an integrated approach for distillation column control

7/28/2019 An Integrated Approach for Distillation Column Control

1/12

AN INTEGRATED APPROACH FOR DISTILLATION COLUMN CONTROL

DESIGN USING STEADY STATE AND DYNAMIC SIMULATION

Donald P. MahoneyHyprotech, Inc.

501 Silverside RoadWilmington, DE 19809

Paul S. Fruehauf

DuPont

P.O. Box 6090Newark, DE 19714-6090

ABSTRACT

Steady state techniques have been used for decades to develop control strategies for distillation columns.While theses techniques are effective for screening out clearly undesirable control structures and suggestingviable candidates, they provide incomplete and sometimes misleading information. To accurately assess theperformance and suitability of alternative control schemes, rigorous dynamic simulation is required. This paperpresents an integrated distillation column control design methodology that involves both steady state anddynamic simulation. The limitations of steady state techniques are discussed, and the need for rigorousdynamic simulation for final selection of a workable and robust strategy is illustrated. An integrated simulationenvironment that encourages the proposed design methodology is also described.

KEYWORDS

Distillation Control, Steady State Modeling, Dynamic Modeling, Computer Simulation, Multiple Steady States,Distillate-Bottoms Control


2/12

INTRODUCTION

Steady state techniques are used extensively in thedevelopment of distillation column control strategies.

For complex, multivariable control systems, theRelative Gain Array (Bristol, 1966) has become quitea popular steady state control analysis technique.Other methodologies that involve steady statesensitivity analysis for control purposes have beenproposed by Tolliver and McCune [8], and extendedby Fruehauf and Mahoney [3]. While all of thesetechniques are useful for screening out unattractivecontrol structures, the results they provide about theremaining alternatives are often incomplete andsometimes misleading. This, along with the fact thata large number of industrial columns still operate inmanual or with ineffectual controls, illustrates that

there is a need for improved distillation control designtechniques.

Rigorous dynamic simulation is clearly a moreaccurate way of evaluating process response and thedynamic performance of various control structures.However, it is not the most efficient way of siftingthrough the often numerous possible controlcandidates. By making use of its solving efficiency,we can employ a steady state technique to screenout unattractive control structures and suggest viablecandidates. Then, using rigorous dynamicsimulation, we can discriminate among the smallernumber of remaining alternatives.

In this paper we present an integrated distillationcontrol design procedure that involves both steadystate and dynamic simulation. The steady statedesign methodology presented is useful forsuggesting viable distillation control candidates andscreening out clearly unworkable schemes. Whilethis methodology is an effective control designtechnique and has been applied to many industrialoperations, when used alone it has a number oflimitations as do all steady state controltechniques. We will examine some of theselimitations and illustrate how rigorous dynamicsimulation can be used to complete the analysis byrevealing important operability and controlperformance information. Having presented theintegrated design procedure, we propose criteria fora suitable simulation environment.

STEADY STATE PROCEDURE

The initial steady state screening procedure issimilar to a design technique proposed by Tolliver

and McCune [8]. However, there are a number ofsignificant differences in the methodologies thatwarrant some discussion. First, we propose the useof mass flows for model specifications as opposed tothe typical molar flows. As we will illustrate in alater example, there can be significant differences inthe results when molar flows are used. Furthermore,most industrial columns measure and control massor mass equivalent flows, not molar flows.

Second, when examining temperature sensitivityfor composition control, we impose the actual controlstructure via appropriate selection of steady statespecifications for the model. For example, if we are

proposing the use of mass reflux to controlcomposition, we would explicitly set mass reflux inthe model specifications. Most techniques simplyvary molar distillate in order to gauge temperaturesensitivity regardless of the proposed compositioncontrol variable.

Third, this technique may be used formulticomponent systems to quantify the incrementalbenefit of using on-line analyzers over temperaturecontrol.

This technique deals exclusively with the designof single-point composition controls. By single-point we mean that the composition of only one end

of the column is controlled directly. Dual-pointcontrol involves controlling top and bottomcompositions. The main benefit of dual-point controlis energy savings, however, since these savings areoften too small to justify the added complexity of thedesign, single-point control seems to predominatethe industry.

