 20microns, 0.2 to 0.8 mils). These droplets accelerate under theforces of the steam acting on them and, when they are carried intothe plane of rotation of the rotating blades, they have reached onlya fraction of the steam velocity. As a result, the blades hit them witha velocity that is almost equal to the circumferential velocity of theblades, which can be as high as 640 m/s (2100 ft/s) in a fossil LPturbine. Water droplet erosion typically occurs in the last two tothree rows of the LP turbines in fossil fired units. The damage ismost common on the leading edge and tip of the blades and alongthe shroud.In turbines operating at low load for long periods, such as

cycling and peaking units, reversed flow of steam caused bywindage and activation of hood spray can erode the trailing edgesof blades. Thin trailing edges with erosion grooves can becomefatigue or corrosion fatigue crack initiation sites. Less frequenterosion damage locations include LP turbine glands and seals,stationary blades, blade attachment sections of discs, disc flowholes (impulse design), and the LP rotor at the gland.The effects of steam and early condensate chemistry on water

droplet erosion are not known, but studies have found that NaCl inthe droplets significantly reduces the incubation period for erosion.In addition, pH was found to have a strong impact on both theincubation period and the erosion rate. At higher pH, both themaximum and steady state erosion rates decrease while the incubationperiod increases (Ryzenkov, 2000; Povarov, et al., 1985).Solutions—There are several options available for reducing the

amount of damage from liquid droplet impact. There are two principaloptions: protection of the leading edge by a hard material andcollection and drainage of moisture. Table 6 outlines these options.

Table 6. Long-Term Actions for Reducing Moisture Erosion on LPTurbine Blades.

Turbine Inspection and Monitoring

Adequate turbine inspection methods are available to detectcorrosion and deposition. These NDT methods include visual,magnetic particle, ultrasonic, dye penetrant, eddy current, andradiographic techniques. Modern turbine designs consider theaccessibility of individual components by inspection probes.Monitoring techniques include stress and vibration monitoring

(i.e., vibration signature), temperature and flow measurement, andwater and steam chemistry sampling and analysis. (Jonas, 1982,1985d, 1986, 1994; EPRI, 1984b, 1994a, 1994c, 1997b, 1997c,1998c, 1999, 2002a; Jonas, et al., 1993, 2000; Jonas and Syrett,1987; Schleithoff, 1984; “Progress in...,” 1981). An advancedexpert system has been developed (EPRI, 1994c) for the use bystation operators and chemists, which automatically determines the


223

problems and recommends corrective actions. There are alsomonitoring methods for online diagnosis of the environments onturbine surfaces and corrosion (Jonas, 1994; EPRI, 1994a, 1997b,1997c, 1999; Jonas, et al., 1993).

MISSING KNOWLEDGE

It is estimated by the author that 70 percent of knowledge tosolve and prevent corrosion problems in steam turbines isavailable. The percentage of available knowledge for understandingthe effects of stress and environment is much lower than that forsolving the problems, about 40 percent. The knowledge that ismissing or needs improvement includes:

• Threshold stress required to initiate SCC in blade attachments.

• Effects of steeple geometry (stress concentrations and size) onSCC and CF.

• Effects of overloads during heater box and overspeed testson stress redistribution and SCC in steeples, blade roots, and disckeyways.

• Effectiveness of grinding out SCC and CF cracks as a correctivemeasure.

• Effects of organic water treatment chemicals, and organic impuritieson SCC, CF, and pitting and composition of water droplets.

• Effects of electrical charges carried by water droplets on corrosion.

• Effects of galvanic coupling of dissimilar materials, such as theblade-steeple, on corrosion.

• Effects of residues of preservatives and Loctite™ on SCC, CF,and pitting of blade attachments.

• Effects of blade trailing edge erosion on cracking.• Accelerated stress corrosion testing.• Effects of variable amplitude loading on CF crack initiationand propagation.

• Effects of water droplet pH and composition on erosion and howto predict erosion.

• Effects of shot peening to reduce stresses and SCC of blade rootsand disc steeples.

• Understanding of the basic mechanisms of stress corrosion,corrosion fatigue, fatigue, and stress induced pitting.

CASE HISTORIES

Stress Corrosion Cracking in Finger-Style Dovetails

Unit:The station consists of three 805 MW, once-through boiler,supercritical, coal-fired units that went into operation between1974 and 1976. Each of the three units has two LP turbines (LPAand LPB).Problem: In 1995, in Unit 1, a blade failed in the L-1 row at a

tiewire hole of the leading blade of a four-blade group (Nowak,1997; Kilroy, et al., 1997). Three damaged groups of blades wereremoved for replacement. A wet fluorescent magnetic particleexamination of the finger-style disc attachments found hundreds ofcrack-like indications, which were later identified as SCC. Cracksran both axially and radially, and were deep (Figure 25).

