American Institute of Aeronautics and Astronautics1

AIAA-99-2825

EXPERIMENTAL AND NUMERICAL RESULTS OF THE COUPLED SEAL CAVITY AND MAINFLOW FOR A LIQUID HYDROGEN ROCKET TURBOPUMP

K. N. OliphantMember

Project EngineerConcepts ETI, Inc.

White River Junction, Vermont

Dr. D. JapikseSenior Member

PresidentConcepts ETI, Inc.

White River Junction, Vermont

Abstract

Effective sealing in secondary flow paths is critical tothe performance of rocket turbopumps. It is importantto be able to predict seal performance and theinteraction of seal flows on the main flow in order forthe designer to properly assess the impact of aparticular seal design on the overall systemperformance. This paper presents the results of acomputational fluid dynamics (CFD) calculation of thefirst stage space shuttle main engine (SSME) LH2rocket turbopump (an alternate design) with sealcavities and compares the CFD results to experimentaldata.

The CFD prediction showed a good match to test datafor the pressure rise through the impeller and thepressure distribution through the front seal cavity.However, the axial thrust prediction was off by asignificant amount, which was probably due toincorrectly calculated pressure distributions in thebackface seal cavity. Also, the predicted impeller inputpower was too low. The disk friction on the backfaceseal is probably not being accurately predicted whichwould account, at least in part, for the discrepancy inthe power and axial thrust.

Copyright © 1999 by Concepts ETI, Inc. Published bythe American Institute of Aeronautics and Astronautics,Inc. with permission.

In addition, the CFD calculations indicated that thelocation of the seal cavity flow injection into theimpeller eye region could be an important design issuefor both the seal cavity and the impeller. The pressuredisturbance from the presence of the impeller leadingedge caused the flow to reverse direction locally fromthe expected direction and to go back into the sealcavity. More study is required to investigate thisphenomenon and determine its effect on the systemperformance.

Introduction

An important design and evaluation activity has beenconducted around the SSME liquid hydrogen highpressure turbopump with emphasis on the first stage.This study was based on a Phase 1 and Phase 2 SmallBusiness Innovative Research (SBIR) grant fromMarshall Space Flight Center, Huntsville, Alabama.The project focused on using advanced pump andcompressor design technology to create an alternativedesign with good throttleability, high efficiency, andreduced part count. Figure 1 displays a photograph of aconventional LH2 first-stage SSME rocket turbopumpimpeller compared with the advanced technologyimpeller. The new impeller has an increased inlet bladecount from 6 to 8 blades, but reduced splitter countfrom 2 rows to a single row giving an exit blade countof 16 blades rather than the 24 blades of the currentturbopump. Hence, the impeller design was simplified.

American Institute of Aeronautics and Astronautics2

Downstream of the impeller a continuous crossoverdiffuser system is employed and the modified designconfiguration is shown in Figure 2. This systemrepresents a substantial reduction in overall diameterand simplicity in component design. Prior rocketturbopump crossover elements were designed as aconveying passage with very thick (vane) regionsbetween each conveying passage. The new crossoverdesign is configured based on a diffusing vane concept.Good performance was established, consequentlyconfirming a reduction in diameter which can betranslated to reduced size and weight of futuregeneration rocket turbopumps.

The project had numerous technology objectives one ofwhich was to understand the front and rear leakagecavities, the coupled cavity-impeller flow character, andthe resulting effects on thrust balance. This paperreports a comparison between basic measurements inthe front cavity and critical observations obtained by adetailed coupled CFD calculation between the cavityflows and the main flow passage.

Fundamental Flow Modeling via CFD

Modern computational fluid dynamic methods providesignificant opportunity for basic or fundamental studiesof fluid dynamic phenomena. When coupled withmeasured results from the laboratory, it is believedthat important insights, sufficient to guide futuregenerations of pump design, will typically result. Toobtain quality CFD calculations, four important issuesmust be dealt with. 1 These include 1) high quality grid,2) realistic handling of numerical viscosity ordiscretization error, 3) turbulence model, and 4) carefulrepresentation of inlet and discharge boundaryconditions.

a) Typical SSME first stage LH2turbopump impeller

b) Photo of redesigned impeller showingbowed blading at inlet.