Having introduced the basis for the steady statedesign and screening procedure, we present thedetailed methodology below.

Step 1 - Developing the design basis.

As with any design effort, it is important to begin byestablishing all of the important criteria that the finaldesign must satisfy. This includes, but may not belimited to identifying and understanding: what the product draw composition specs are

Page 2


3/12

whether the specifications are one- or two-sided1

which stream is the demand stream2

what the expected disturbances to the columnare

what the operating constraints are what the base case or normal operating

condition for the column is.It is particularly important to understand the nature ofthe disturbances that are likely to upset the column.Accurate predictions of feed rate and feedcomposition disturbances are a key element toeventually developing a robust and workable controlstructure. It is perhaps worth mentioning here aswell, that if the design is for an upgrade or expansionto an existing process, it is important to understandthe existing control structure and why it isimplemented the way it is. Reasons for a particular

control structure may be very subtle, yet criticallyimportant to the plant-wide operability. On the otherhand, many controls are left-overs from old designsand have not been changed simply because no onehas bothered to improve them.

Step 2 - Selecting candidate con trol schemes

The second step in the steady state procedure is toselect a candidate control structure. Two-productdistillation is typically viewed as a 5X5 controlproblem. There are five degrees of freedom in atypical two-product distillation column3, representedby:1. feed valve2. reflux valve3. distillate valve4. heat input valve5. bottoms valve.We do not consider pressure controls here since weare normally able to achieve tight pressure control viainerts blanketing and venting, or with low-boilerventing. Under such conditions, pressure may beconsidered fixed, and thus does not affect the

1 One-sided spec's must remain at or below a certain value,

two-sided specs must remain within a specified range.

2 The demand stream is set independently by an upstream or

downstream process and is therefore not available for control

purposes.

3 Because steady state calculations do not take into account

the three inventory variables of condensate level, bottoms level,

and pressure, steady state models for binary distillation have

only two degrees of freedom. For real control purposes,

however, we clearly have five.

degrees of freedom available for the remainingcontrols.

In a 5X5 system, there are 5! or 120 possiblesingle-input, single-output control combinations.

However, once all of the constraints of the processare considered, normally only a few combinations areleft.

First, we must determine which of the streams isthe demand stream: the one which is set by someupstream or downstream process and thereby setsthe production rate for the column. Typically the feedis the demand stream, however, occasionally we seethe distillate or the bottoms being set independently.

Next we examine the overhead and bottomsinventory controls in light of the base case, or normaloperating conditions. We compare the relativemagnitude of the reflux vs. distillate, and bottoms vs.

boilup. If there is a difference of 10:1 or more, wetypically select the larger of the streams to controllevel. An example where this ratio applies is a tarstill. Here we are typically trying to remove a smallquantity of high boiler. It is not uncommon for theboilup to be 100 times the bottoms flow. In this case,the bottoms flow is too small to compensate for manydisturbances, thus the boilup must be used to controllevel.

The next step is to consider the economics andconstraints of the system and select a suitable feed-split control scheme for composition control. If thefeed is the demand stream, and we do not have a tar

still, the structure often reduces to one of the twoschemes shown in Figures 1 and 2.

In Figure 1 we have what is called a direct feed-split scheme. In this case, distillate is manipulateddirectly to control the composition profile. Thisstructure is often used when heat input to a column islimited or must be fixed.

In Figure 2 we have an indirect feed-split controlscheme. Here, the distillate is changed indirectlythrough the heat input-composition controller. If thecontrol temperature is too low, heat input flow isincreased. This increase throws more vaporoverhead and results in an increase in distillate asthe condensate drum level increases. There are twobasic advantages to this configuration. First, thetemperature-boilup loop has a faster closed-loopresponse, and provides superior disturbancerejection. The second advantage is that we canmake use of the condensate tank inventory to

Page 3


4/12

Feed (F)

Bottoms (B)

Reflux (L)Distillate (D)

RefluxDrumLT

SteamReboiler

LT

Condenser

LC

LC

TT TC

FC

FC

Figure 1 - Direct Feed-Split Control

Feed (F)

Bottoms (B)


RefluxDrum LT

SteamReboiler

LT

Condenser

LC

LC

TT TC

FC

FC

Figure 2 - Indirect Feed-Split Contro l

achieve substantial flow smoothing in the distillate tothe benefit of downstream processes4.