Figure 25. Locations in Finger and Pin Attachments Where SCCHas Been Found. (Courtesy of EPRI, 1997)

Inspection: Inspection procedures and the acceptance criteriathat would be applied were developed before the outage.Eventually NDE inspection found some damage in each of the 12ends of all six rotors. Severe cracking was found in all four-rotorends in Unit 2 and in Unit 3 on both ends of LPB. Next in severitywere both ends of Unit 1 LPB and Unit 3 LPA; with a few indicationsin Unit 1 LPA. In the heavily cracked rotors, damage was mostsevere in Fingers 3, 4, and 5 with little or no cracking in Fingers 1or 6. Cracking was evenly distributed between admission anddischarge sides in Fingers 3 and 4 with somewhat more crackingon the admission side in Finger 5.Results of metallurgical analysis: A metallurgical evaluation

confirmed the presence of extensive pitting. Cracks were found to beintergranular, highly branched, and oxide-filled. The metallurgicalexamination was unable to detect the presence of specific contaminantson the fracture surface. Sampling of deposits, which had occurredelsewhere in the cycle (crossover piping, bucket pins, and bucketfingers) during prior outages, had found indications of sodium, sulfur,and chloride and sodium hydroxide was found by x-ray diffraction incrossover piping.Analysis of samples from two rotors showed that the chemical

composition was within the specification for ASTM A470 Class 7material. Tensile strength averaged 126.5 ksi (870 MPa) for the twospecimens; yield averaged 112.5 ksi (775 MPa).Review of cycle chemistry: Several cycle chemistry changes

had been made over time in response to events such as improvedtechnology and information, and upsets caused by condenser tubeleaks, demineralizer breaks, and variations in boiler water makeup.The main water chemistry problem was operation with morpholine,which resulted in poor condensate polisher performance and highconcentrations of sodium hydroxide in feedwater and steam. Aheavy deposit of sodium hydroxide was found in the crossoverpiping. The unit had been changed to oxygenated treatment aboutthe time of the discovery of the stress corrosion cracks, which hassince resulted in better water and steam purity.Results of stress analysis: A finite element stress analysis showed

that Fingers 3, 4, and 5 of the group were the most highly stressed.The maximum equivalent elastic stresses around the pin holes were~222 ksi (1530 MPa), ~114 ksi (786 MPa) at the inner land, and ~89ksi (614 MPa) at the outer land (closer to disc outer diameter [OD]).The differences between inner and outer land stresses agreed with theobservation that field cracking was more severe at the inner land.However, no SCC was discovered in the flat portion of the discfingers around the holes, where the stress was nearly twice as high.Closer examination found that the flat surface near the pinholes

did not show cracks nucleating from the bottom of the pits,whereas intergranular cracks appeared in the pits along the ledgewhere the pits were linked to form continual flaws.Root causes: There were two root causes of this massive

problem: high design stresses and improper feedwater chemistryusing morpholine, which resulted in the presence of NaOH andother impurities in steam. One can also speculate about thecontributions of the local stress concentration, poor surface finish,and residual machining stresses.Economic analysis: An economic analysis was performed and

the costs considered included: replacement of rotors in-kind,replacement with an improved rotor steam path that would improveunit heat rate by 1.2 percent, rotor weld repair, outage durationcosts, performance changes, reduced generation capacity frompressure plates, fuel pricing, and replacement energy costs.Actions: As a temporary fix, the affected blade rows were

removed and nine pressure plates were installed out of the 12possible locations. A pressure plate is a temporary device thatprovides a pressure drop when installed as a replacement for aremoved blade row. A decision was made to purchase two new fullybladed rotors and to refurbish the existing rotors. For Unit 1, newrotors were purchased from the original equipment manufacturer(OEM). The two LP rotors removed from Unit 1 were weld repaired


using 12 percent chromium material applied with a submerged arcweld process, and installed in Unit 3. Material testing and analysiswere used to determine the expected lifetime of the refurbishedparts against damage by SCC, and both high- and low-cycle fatigue.Turbine and crossover pipes were cleaned to remove deposits.

Massive SCC of Disc Rim—SupercriticalOnce-Through Unit after Only Five Years of Service

Problem: Stress corrosion cracking of the L-1 stage disc wasdiscovered during routine inspection (Figure 26).

Figure 26. Massive SCC of L-1 Disc Caused by HighConcentration of NaOH in Steam.