Figure 1. Comparison of a conventional LH2 first-stage SSME rocket turbopump impeller (a) withthe advanced technology impeller (b).

Figure 2. Modified design configuration showing acontinuous crossover diffuser system employeddownstream of the impeller.

American Institute of Aeronautics and Astronautics3

For the present investigation, a time-marching multi-grid code called FINE/Turbo* was used for thecalculations.2 It employs a second order accuratecentral difference discretization scheme with precondi-tioning to allow for computation of incompressibleflows.3 A typical grid display is shown in Figure 3.The mesh was prepared using GridPro† whichprovides excellent multiblock definition of very highquality grids.

A Baldwin-Lomax turbulence model was employed inthis computation. It provides adequate performance inboundary layer dominated flows, which is prevalent inthe main flow path of the impeller, provided theboundary layer is resolved well enough. For this casethe mesh was clustered towards the walls for a firstnode y+ value of approximately 2. To be sure, this * FINE/Turbo is a trademark of NUMECAInternational† GridPro® is registered by Program DevelopmentCorporation

turbulence model will break down in the more complexstress-strain fields of the seal cavities. However, thestandard one or two equation models that are commonlyavailable to the industrial user are all based on the sameeddy viscosity assumption, which is not valid incomplex stress-strain fields. The use of more advancedturbulence models was outside of the scope of thiswork.

The impeller displayed in Figures 1 and 3 was, in fact,tested behind the bill-of-materials SSME LH2 rocketturbopump inlet element as shown in Figures 4 and 5.This element is a side inlet housing with an inlet guidevane system that imparts approximately 45° of swirl tothe impeller flow path. Fortunately, some flowmeasurements have been conducted on the flow leavingthis device and are available to workers concerned withthe SSME rocket turbopump investigation. Hencemeasured inlet velocity and flow angle distributions justupstream of the eye of the impeller were used for theboundary condition at inlet. The static pressure wasimposed as the boundary condition at the discharge.

Figure 3. Meridional view of mesh with front and backface seal cavities.

American Institute of Aeronautics and Astronautics4

As Figure 3 displays, the flow field includes not onlythe main passage of the impeller, but also the frontcavity flow field and the rear cavity flow field. Eachhas its own block description and detailed flowcharacteristics are calculated in each component and allelements (the main flow path, the front cavity block,and the rear cavity block) are solved concurrently to asingle converged final solution. Because the returnchannel was not modeled in this computation, the massflow through the rear seal cavity was imposed. Themass flow was adjusted until the pressure drop throughthe rear seal cavity matched the data.

Figure 4. Impeller tested behind the bill-of-materials SSME LH2 rocket turbopump inlet element.

Figure 5. Turbopump test stand showing magneticbearings, pump inlet housing, and inlet duct plusinstrumentation. Magnetic bearing is enclosed in thecasing shown on the left side, most stage hardware is inthe insulated section just under the white inlet duct, andthe crossover is located in the aft square section withdischarge into an axisymmetric collector with four exitports.

American Institute of Aeronautics and Astronautics5

Computed and Measured Results

Table 1 displays a few of the basicparameters involved in describing thisstage and the resulting calculations. Theevaluations were carried out at a flow ratecorresponding to a Q/N which typifiesnormal operation which, for water-rig testpurposes, yields a rate of 54.3 lbm/s. Testspeed for this water-rig evaluation was 889rpm. It will be observed that the exit staticpressure from the impeller was closelyreplicated in the CFD calculations (33.3psi measured versus 32.9 psi computed)and that the exit total pressure (42.3 psicalculated) is in close proximity to a singletotal pressure probe located downstream atthe diffuser throat (42.0 psia) with somepressure drop anticipated between thecomputed and the downstream measuredlocations. The actual efficiencies for theimpeller are slightly incorrect with a fivepoint discrepancy in the impeller total-to-total efficiency but only a two-pointdiscrepancy in the impeller total-to-staticefficiency. The total-to-static may be amore realistic comparison as the estimateof measured total-to-total efficiency maybe considered less precise. This followssince the exit static pressure could bereadily measured, but the exit total pressureis not measured in this test rig with detailedprecision. Further understanding of thecomparison between measured data andcomputed results is shown in Figure 6. Inthis figure, the static pressure rise throughthe front cavity is displayed. Measuredpressure data, at various radial locations,are shown as the large solid square symbol.It may be observed that the agreement isgenerally excellent.