Finally, we consider a ratio alternative that mightreduce utility costs or improve the dynamic response.One alternative for the scheme shown in Figure 2would be a controller that keeps the ratio betweenthe feed and the reflux streams constant. Thisscheme is likely to use less energy since for smallerfeed rates, we have less reflux, and thus less heatinput is required.

While we present a procedure here for two-

product distillation, this technique may easily beextended to multiple-draw, multiple-feed columns. Ifwe use a partial condenser with a vent stream, wehave and additional degree of freedom, representedby the vent valve. If this valve is used for pressurecontrol, then the analysis is the same. If we have anadditional draw not used for pressure control, thenwe simply have one more degree of freedom. In thatcase, we may chose to try to manipulate this valve inorder to control another variable, or we may chosesimply to ratio this stream to another key stream.Recognize, however, that as we increase the numberof manipulated variables, we tend to increase thedegree of interaction between controls. This makesthe column more difficult to control and requires alonger time to recover from disturbances. Ratioing

4 To achieve the best flow smoothing, the condensate tank

level control should not be tightly tuned. Proportional-only or

averaging level control tuning is preferred.

additional draws to other key streams avoids suchinteraction and is often adequate.

Step 3 - Open loop testing

The third step in the process involves evaluating theopen loop sensitivity of temperature (composition) tothe candidate composition control variable. The goalhere is to identify a sensitive region in the column fortemperature measurement. By holding all otherinputs constant and running a number of case

studies with different values of the manipulatedvariable, we can generate a family of temperatureprofiles around the base case. Typically, changes inthe manipulated variable of 1%, 2%, 5%, and 10% are sufficient.

Examining the curves, we look for sensorlocations where temperature changes are significant,and linear. We can often control temperaturesaccurately to within 0.5C. Thus a temperaturechange of 1C, while not ideal, is often significantenough. By linear, we mean that temperaturechanges are roughly equal in magnitude when themanipulated variable is changed by the same amountin both directions around the base case.

As mentioned earlier, we advocate using massflows when examining the temperature sensitivity. Inpractice, we control mass flows, volumetric flows, orflows measured by a pressure differential. The lattertwo are essentially the same as mass flow, and aredifferent from molar flows if molecular weight varies.

Page 4


5/12

Figure 3 shows a group of temperature profilesfor an industrial column currently in operation. Usingthe criteria described above, we might select tray 38for temperature control. Tray 10 would be a poor

location as there is no sensitivity to negative changesin the manipulated variable.

If we had additional draws, we could evaluate thesensitivity of temperature to changes in the draw aswell. We may find that controlling temperature withthe side draw is better than using any of the otheravailable streams. As mentioned earlier, we maychoose to try to control an additional variable with theextra draw, or simply ratio the flow to another keystream. If we want to control an additional variable(another temperature for example) we would need toexamine the sensitivity of that variable to changes inthe draw. This kind of analysis should reveal

whether the extra draw represents an attractivecontrol candidate, or whether it is best to simply ratiothe stream to another. Recall again, that the morevariables we try to manipulate independently, themore interactions we have and the more difficult thecolumn will be to control in practice. Dynamicsimulations in this case are particularly useful.

An important issue to consider when selecting atemperature sensor location is its proximity to the endof the column with the more important purityspecification. Normally we prefer to measuretemperature as close to the composition of interestas possible. While there are a number of techniques

in the literature for selecting an appropriatetemperature sensor location [1], we believe thatfurther research in this area is needed. As we willillustrate later, dynamic simulation is particularlyhelpful for identifying a suitable temperature sensorlocation.