Root cause: The SCC was caused by poor performance of thecondensate polishers that operated in H-OH form throughammonia breakthrough when they released Na+. The sodiumreacted with water forming NaOH and deposited in the turbine.Actions: Rotor replaced and operation of condensate polishers

and monitoring was improved.

Stress Corrosion Cracking ofBolted-on Discs in One Type of LP Turbine

Problem: In the effort to accommodate longer L-0 blades, onetype of LP turbine was originally designed with the bolted-on lastdisc made of NiCrMoV low-alloy steel and heat treated to a yieldstrength of up to 175 ksi (1200 MPa). This steel was found to besusceptible to SCC.Root cause: The root cause of this problem was design with a

high strength material that is very susceptible to SCC.Solution: All of these discs in many power plants had to be

replaced with a lower strength material because of the danger ofSCC failures in all types of steam environments.

Corrosion Fatigue of a Blade Airfoil

Unit: A 400 MW reheat unit with a once-through boiler,seawater cooling, and mixed bed condensate polishers.Problem: During a three-month period, condenser cooling

water leakage occurred periodically. Cation conductivity in thecondensate and feedwater increased up to 2 µS/cm for about 30 to60 minutes per day. At the end of the three-month period, vibrationwas detected in the LP turbine (EPRI, 1998a)Damage: The turbine was opened and five broken freestanding

L-2 rotating blades were found. According to the turbine designdata, the broken blades were in the phase transition zone. Theblades were broken at the transition between reddish deposits at theblade foot and clean metal in the upper part of the blades (phasetransition on the blade). A laboratory investigation found chlorideand sodium at the crack site and confirmed corrosion fatigue as theunderlying mechanism. A calculation of vibration frequencies didnot show abnormal conditions.Root cause: Improper water chemistry conditions caused by the

condenser tube leak.Actions: The broken blades were replaced, the condenser leak

was repaired, and a dampening (lacing) wire was introduced intothe blade design. By doing so, both the environmental and stresscontributors were reduced.

Corrosion Fatigue of Numerous Modifications of L-1 Blades

Problem: One type of LP turbine experienced corrosion fatiguefailures of the L-1 blade airfoil, which was modified andredesigned over 10 times. Fatigue failures occurred in periodsranging from six weeks to over 10 years of operation. The presenceof chloride in the blade deposits at concentrations above 0.25percent caused pitting, which accelerated crack initiation. In onecase, new 16 inch (40 cm) blades were forged in two dies and mostof the blades forged in one of the dies (but none from the other die)failed within about six weeks.Root cause:The root cause of this fast CF failure was an off-design

blade geometry caused by wrong die dimensions, which broughtthese blades into resonance. The failure acceleration was caused bycorrosive impurities in steam, mainly chloride.Actions: Redesigned, better tuned blades were installed and

improved control of water and steam chemistry was initiated.

Massive Pitting of a Turbine (HP, IP,and LP) after Brackish Water Ingress

Problem: A separation of the welded end of the condensatesparger caused breakage of condenser tubes and ingress ofbrackish water into a once-through boiler cycle. Because ofthe poor reliability of the water chemistry instrumentation, theinstrument readings were ignored and the trouble was noticed afterthere was almost no flow through the superheater because of heavydeposits. The turbine with sea salt deposits was left assembled inthe high humidity environment and was only opened 11 days afterthe condenser tube failure. It was found that the whole turbine wasseverely rusted and pitted and it was eventually replaced.Mixed bed condensate polishers were not able to protect the

cycle against the massive ingress of brackish water. They wereexhausted within a few minutes.Root cause: The root cause of the impurity ingress was a failure of

the condensate sparger. Poor water chemistry monitoring and controland the long delay in beginning turbine damage assessment andcleaning significantly contributed to the amount of damage caused.Actions: The rotors were replaced and new chemistry monitoring

instrumentation was installed.

L-0 Blade Corrosion FatigueCracking Caused by Trailing Edge Erosion

Problem: After 14 to 18 years of service of one type of 400 MWturbine, there were five cases of L-0 33.5 inch (85 cm) longshrouded blade failures about 5 inches (12.7 cm) below the tip. TheCF cracks originated at the eroded and thinned trailing edge. Whenthe blade tips separated, the blade fragments had such kinetic energythat they penetrated over 2 inches (5 cm) of carbon steel condenserstruts. The blade material was martensitic 12%Cr stainless steel witha Brinell hardness of ~345. All affected units were similar drumboiler units, some on all volatile treatment (AVT) and some onphosphate treatment. Steam chemistry in the affected units was goodand did not play a role in the rate of erosion or cracking.Root cause: Erosion caused by frequent operation at low load

with the hood sprays on and reversed steam flow, lack of properearly inspection, and blade design with thin trailing edge.Actions: Heavily eroded and cracked blades were replaced,

shallow erosion damage was polished, and similar turbines wereinspected for damage.