Figure 7 shows the calculated staticpressure distribution through the rear sealcavity compared to measured test data.Recall that the mass flow through therear seal cavity was adjusted until thepressure drop through the cavity wasapproximately equal to the test data.Hence, the good agreement between thetest data and the CFD is meaningless. Thecomputed static pressure distributionthrough the rear seal cavity is, however, ofinterest. It shows that the rear damper seal

Static Pressure Along Front Seal Cavity Wall

16.00

18.00

20.00

22.00

24.00

26.00

28.00

30.00

32.00

34.00

3.40 3.60 3.80 4.00 4.20 4.40 4.60 4.80 5.00 5.20 5.40 5.60 5.80 6.00 6.20

Radius

Sta

tic

Pre

ssu

re (

psi

)

CFD Results

Test Data

Figure 6. Comparison of CFD predictions to test data of thestatic pressure along the front seal cavity outer wall.

Static Pressure Along Rear Seal Cavity Wall

30.00

31.00

32.00

33.00

34.00

35.00

36.00

37.00

38.00

39.00

40.00

2.00 2.40 2.80 3.20 3.60 4.00 4.40 4.80 5.20 5.60 6.00 6.40

Radius (in.)

Sta

tic

Pre

ssu

re (

psi

)

CFD Results

Test Data

Figure 7. Comparison of CFD predictions to test data of thestatic pressure along the rear seal cavity wall.

Table 1. Test data parameters compared to CFD results.

Parameter Units Data CFDMass Flow lbm/s 54.3 54.4Wheel Speed RPM 889 889Inlet Total Pressure (P00) psia 22.42 22.43Impeller Exit Static Pressure (P2) psia 33.28 32.89Impeller Exit Total Pressure (P02) psia - 42.25Diffuser Throat Total Pressure (P04) psia 42.04 -Shaft Power hp 5.22 4.82Total-to-Total Impeller Efficiency (ηtt) 0.89 0.94Total-to-Static Impeller Efficiency (ηts) 0.47 0.49Axial Thrust lbf 768 840Front Seal Cavity Flow lbm/s - 0.50Rear Seal Cavity Flow lbm/s - 0.33

American Institute of Aeronautics and Astronautics6

at a radius of 2.1 inches (see Figure 4) is the singlebiggest factor in determining the pressure distributionon the impeller backface. Calculating the right pressuredrop and mass flow through that seal is of primaryimportance for an accurate axial thrust prediction.

The computation included the front and rear sealcavities; therefore, it was possible to calculate a CFDpredicted axial thrust to compare to the measured thrustfrom the magnetic bearing rig. The thrust wascalculated by summing the pressure and momentumflux forces around the computational domain. Inaddition, the forces on the impeller shaft, which werenot part of the CFD model, were taken from the testdata and added to the CFD predictions in order toprovide a proper comparison to the test data.4 Table 1shows the predicted axial thrust and the measuredresults. The prediction is 9.4% higher than the testedresults. Part of this difference comes from the smalldiscrepancy between the test data and the CFDpredicted pressure in the front cavity near the impellerexit (see Figure 6). Integrating the difference in thepressures over the area between the two data pointsclosest to the impeller exit results in an axial thrustdifference of about 19 pounds force. This accounts forabout 2.5% of the higher thrust value. The other 6.9%is most likely due to an incorrectly predicted pressuredistribution over the impeller backface.

The flow in the seal cavities is highly complex whichincludes laminar to turbulent transition, non-equilibrium turbulent flow, jets and wakes, and largeseparation regions. The turbulence model which hasbeen tuned for boundary layer dominated flowsis not able to handle these complex flowphenomena. In addition, there were somegeometric features (i.e., bolt heads) that werenot modeled in the CFD which could have aneffect on the results. This is especially true forthe bolt pattern that is close to the impeller tip(see Figure 4) on the backface since the diskfriction is proportional to the diameter to thefifth power and the pressure distribution will beaffected by the disk friction.