A practical implementation detail is in order here.Once a control scheme has been designed, tested,and selected, we recommend specifying two

additional temperature sensors; one theoretical stageabove, and one theoretical stage below the primarynozzle. This accounts for any inaccuracies that mayexist in the model, and is far cheaper to specify at

this point than after start-up.

Step 4 - Closed loop testing

Tray

Temperature[C]

150

155

160

165

170

175

180

185

190

0 10 20 30 40 50 60

Base Case -1%+1%

ControlPoint

Top Bottom

Figure 3 - Temperature Sensiti vity

Once the candidate control structure has beendefined and tested for open loop sensitivity, we areready to perform closed loop testing by subjecting themodel to the expected feed rate and compositiondisturbances. Before we do so, however, we mustdetermine the operating conditions that maximize thedemand for the fixed flow or flow-ratio variable. Touse the structure of Figure 2 as an example, we mustdetermine the value of fixed reflux that is required to

maintain the purity specifications for both ends of thecolumn when the most severe expected feedconditions are encountered. In the case of a ratioscheme, we must determine the ratio required tosatisfy the specifications at the most extreme feedconditions. This may be an iterative process thatinvolves making changes to the feed rate and feedcomposition and observing which conditions placethe highest demands on the column when thecomposition specs are met or exceeded. In manyinstances, we may use our knowledge of distillationto determine these conditions.

Once this is determined (the maximum demand

reflux rate in the case of Figure 2), we must note thetemperature at the selected control location as thiswill be used as our setpoint. At this point, the schemeis fully defined and we are ready to test the structureon the full range of operating conditions. It isimportant here that the actual control structure beenforced on the model by careful selection of thesteady state specifications. For example, the controlstructure shown in Figure 2 would require atemperature specification for the selected tray, andmass reflux specification equal to the mass refluxrate equal to the maximum load rate. Here, the massreflux will be fixed, and the value of heat input will bemanipulated by the steady state solver in order toachieve the temperature setpoint. In the case of aratio scheme, the desired ratio may be maintainedwith the use of an adjust or set operation. Thesame specifications would apply.

Page 5


6/12

To fully test the candidate structures, werecommend testing the model with all combinationsof low, mid, and high values for both feed andcomposition. Figure 4 shows the 9 cases that would

be required for a simple binary distillation column.For multiple components, more cases arerecommended.

While we do not explicitly call it out here, there isan implied Step-5 in this procedure that involvesiterating back to any of the earlier steps 2 through 4in the event that the closed loop testing is notsatisfactory. What we should have after successfullycompleting Step 4, is one or more control structuresthat appear to be viable candidates. The procedureto this point will likely have screened out thoseschemes that are clearly unworkable, leaving onlythose alternatives that satisfy the controlrequirements to the extent that steady state modelingis able. As was mentioned earlier, the designmethodology presented up to this point has beensuccessfully applied to numerous industrial columns.It does, however, provide incomplete information onmuch of the dynamic operability of the candidateschemes, and has several other limitations which willbe discussed in the next section.

LIMITATIONS OF STEADY STATE TECHNIQUES

There are a number of limitations associated withusing a steady state modeling approach for thedesign of an inherently dynamic process. As wehave shown, aspects such as sensitivity and steadystate response to upsets may be revealed throughsteady state design techniques. Attractive controloptions can be identified and clearly unworkable

structures eliminated. However, these studies revealvery little about the dynamic operability of the controlstructures being considered; or the process itself forthat matter. Further, it is often difficult to discriminate

among the viable control alternatives using steadystate techniques alone.Feed Rate(Low, Med, High)

Composition

(1,2,3

)

L1 M1 H1

L2

L3

M2

M3

H2

H3

Figure 4 - Nine Cases for B inary Distil lation

Two very significant shortcomings of steady statetechniques are that 1) the effect of holdups in thesystem are not considered, and 2) the high-frequency, or initial response of the system is notconsidered. These two issues account for many ofthe problems, and much of the misleadinginformation that steady state control designtechniques can produce. To illustrate the concepts,two very interesting cases will be examined.