Stress Corrosion Cracking of Dovetail Pins

Problem: A dovetail pin, which penetrates both the rotatingblade and the wheel dovetails, holds a bucket in place on the wheelof a rotor. In many plants during the 1970s and earlier, dovetail pinshad suffered stress corrosion cracking, although no lost blades hadresulted. The material originally used for the dovetail pins wassimilar to ASTM A681 Grade H-11 tool steel, with a chemistry ofFe-5.0Cr-1.Mo-0.5V-0.4C at a strength level of 250 to 280 ksi


225

(1715 to 1920 MPa). This material is still used in rare cases wherethe highest strength is required.Root cause: Use of high strength material, which is susceptible

to SCC, combined with high bending stresses.Actions: The approach to solving the cracking problem was

twofold: changing the material chemistry and strength and usingsteel ball shot peening to impart a compressive layer to the surfaceof the finished pins. The new high strength material is 5CrMoVlow alloy steel at a strength level of 240 to 270 ksi (1645 to 1850MPa). For lower stress applications, a new 1CrMoV low alloy steelwith a strength of (170 to 200 ksi) is used.Results: No cracking has been observed in dovetail pins since the

change in materials and the introduction of the shot peening practice.

Turbine Destruction—Sticking Valves

Problem:After only 16 hours of operation of a new 6 MW steamturbine installed in a fertilizer plant, an accidental disconnect of theelectrical load on the generator led to a destructive overspeed. Theoverspeed occurred because high boiler carryover of boiler watertreatment chemicals, including polymeric dispersant, introducedthese chemicals into the bushings of all turbine control valves,gluing the valves stuck in the open position.Root cause: Poor control of boiler operation (drum level)

together with the use of the polymeric dispersant that, afterevaporation of water, becomes a strong adhesive. Controls of theelectric generator allowed accidental disconnect.Actions: New turbine generator installed, boiler and generator

controls fixed, turbine valves reused after dissolution of thebushing deposits in hot water.

CONCLUSIONS

• Steam turbines can be a very reliable equipment with life over30 years and overhaul approximately every 10 years. However,about 5 percent of the industrial and utility turbines experiencecorrosion and deposition problems. Mostly due to LP blade and bladeattachment (disc rim) corrosion fatigue or stress corrosion failures.

• The root causes of the blade and disc failures include design withhigh stresses, bad steam chemistry, and use of high strength materials.

• Other steam turbine problems include: low cycle thermalfatigue, pitting during unprotected layup and operation, loss ofMW/HP and efficiency due to deposits, water droplet erosion, flowaccelerated corrosion, solid particle erosion by magnetite particlesexfoliated from superheater, turbine destructive over speed causedby the control valves stuck open because of deposits in thebushings, and water induction-water hammer.

• All the problems are well understood, detectable, and preventable.Monitoring, inspection, and defects evaluation methods are available.These methods include design reviews and audits of operation andmaintenance, NDT, life prediction, vibration monitoring, vibrationsignature analysis, water, steam, and deposit chemistry monitoringand analysis, valve exercise, and control of superheater temperatures.

• Steam cycle design and operation influences turbine problemsby causing high steady and vibratory stresses, by thermal stressesrelated to load and temperature control, and by water and steampurity and boiler carryover.

REFERENCES

Amos, D., Lay, E., and Bachman, S., 1997, “Qualification ofWelding Rotors with 12Cr Stainless Steel to Improve SCCResistance,” Proceedings: Steam Turbine Stress CorrosionWorkshop, EPRI, Palo Alto, California, TR-108982.

ASM International, 1989, ASM Metals Handbook, 9th Edition, 17,Nondestructive Evaluation and Quality Control.

ASME, 1982, ASME Committee on Water in Thermal PowerSystems, Industrial Subcommittee, Steam Purity Task Group,

Results of 1982 Survey, American Society of MechanicalEngineers, NewYork, NewYork.

ASME, 1989, The ASME Handbook on Water Technology forThermal Power Systems, American Society of MechanicalEngineers, NewYork, NewYork.

ASTM 2001, 2001 Annual Book of ASTM Standards, Section 3:Metals Test Methods and Analytical Procedures, 03,Nondestructive Testing,West Conshohocken, Pennsylvania.