The comparison just presented between some ofthe measured results and computed results isquite interesting, but the greater value probablylies in the insights obtained into the character ofthe basic flow field which is made possible bycareful study of the CFD results. This study is now shown. Attention is first focused onFigure 8. The contour plot is actually the gapcavity shown in Figure 3 along the shroud line

just at impeller inlet. In other words, there is a smallgap where flow can move radially up or down from theseal cavity area into the main flow area. The littlecircumferential strip which forms the interface betweenthe main flow and the seal cavity flow is the strip dis-played in Figure 8. These are radial velocities.Normally, one thinks of the flow moving radiallyinward from the seal leakage regime down into themain flow path (negative Cr). However, a careful ex-amination of these results, shows that this is not alwaystrue. The expected flow direction, using negative valueof Cr corresponds to all contours less than the numbernine. Immediately in front of the main blade, however,there are a variety of iso-contours running from 9 up to14. In this region, the flow is actually reversed and theflow is going radially from the main flow up into thecavity region. This is brought about by a stagnationeffect of the flow as it approaches the main blade. Theincrease in local static pressure as the flow is tendingtowards stagnation, or in the vicinity of stagnation, issufficient to cause the flow direction to reverse and togive a radial outflow. Of course this flow must bebalanced and eventually is swept, in an adjacentcircumferential position, back into the eye of theimpeller. A similar presentation is shown in Figure 9concerning the flow near the impeller exit. In this case,instead of having the radial velocity component as usedin Figure 8, Figure 9 shows the axial velocitycomponent which can be visualized again by referenceto Figure 3. The gap for the front cavity in the vicinityof the impeller exit has an axial orientation and it isthrough this strip that the flow is now being studied.Consequently, the velocities are axial. Negative axial

14 12 10

98

768

76 5 4

13

Level cr14 0.5013 0.4012 0.3011 0.2010 0.109 0.008 -0.107 -0.206 -0.305 -0.404 -0.503 -0.602 -0.701 -0.80

Main Blade

U

Figure 8. Radial component of velocity contours at locationof seal cavity injection into impeller eye.

American Institute of Aeronautics and Astronautics7

velocity is the direction in which a flow would beexpected to leak from the impeller tip into thefront leakage cavity. However, some of thesevelocity values are, in fact, positive indicatingflow is going back out into the main stream. Thepositive values are found for contours above 12.Indeed, if one carefully examines these contourdesignations, there is a light gray region of flowwhich falls above 12 in this diagram. Hence asmall pocket of flow is moving from the sealcavity back into the main flow (although most ofthe flow is in the expected direction).

Important streamline information follows next.Figure 10 displays the streamlines in the mainflow path and some aspects of the streamwiseflow in the adjacent cavities. In general, the basicflow character is much as expected. The onlyanomaly is a small separation bubble up in thevaneless diffuser; frequently, these can occureither in the physical laboratory or in the

computational laboratory and can be avoided by usingpinch in the vaneless diffuser (a vaned diffuser wasused for testing – the vaneless is for computationalpurposes only), but this was not done for the presentinvestigation. It will imply a small distortion in theaxial components at the impeller discharge whencompared with the actual configuration. Figure 11shows pertinent details at the impeller inlet especially inthe seal cavity regime. The streamlines reflect the basicpattern which might be anticipated for a flow of thisnature. The strong jet-vortex nature of flow within eachcavity is, of course, the intended objective: a seal mustcreate maximum losses in order to minimize leakageflow rate. Further details can be observed in Figure 12where a classic front cavity recirculation, roughlylooking like a Couette type of flow field, has beenestablished in the front passage. The boundary layeraround the rotating wall is moving radially outward dueto centrifugal effects on the shear layer; the effect of thedifferential pressure across the cavity develops theradial inflow along the outer wall. Incidentally, thevery thick shroud surface may appear to the reader assomething of an anomaly. There is a good reason forthis. The external contour of the impeller shroud isidentical between the redesign, utilized in this paper,and the original LH2 impeller configurations. For theredesign, a narrower passage width was employed inorder to obtain desired flow control. Consequently, athicker shroud or cover was employed for theexperimental investigations. In an actual applicationthis would be much thinner and the entire front cavityflow problem would be translated closer to the mainflow path. Flow phenomena in the rear cavity, see

119

1011

12

12

12

12

11

10

9

Level wz13 0.1012 0.0011 -0.1010 -0.209 -0.308 -0.407 -0.506 -0.605 -0.704 -0.803 -0.902 -1.001 -1.10

Main Blade Splitter Blade

U

Figure 9. Axial component of velocity contours at locationof seal cavity injection from impeller exit.