Ignoring ho ldup effects - the Distillate-Bottoms

control structure

If we were to devise a control strategy aimed atcontrolling both the top and bottoms compositionsusing the distillate and bottoms streams respectively,we might end up with a control structure similar tothat shown in Figure 5. While this is clearly aconfiguration that we can set up physically, is itpossible to actually control the process using thisscheme?

From a purely steady state standpoint, theanswer would be no. At steady state, the relationshipbetween the feed rate, and the distillate and bottoms

flows are not independent. Here, the relationship F =D +B must hold, and thus L and D may not be varied

LC

Feed (F)

Bottoms (B)


RefluxDrumLT

SteamReboiler

Condenser

LC

XC

FC

XC Figure 5 - Distillate-Bottoms Control

Page 6


7/12

independently. This relationship is so fundamental tothe steady state description of distillation that, untilonly recently, control experts considered this strategyimpossible and unworkable. Those familiar with

the Relative Gain Array, or RGA technique for controldesign will recognize that the gain matrix using theDistillate-Bottoms strategy is singular; thus provingthat such a control strategy is not possible. Howeverin 1989, Finco et al. [2] demonstrated for the first timethat the Distillate-Bottoms structure is indeedfeasible. Papastathopoulou et al. [6] proposed atuning methodology, and Skogestad et al. [7] haveprovided an explanation for reason it actually works.

In short, the liquid lag from the top of the columnto the bottom due to holdup, effectively decouples thedistillate and bottoms responses at high frequency[7]. High frequency, or initial response, is where

most of the control action takes place. Thus theDistillate-Bottoms control scheme is not onlypossible, but it turns out that it is relatively easy tocontrol [7]. The notion that steady state data mayprovide misleading information for control analysis issomewhat intuitive, yet it is often forgotten whendeveloping multivariable controls.

Steady state techniques are clearly valuable, butmust be used with caution and good judgment. Thenext example illustrates how even steady statesensitivity may be misleading, and that it is importantto examine control system initial response usingdynamic simulation.

Ignoring high-frequency response - the control of

columns exhibiting multiple steady states

As part of the steady state screening procedure, wedescribed the need to use mass flows as inputs tothe steady state sensitivity analysis. The reasonsbeing 1) we typically control mass or mass equivalentflows in practice, and 2) we can get significantlydifferent results when molar flows are used.

Since molar flows enter directly into the materialbalance calculations used in the column solver andthus determine the separation, we are often inclinedto work on a molar basis. Further, we often considermass reflux to increase monotonically with molarreflux, thus making them rough equivalents forcontrol purposes. While this is often true, aninteresting example that illustrates where it is not, isthe case of distillation columns that exhibit multiplesteady state solutions when mass inputs arespecified. The development explaining the multiple

steady state behavior for binary distillation with fixedboilup is shown in Appendix A.

Figure 6 shows the steady state sensitivity ofoverhead purity for changes in mass reflux with fixed

heat input to the column. Notice how between about0.001% and 0.35% overhead impurity, we see areversal in sign of the mass reflux effect. If we usedthis sensitivity plot to decide how to control overheadimpurities in the neighborhood of 0.01%, we might beinclined to set up our controls to decrease massreflux, in order to increase overhead purity. After all,our steady state analysis shows that decreasingreflux here should reduce overhead impurity.

As it turns out, all of this steady state data hasvery little impact on the way the column actuallycontrols. Dynamic simulation of such a systemreveals that the high frequency response is much

different from the low frequency, or steady stateresponse. In fact, the high frequency, initial responseis almost the same regardless of whether mass ormolar reflux is manipulated [5]. To see what actuallyhappens, it may be helpful to consider a casedisturbance.

Consider a column that has lined out to somesteady operating point. We now make a stepincrease in the overhead impurity set-point. If our

control action is to decrease mass reflux, Lw, we willsee an initial decrease in molar reflux, L, since themolecular weight is unchanged (Lw = L MW). Thishappens quickly and correctly increases overhead

impurity by some amount, yD1, regardless of thesteady state compositional effects at play. As theincrease in overhead impurity increases themolecular weight, the molar reflux actually begins todecrease further. If this second decrease in molarreflux causes a further increase in overhead impurityby another yD1 or more, the control action mayincrease mass reflux back above its original value.