ASME, 2002, Consensus for the Lay-up of Boilers, Turbines,Turbine Condensers, and Auxiliary Equipment, ASMEResearch Report, CRTD-Vol. 66, 2002. American Society ofMechanical Engineers, NewYork, NewYork.

Atrens, A., et al., 1984, “Steam Turbine Blades,” Corrosion inPower Generating Equipment, Plenum Press.

Baboian, R., Editor, 1995, “Corrosion Tests and Standards—Application and Interpretation,” ASTM, West Conshohocken,Pennsylvania.

BLADE-ST™ Version 3, 2000, STI Technologies, Rochester,NewYork.

Cheruvu, N. and Seth, B., 1993, “Key Variables Affecting theSusceptibility of Shrunk-On Discs to Stress CorrosionCracking,” The Steam Turbine Generator Today: Materials,Flow Path Design, Repair, and Refurbishment, ASME, NewYork, NewYork.

Clark, W. G., et al., 1981, “Procedures for Estimating the Probabilityof Steam Turbine Disc Rupture from Stress CorrosionCracking,” ASME/IEEE Power Generation Conference, PaperNo, 81-JPGC-PWR-31, NewYork, NewYork.

Cotton, K., 1993, Evaluating and Improving Steam TurbinePerformance, NewYork, NewYork: Cotton Fact, Inc.

EPRI, 1981, “Steam Turbine Blades: Considerations in Design anda Survey of Blade Failures,” Palo Alto, California, CS-1967.

EPRI, 1982a, “Steam Turbine Disc Cracking Experience: Volume1—Literature and Field Survey, Volume 2—Data Summariesand Discussion, Volume 3—Stress Corrosion Cracking ofLow-Alloy Steels, Volume 4—Factors Determining ChemicalComposition of Low Pressure Turbine Environments, Volume5—Characteristics and Operating Histories of U.S. PowerPlants, Volume 6—Description of Turbine Rotor Models,Volume 7—Metallurgical Analysis of Cracked Discs from 10U.S. Power Plants,” Palo Alto, California, NP-2429-LD.

EPRI, 1982b, “Metallurgical Evaluation of a Failed LP TurbineDisc,” Palo Alto, California, NP-2738.

EPRI, 1983, “EPRI Research Related to the Steam TurbineGenerator,” Palo Alto, California, NP-3288-SR.

EPRI, 1984a, “Properties of Turbine Disc Materials,” Palo Alto,California, NP-3634.

EPRI, 1984b, “The Effects of Phosphate Environments on TurbineMaterials,” Palo Alto, California, CS-3541.

EPRI, 1984c, “Stress Corrosion Cracking in Steam Turbine Discs:Survey of Data Collection, Reduction, and ModelingActivities,” Palo Alto, California, NP-3691.

EPRI, 1984d, “Corrosion Fatigue of Steam Turbine-Blading Alloysin Operational Environments,” Palo Alto, California, CS-2932.

EPRI, 1984e, “The Effect of Variables on the Fatigue Behavior ofTi-6Al-4V,” Palo Alto, California, CS-2934.

EPRI, 1985a, “Stress Corrosion Cracking in Steam Turbine Discs:Analysis of Field and Laboratory Data,” Palo Alto, California,NP-4056.

EPRI, 1985b, “Survey of SteamTurbine Blade Failures,” PaloAlto,California, CS-3891.


EPRI, 1985c, “Proceedings: Steam Turbine Blade ReliabilitySeminar and Workshop,” Palo Alto, California, CS-4001.

EPRI, 1985d, “Stress Corrosion Cracking and Corrosion Fatigue ofSteam Turbine Materials,” Palo Alto, California, NP-4074M.

EPRI, 1986, “Interim Consensus Guidelines on Fossil PlantChemistry,” Palo Alto, California, CS-4629.

EPRI, 1987a, “Guide for Use of Corrosion-Resistant Coatings onSteam Turbine Blades,” Palo Alto, California, CS-5481.

EPRI, 1987b, “Stress Corrosion Cracking of A471 Turbine DiscSteels,” Palo Alto, California, NP-5182.

EPRI, 1987c, “Steam Turbine Blade Reliability Seminar andWorkshop,” Palo Alto, California, CS-5085.

EPRI, 1989, “Guidelines for Predicting the Life of Steam TurbineDiscs Exhibiting Stress Corrosion Cracking,” Palo Alto,California, NP-6444.

EPRI, 1991a, “Proceedings: Fossil Steam Turbine Disc CrackingWorkshop,” Palo Alto, California, GS-7250.

EPRI, 1991b, “Florida Power and Light Uses BLADE Program toAnalyze Low Pressure Turbine Blade Failures,” Palo Alto,California, 100229.