Figure 10. Streamlines in main flow path and thefront and rear seal cavities.

Figure 11. Velocity vectors and streamlines showthe details of the flow in the front cavity labyrinthseals.

American Institute of Aeronautics and Astronautics8

Figure 12, may also be observed to be quite similar innature to that which is displayed in the front cavityexcept for the interesting radial jet-like flowphenomena which are observed in the labyrinth sealareas. The flow character in these cavities is rationalbut a bit different than one might anticipate from thefront cavity examination. The flow is being pumpedinto a radial outward direction. When one considers thethird component (into the figure) it can be appreciatedthat the flow may not be axisymmetric in the sealregions.

Further detail in the inner cavity region of the rear disccavity is displayed in Figure 13. Anticipated cellularflows are observed.

The final evaluation of flow states concentrates on theimpeller inlet. Figures 14-17 give details of the inletleakage recirculation problem at different positionsclocked between one main blade and the adjacent.Since there are 8 blades, there is a 45° separationbetween each of the main blades. Figures such as 14-17 were prepared in 5° increments between a pair ofadjacent main blades. Four of these are shown here forillustration. Figure 14 corresponds to the θ = 0 positionand the velocity vector pattern appears in a commonlyexpected form. The θ position corresponds to thedistinct line in the inset picture. At 5°, see Figure 15,the study section is shown as a quasi-orthogonal darksheet right at the leading edge of the (bowed) mainblade (see insert). In this case, the effect of the pressurefield from the main blade is felt and some of the flow ispushed up radially into the cavity region. This flow,however, does not make it back even to the first seal

Figure 12. Velocity vectors and streamlines showthe details of the flow in the front seal cavity and inthe rear cavity labyrinth seals.

Figure 13. Velocity vectors and streamlines showthe details of the flow in the back cavity near thedamper seal.

Figure 14. Velocity vectors at impeller eyeinjection location (θ = 0.0°).

Figure 15. Velocity vectors at impeller eyeinjection location (θ = 5.0°).

American Institute of Aeronautics and Astronautics9

tooth. It is simply re-entrained locally. In addition,there is a small backflow recirculation region betweenthe leading edge of the main blade and the front cavitygap. A further illustration at the 10° position is shownin Figure 16. It is observed that some radial outflowstill exists in this region. However, by 15° there is littleor no recirculation and between 15° and 45° the flowpattern is much like that shown in Figure 17corresponding to the 25° location. Hence, the positiveoutflow of gap leakage is constrained to a small sectornear the main blade location.

Closure

This study has presented certain measured dataconcerning the flow through an alternative SSME ATD

first stage impeller and the flow through the adjacentcavities. An unusual phenomenon has been discoveredand identified which corresponds to radial outflow inthe front seal cover gap and also reverse flow at theimpeller tip but with little adverse impact on theremainder of the flow. The computation of the flow inthe front cavity is in excellent agreement with themeasured static pressure change through this regime.The axial thrust prediction was too high, however, itcame reasonably close given the simplicity of theturbulence model. CFD calculations, using multiblockschemes, should form a good basis for accurate thrustcalculations on a closed impeller with the use of moresophisticated turbulence models.

References

1. Hirsch, C., “CFD Methodology and Validation forTurbomachinery Flows,” presented at an AGARDLecture Series, “Turbomachinery Design UsingCFD,” May to June 1994.

2. Hirsch, C., Lacor, C., Dener, C., and Vucinic, D.,“An Integrated CFD System for 3DTurbomachinery Applications,” AGARD-CP-510,1991.

3. Hirsch, C., and Hakimi, N., (Edited by W. G.Habashi), Preconditioning Methods for Time-Marching Navier-Stokes Solvers, SolutionTechniques for Large-Scale CFD Problems, J.Wiley & Sons, 1995, pp. 333-353.

4. Japikse, D., Marscher, W. D., and Furst, R. B.,Centrifugal Pump Design and Performance,Concepts ETI, Inc., 1997, pp. 9-34 to 9-36.

Figure 16. Velocity vectors at impeller eyeinjection location (θ = 10.0°).

Figure 17. Velocity vectors at impeller eyeinjection location (θ = 25.0°).