Mass Reflux Flow [kg/h]

OverheadIm

purity

[MassFrac]

0.0001

0.001

0.01

0.1

1

5500 6000 6500 7000 7500 8000

Figure 6 - Overhead Impurity vs. Mass Reflux

Page 7


8/12

Thus as the controller lines out again, we mayactually end up with a higher mass reflux with theincreased overhead impurity. Whether this occurs ornot depends largely on the values of dy LD / and L

(see Appendix A).Figures 7 and 8 show the simulated5response of

a mass reflux controller to a step increase in theoverhead impurity set-point. Figure 7 shows thetypical response which we observe in the lowimpurity region (below 0.001% overhead impurity).Here, as we might intuitively expect, the mass refluxdecreases initially, and lines out to a lower value asthe higher impurity set-point is achieved. Figure 8shows the response of the same control structure,however this time, in the region of steady state signreversal (between 0.001% and 0.35% impurity). Theinitial response here is the same as that shown in

Figure 7. However in this case, we line out to ahigher value of reflux after the higher impurity isreached. This inverse response clearly illustrates thedifference between the initial, high frequencyresponse, and the steady state response. Since

5 These trend charts were taken from a rigorous dynamic

simulation of a real process using HYSYS from Hyprotech,

Ltd.

control is based extensively on high-frequencybehavior, it is important to use steady stateinformation intelligently, and examine the dynamicsof the system before selecting the controls.

THE DESIGN APPROACH CONTINUED

As these examples have shown, steady stateanalysis for control, when used alone, can provideincomplete and sometimes misleading information.We have proposed a steady state screening anddesign procedure which is useful for eliminatingundesirable structures, and suggesting viablecandidates. Using dynamic simulation to rigorouslymodel the remaining control structures, we have anintegrated design approach that is both efficient andaccurate.

The dynamic simulation part of the designprocedure does not lend itself as much to a detailedstep-by-step process as the steady state part does,however, we can trace some general steps.

Step 5 - Supplying holdup information

Mass Reflux

Overhead Impuri ty

Figure 7 - Typical response to step decrease in

overhead purity set-point

Steady state simulations do not typically involveholdup information in the solutions. Thus, in thedynamic analysis, we typically begin by assigning therelevant holdup information to condensers, traysections, reboilers, and any other ancillary equipmentthat possesses a material holdup. Since holdup

volumes play a critical role in defining the systemtime constants, and thus influence response times,disturbance rejection, and overall controllability, it isimportant to specify holdups properly.

Mass Reflux

Overhead Impurity

Figure 8 - Inverse response to decrease in overhead

purity set-point

Step 6 - Identifying the dominant dynamics

Controls and instrumentation are also not normallypart of a steady state model, thus we must considerthe measurements and actuators that we are likely tohave at our disposal when we go to operate theplant. While it is important to impose each candidatecontrol structure on the model in the same way that itis likely to be implemented in the plant, we do notwant to introduce unnecessary complexity. Forexample, we may need to add lags or dead times toprocess measurements if they are likely to beupdated only periodically from lab samples. Dead

Page 8


9/12

times have a very large effect on closed loopdynamics and therefore need to be accuratelyrepresented. We may also need to add cascadecontrols in loops where they will apply in the real

plant. These are likely to influence the dynamics to asufficient extent that they ought to be included in themodel. Other pieces of instrumentation andequipment will not significantly affect the dynamics ofinterest, and thus need not be modeled. Thedynamic response of instruments such as valves oron-line sensors, for example, are seldom includeddue to their negligible effect on the dynamics thatdominate the process. The level of detail used in thesimulation must be driven by a judicious balancebetween accuracy and usability.