EPRI, 1993a, “Solid Particle Erosion Technology Assessment,”Final Report, Palo Alto, California, TR-103552.

EPRI, 1993b, “Proceedings of the Steam and Combustion TurbineBlading Conference and Workshop—1992,” Palo Alto,California, TR-102061.

EPRI, 1994a, “Steam Chemistry and Corrosion,” Palo Alto,California, TR-103738.

EPRI, 1994b, “BLADE Code Helps TVA Diagnose Blade FailureQuickly and Avoid Blade Replacements,” Palo Alto,California, IN-10420.

EPRI, 1994c, “Cycle Chemistry Guidelines for Fossil Plants:Oxygenated Treatment,” Palo Alto, California, TR-102285.

EPRI, 1995, “Low-Pressure Turbine Blade Design Evaluation,”Palo Alto, California, TR-105839.

EPRI, 1996, “Flow-Accelerated Corrosion in Power Plants,” PaloAlto, California, TR-106611.

EPRI, 1997a, “Proceedings: Steam Turbine Stress CorrosionWorkshop,” Palo Alto, California, TR-108982.

EPRI, 1997b, “Turbine Steam, Chemistry and Corrosion:Experimental Turbine Tests,” Palo Alto, California, TR-108185.

EPRI, 1997c, “Steam, Chemistry, and Corrosion in the PhaseTransition Zone of Steam Turbines: Volume 1: Key Results,Summary, and Interpretation; Volume 2: Individual Contributionsof Participants,” Palo Alto, California, AP-108184.

EPRI, 1997d, “Low-Pressure Rotor Rim Attachment CrackingSurvey of Utility Experience,” PaloAlto, California, TR-107088.

EPRI, 1997e, “Moisture Nucleation in Steam Turbines,” Palo Alto,California, TR-108942.

EPRI, 1998a, “Turbine Steam Path Damage: Theory and Practice,”Palo Alto, California, AP-108943.

EPRI, 1998b, “Cycling, Startup, Shutdown, and Layup—FossilPlant Cycle Chemistry Guidelines for Operators andChemists,” Palo Alto, California, TR-107754.

EPRI, 1998c, “Interim Cycle Chemistry Guidelines for CombinedCycle Heat Recovery Steam Generators (HRSGs),” Palo Alto,California, TR-110051.

EPRI, 1998d, “Proceedings: Workshop on Corrosion of SteamTurbine Blading and Discs in the Phase Transition Zone,” PaloAlto, California, TR-111340.

EPRI, 1999, “Turbine Steam, Chemistry and Corrosion-Generationof Early Liquid Films in Turbines,” Palo Alto, California,TR-113090.

EPRI, 2000a, “Productivity Improvement Handbook for Fossil SteamPower Plants: Second Edition,” Palo Alto, California, TR-114910.

EPRI, 2000b, “Corrosion of Low Pressure Steam TurbineComponents,” Palo Alto, California, 1000557.

EPRI, 2001a, “Cost of Corrosion in the Electric Power Industry,”Palo Alto, California, 1004662.

EPRI, 2001b, “Steam Turbine Efficiency and Corrosion: Effectsof Surface Finish, Deposits, and Moisture,” Palo Alto,California, 1003997.

EPRI, 2002a, “Cycle Chemistry Guidelines for Fossil Plants:All-Volatile Treatment,” Palo Alto, California, 1004187.

EPRI, 2002b, “Advances in Life Assessment and Optimization ofFossil Power Plants,” Proceedings, PaloAlto, California, 1006965.

EPRI, 2002c, “Priorities for Research and Development for theElectric Power Industry,” Palo Alto, California, 1007274.

Evans, D., 1993, “Comparison of Unstalled Flutter Predictionsand Field Measurements for Steam Turbine Blades,”Proceedings of the Steam and Combustion Turbine-BladingConference and Workshop—1992, EPRI, Palo Alto,California, TR-102061.

Haas, H., June 1977, “Major Damage Caused by Turbine orGenerator Rotor Failures in the Range of the Tripping Speed,”Der Maschinenschaden.

Heymann, F., 1970, “Toward Quantitative Prediction of LiquidImpact Erosion,” ASTM STP 474, American Society forTesting and Materials.

Heymann, F., 1979, “Conclusions from the ASTM InterlaboratoryTest Program with Liquid Impact Erosion Facilities,”Proceedings of the Fifth International Conference on Erosionby Liquid and Solid Impact, Cavendish Laboratory, Universityof Cambridge, United Kingdom.

Heymann. F., 1992, “Liquid Impingement Erosion,” ASMHandbook, Volume 18: Friction, Lubrication, and WearTechnology, ASM International.

Holdsworth, S., 2002, “Prediction and Prevention of StressCorrosion and Corrosion Fatigue Cracking LP SteamTurbines,” Advances in Life Assessment and Optimization ofFossil Power Plants—Proceedings, EPRI, Palo Alto,California, 1006965.

Jaffee, R., Editor, 1983, Corrosion Fatigue of Steam Turbine BladeMaterials, NewYork, NewYork: Pergamon Press.

Jonas, O., 1977, “Major Damage to Steam-Turbosets,” PanelDiscussion at the 9th Allianz Forum, “Technology andInsurance,” 1976, and Der Maschinenschaden, 1977.

Jonas, O., 1978, “Tapered Tensile Specimen for Stress CorrosionThreshold Stress Testing,” ASTM Journal of Testing andEvaluation, 6, (1), pp. 40-47.

Jonas, O., September 1982, “Beware of Organic Impurities inSteam Power Systems,” Power, 126, (9).

Jonas, O., February 1985a, “Steam Turbine Corrosion,” MaterialsPerformance.

Jonas, O., March 1985b, “Control Erosion/Corrosion of Steels inWet Steam,” Power.

Jonas, O., 1985c, “Design Against Localized Corrosion,” SecondInternational Symposium on Environmental Degradation ofMaterials in Nuclear Power Systems—Water Reactors,Monterey, California.


227

Jonas, O., 1985d, “Transport of Chemicals in Steam Cycles,” PaperNo.245, Corrosion/85, NACE.

Jonas, O., 1986, “Cost of Corrosion and Scale in Electric Utilities,”1985 Fossil Plant Water Chemistry Symposium: Proceedings,EPRI, Palo Alto, California, CS-4950.

Jonas, O., 1987, “Corrosion of SteamTurbines,”Metals Handbook, 9thEdition, 13, Corrosion,ASM, International, Metals Park, Ohio.

Jonas, O., May 1989, “Developing Steam-Purity Limits forIndustrial Turbines,” Power.

Jonas, O., 1994, “On-Line Diagnosis of Turbine Deposits and FirstCondensate,” 55th Annual International Water Conference,Pittsburgh, Pennsylvania.

Jonas, O., 1995, “Steam Generation,” Corrosion Tests andStandards—Application and Interpretation, ASTM, WestConshohocken, Pennsylvania.

Jonas, O., 1997, “Effects of Steam Chemistry on MoistureNucleation,” Moisture Nucleation in Steam Turbines, EPRI,Palo Alto, California, TR-108942.

Jonas, O. and Dooley, B., 1996, “Steam Chemistry and its Effectson Turbine Deposits and Corrosion,” 57th International WaterConference, Pittsburgh, Pennsylvania.

Jonas, O. and Dooley, B., 1997, “Major Turbine Problems Related toSteam Chemistry: R&D, Root Causes, and Solutions,”Proceedings: Fifth International Conference on Cycle Chemistryin Fossil Plants, EPRI, Palo Alto, California, TR-108459.

Jonas, O. and Syrett, B. C., 1987, “Chemical Transport and TurbineCorrosion in Phosphate Treated Drum Boiler Units,”International Water Conference, Pittsburgh, Pennsylvania.

Jonas, O., Dooley, B., and Rieger, N., 1993, “Steam Chemistry andTurbine Corrosion—State-of-Knowledge,” 54th InternationalWater Conference, Pittsburgh, Pennsylvania.

Jonas, O., et al., 2000, “EPRI ChemExpert: Cycle ChemistryAdvisorfor Fossil Power Plants,” EPRI 6th International Conference onCycle Chemistry in Fossil Plants, Columbus, Ohio.

Jonas, O., et al., March 2007, “Water Hammer and other HydraulicPhenomena,” Power.

Kilroy, R., et al., 1997, “A 12% Chrome Weld Repair IncreasesStress Corrosion Cracking Resistance of LP Finger Type RotorDovetails,” Proceedings of the International Joint PowerGeneration Conference, 2, Power.

Kleitz, A., 1994, “Erosion and Corrosion Phenomena inWet SteamTurbines,” ESKOM Power Plant Chemistry Symposium.

Kramer L. and Michael, S., September 1975, “Service Experienceand Stress Corrosion of Inconel 600 Bellows Expansion Joinsin Turbine Steam Environments,” Materials Performance.

McCloskey, T., 2002, “TroubleshootingTurbine Steam Path Damage,”Advances in Life Assessment and Optimization of Fossil PowerPlants—Proceedings, EPRI, Palo Alto, California, 1006965.