Step 7 - Applying the control struc ture

Once we have supplied holdup information, andidentified the important additional dynamics, we mayapply the candidate control structure to the processmodel. One of the benefits of having suppliedaccurate holdup information and significantinstrumentation dynamics, is that we may now useone of many tuning techniques6 to generatepreliminary controller tuning constants.

Step 8 - Exercising the model

Once we have the control structure implemented,

and the controllers tuned, we are ready to exercisethe model. At this point we are able to evaluate anumber of different scenarios. Start-up and shut-down performance may be studied. Also, feed rateand composition upsets, as well as other likely loaddisturbances may be applied to the model to test thedynamic operability and disturbance rejectioncapabilities of each control scheme underconsideration. It is a good practice to design a suiteof test disturbances to which all of the candidatecontrol structures will be subjected. This provides acommon basis useful when comparing theperformance of competing schemes.

Step 9 - Selecting the strategy; and beyond

Steps 7 and 8 may be repeated for each candidatecontrol strategy. While the best strategy may emerge

6 Ziegler-Nichols, the IMC, and the ATV, or Auto-Tune

Variation Technique have been found useful.

quite clearly in some cases, the differences in controlperformance are often subtle, and may involve trade-offs between things like ease of start-up and shut-down, disturbance rejection capabilities, response to

production rate changes, complexity of the controlstrategy, the degree of interaction between controls,and the number of controllers required. Theseissues and others must be considered carefully bythe designer, and hopefully by others involved in thesubsequent operation of the process.

Once the final control scheme is selected, thereare still a number of ways in which dynamicsimulation can be extended to add value. Byallowing links between the computer processsimulation and various DCS platforms, the dynamicmodel used to define the process controls may beused to check-out the commissioned control strategy

resident on the DCS. This exercise not only checksthe integrity of the control software configuration, butalso allows for preliminary tuning, and operatortraining. Beyond plant start-up, such a link may alsobe used as part of an on-line control system as wellas for on-going process improvement andoptimization studies.

SIMULATION CRITERIA

While the appeal of an integrated steady stateand dynamic control design approach is strong, therehave been no commercial simulation environments

available to encourage such activity. Havingexamined some of the major issues involved indesigning robust and workable control strategies, wecan begin to see what a truly useful simulationenvironment should look like.

Accurate. Clearly a solid engineering foundationis critical. Without accuracy, modeling is not only awaste of time, but may make predictions that couldlead to poor design decisions. Creating this solidfoundation, however, is more than simply capturingthe latest process engineering knowledge andpresenting it together in a simulation package. Itrequires a judicious balance of rigor and performancethat yields a tool that is at the same time useful andusable.

Integrated steady state and dynamics. As wehave illustrated, the benefits of having both steadystate and dynamics functionality available togetherare clear. While independent steady state anddynamic simulation packages are often positioned asproviding seamless integration, many actually make

Page 9


10/12

use of flat file exchanges and do not share acommon simulation environment. For an integratedsystem to be truly useful, it must be able to eliminatethe duplication of effort experienced when separate

models are required for each mode. Ideally, onebuilds the model topology once, then executessteady state or dynamic solvers depending on theneed.

Fast. Experience has shown that much of theprocess understanding that comes from simulationoccurs during the model-building phase. Interactionwith, and immediate feedback from the model arekey elements to the effectiveness of any processmodeling endeavor. Submitting runs batch-style andwaiting for results is not only inconvenient andinefficient, but it removes the valuable live linkbetween the engineer and the model.

Broadly Applicable. Ideally, all of the functionalrequirements for all types of applications should beavailable in one place. Conceptual design, processdesign, dynamic operability analysis, control strategydevelopment, DCS interfacing and check-out,operator training, and on-going process improvementfor all kinds of processes should share a commonenvironment. This environment, however, needs tobe more than a group of functional engineering toolsartificially linked through file sharing or swapping. Byseamlessly integrating all of the functionalcapabilities into a single environment, informationgenerated in one mode is fully and immediately

available for all others.Easy to Learn and Use. In order to break the

barriers that prevent its wide spread use, processsimulation must be both easy to learn and to use. Byusing an intuitive, graphical user environment, and acomprehensive selection of configurable unitoperation modules, simulation tools can maketremendous improvements in this area. Literallythousands of man-years of process engineering andmodeling experience have been accumulated in thesimulation industry. With creative ways of packaging

this information available, there are very few reasonswhy engineers should have to write custom code orcompile input files or subroutine calls in order to runsimulations. By making simulation technology easy

to use and available to all engineers via configurablemodules, we are placing the process understandinginto the hands of those who are most able to put it toeffective use.