McIntyre, P., 1979, “U.K. Experimental Work on Turbine Disc andRotor Cracking: An Overview,” Working on SCC in TurbineRotors and Discs, EPRI-CEGB, Leatherhead, England.

NERC, 2002, pc-GAR (Generating Availability Report, StatisticalData 1982-2000), North American Electric Reliability Council.

Newman, J., 1974, “The Stress Corrosion of Turbine Disc Steels inDilute Molybdate Solutions and Stagnant Water,” CERL LabNote No. RD/L/N 215/74.

Nowak, 1997, “Low Pressure Turbine Stress Corrosion CrackingInvestigation at the Navajo Generating Station,” Proceedings:Steam Turbine Stress Corrosion Workshop, EPRI, Palo Alto,California, TR-108982.

Parkins, R., January 1972, “Stress Corrosion Spectrum,” BritishCorrosion Journal, 7.

Petr, V., Kolovratnik, M., Jiricek, I., and Jonas, O., 1997,“Experimental Investigation of the Effects of Steam Chemistryon Droplet Nucleation,” Moisture Nucleation in SteamTurbines, EPRI, Palo Alto, California, TR-108942.

Povarov, O., et al., June 1985, “Basic Regularities of Erosion ofSteam-Turbine Blade Materials under Impingement of WaterDroplets,” Strojarstvo.

“Progress in Understanding Improves Outlook for Control ofTurbine-Steam Chemistry,” March 1981, Power.

Pryakhin, V. A., et al., October 1984, “Problems of Erosion of theRotating Blades of Steam Turbines,” Teploenergetika.

Rezinskikh, V., Vaiman, A., and Melamed, M., November 1993,“The Mechanism of Damaging the Metal of Blades of SteamTurbines Working in the Zone of the Phase Transition,”Teploenergetika.

Rosario, et al., 2002, “Evaluation of LP Rotor Rim-AttachmentCracking Using LPRimLife,” Advances in Life Assessmentand Optimization of Fossil Power Plants—Proceedings, EPRI,Palo Alto, California, 1006965.

Ryzenkov, V., 2000, “The State of the Problem and Ways forIncreasing the Wear Resistance of the Power Equipment atThermal Power Stations,” Thermal Engineering, 47, (6).

Sakamoto, T., Nagao, S., and Tanuma, T., 1992, “Investigation ofWet Steam Flow for Steam Turbine Repowering,” SteamTurbine-Generator Developments for the Power GenerationIndustry, ASME Power, 18.

Sanders, W., 2001, Steam Turbine Path Damage and MaintenanceVolumes 1 (February 2001) and 2 (July 2002), Pennwell Press.

Scegljajev, V., 1983, Parni Turbiny (Steam Turbines), Prague,Czechoslovakia: SNTL.

Schleithoff, K., 1984, “Influence of Water Steam Chemistry onStress Corrosion Cracking of Steam Turbine LP Discs,”Corrosion in Power Generating Equipment, Plenum Press.

Singh, M., Matthews, T., and Ramsey, C., “Fatigue Damage ofSteam Turbine Blade Caused by Frequency Shift Due toSolid Buildup—A Case Study,” http://www.dresser-rand.com/e-tech/tp110/tp110prt.htm

Speidel, M. and Atrens, A., Editors, 1984, Corrosion in PowerGenerating Equipment, Plenum Press.

Speidel, M. and Bertilsson, J., 1984, “Stress Corrosion Cracking ofSteam Turbine Rotors,” Corrosion in Power GeneratingEquipment, Plenum Press.

Stastny, M., Jonas, O., and Sejna, M., 1997, “Numerical Analysisof the Flow with Condensation in a Turbine Cascade,”MoistureNucleation in Steam Turbines, EPRI, Palo Alto, California,TR-108942.

Svoboda, R. and Faber, G., 1984, “Erosion-Corrosion of SteamTurbine Components,” Corrosion in Power GeneratingEquipment, Plenum Publishing Corporation.

Syrett, C. and Gorman, J. A., February 2003, “Cost of Corrosionin the Electric Power Industry—An Update,” MaterialsPerformance.

Syrett, C., Gorman, J. A., Arey, M. L., Koch, G. H., and Jacobson,G. A., March 2002, “Cost of Corrosion in the Electric PowerIndustry,” Materials Performance.

Turner, D., 1974, “SCC of LP Turbines: The Generation ofPotentially Hazardous Environments from MolybdenumCompounds,” CERL Lab Note No. RD/L/N 204/74.


turbo machinery presentation collection

Documents