Hyprotech, Ltd. is taking a lead in this area ofprocess simulation. Based on the criteria discussedabove, Hyprotech has developed an integratedsimulation environment called HYSYS thatcombines steady state and dynamics functionality inone package, and also provides links to popular DCSplatforms for control system check-out, operatortraining, and the potential for on-line dynamic model-based control. We believe that this technology will

have a tremendous impact on not only the wayengineers approach control strategy development,but on how we approach modeling in general.

SUMMARY

We have presented a proven steady state screeningand design technique for distillation column controls.We have also highlighted some of the limitations andweaknesses of such methods when used alone.Dynamic simulation completes the analysis byproviding the necessary high-frequency or initialresponse information that allows for proper

development, evaluation, and selection of candidatecontrol structures. Having presented an integrateddesign solution, we saw the need for new simulationenvironments that encourage such a designmethodology. The features of such simulation toolsinclude: accuracy, integration of steady state anddynamics functionality, fast execution, broadapplicability, and ease of use.

REFERENCES

1. Buckley, P. S., Luyben, W. L., Shunta, J . P., Design of Distillation Column Control Systems. ISA, Research Triangle

Park, NC, 1985

2. Finco, M. V., Luyben, W. l., Polleck, R. E., Control of Distillation Columns With Low Relative Volatilities. Ind Eng

Chem Res, V28, n1, J an 1989, pp. 75-83.

3. Fruehauf, P. S., Mahoney, D. P., Improve Distillation Column Control Design. Chemical Engineering Progress,

March, 1994.

Page 10


11/12


12/12

APPENDIX A

Development of the multiple steady state conditionfor binary distillation with fixed mass inputs [5].

L LM M y MW y MWD D= = + 1 1( ) 2

where Lw = mass refluxL = molar refluxM = reflux molecular weightyD = mole fraction of light keyMW1 = molecular weight of light keyMW2= molecular weight of heavy key

If we take the partial derivative ofLw with respect toL, we have:

L

LM L

M

L

L

LM L MW MW

y

L

w

w

= +

= + ( )1 2D .

IfMW1 >MW2, as is often the case, then

L

Lw may

be negative depending on the difference in molecular

weights, the magnitude ofdy

LD

, and the magnitude

ofL. A negative sign here suggests that under someconditions, increasing mass reflux will actuallydecrease molar reflux. Figure A1 shows therelationship between mass reflux and molar reflux fora system exhibiting multiple steady state solutions.Figure A2 shows how overhead purity changes with

changes in molar reflux (i.e.,dy

LD

). Recall that this

relationship, along with the difference in molecularweights and the value of L determines the sign of

L

Lw . At very low and very high overhead impurities,

dyL

D

is small (see Figure A2), and wee see that

molar reflux increases monotonically with increasingmass reflux. However, between about 0.001% and

0.35% overhead impurity,dy

LD

becomes very large

(steep slope in Figure A2), thus driving

L

Lw

negative such that molar reflux actually decreaseswith increasing mass reflux.

Molar Reflux Flow [kgmol/h]

O

verheadImpurity

[MassFrac]

0.0001

0.001

0.01

0.1

1

120 140 160 180 200 220 240 260

Figure A2 - Overhead Impuri ty vs. Molar Flow

Mass Reflux Flow [kg/h]

M

olarRefluxFlow

[kgmol/h]

120

140

160

180

200

220

240

260

5500 6000 6500 7000 7500 8000

Figure A1 - Molar Reflux vs. Mass Reflux

Page 12

an integrated approach for distillation column control

Documents