DISTILLATION COLUMN CONTROL DESIGN USING STEADY STATE MODELS: USEFULNESS AND LIMITATIONS

Transcription

1 DISTILLATION COLUMN CONTROL DESIGN USING STEADY STATE MODELS: USEFULNESS AND LIMITATIONS Paul S. Fruehauf, PE Engineering Department E. I. du Pont de Nemours & Co., Inc. P.O. Box 6090 Newark, Delaware & Donald P. Mahoney Hyprotech, Inc. 501 Silverside Road Wilmington, DE KEYWORDS Computer Aided Engineering, Simulated Distillation, Chemical Processing, Distillation Control, Steady State Modeling ABSTRACT Steady state models continue to be powerful and efficient tools for designing control systems for distillation columns. This paper presents a control design procedure and an example application of this technique to an actual column.

2 INTRODUCTION Steady state process models have long been used to assist the control engineer in designing control strategies for distillation columns. However, with the large number of industrial columns still operating in manual or with ineffectual controls, there remains a need for sound distillation column control design techniques. We believe that Tolliver and McCune (1978) have made the greatest contribution to the development of this type of design procedure. Two other very good papers on this subject are by Thurston (1981) and Roat, et al (1988). While our procedure is an extension of that proposed by Tolliver and McCune, we have improved the procedure in the following ways: We advocate that mass flows be used in models versus the previous standard of molar flows. We have determined independently that use of molar flows can lead to incorrect results. A recent review article on distillation column control by Skogestad (1992) confirms these findings. We also advocate that the actual control structure be enforced when using the steady state simulation to identify a temperature sensor location for composition control. This is accomplished by a careful choice of independent variables when defining the model solution conditions. Tolliver and McCune advocate varying only molar distillate flow regardless of the proposed control structure. This too can lead to incorrect results. We show that this technique can be used for multicomponent columns to quantify the incremental benefit of composition control using on-line analyzers versus temperature control. This paper deals exclusively with the design of single point composition controls. The vast majority of columns have one sided composition specifications; those in which a single point composition control scheme can keep both top and bottom product compositions at or below limits for a wide range of disturbances. This does not have to be accepted on faith because the design procedure explicitly tests this hypothesis. The predominance of one sided specifications leaves the main incentive for dual point control schemes to be energy savings. In most cases, the energy savings is small and does not justify the added difficulty of implementing and maintaining dual point control. Luyben (1975) presents the potential energy savings for many different types of separations. Additionally, dual point schemes often have significantly longer recoveries from upsets due to interactions between the control loops. We believe it is appropriate to contrast steady state and dynamic models as control design tools. While both tools have a place, we have found that using steady state models coupled with experience and a general knowledge of distillation column dynamics is adequate for many problems and can be more efficient than using dynamic models. For a good development on the rationale behind using steady state models refer to the chapters on Quasi-Static Analysis in Rademaker, et al (1975). One obvious limitation of steady state modeling is that it tells us nothing about the dynamic response, making it difficult to compare the dynamic disturbance rejection capability of alternative control schemes. When we encounter a difficult and important problem we invest the extra engineering time to develop a dynamic model. The ideal design tool would be one that has both steady state and dynamic capabilities. This tool would provide the efficiency of steady-state analysis, but would also have the added benefit of comparing the disturbance rejection capabilities of different schemes with the dynamic model. The combined tool would allow the designer to perform both tasks without requiring an investment in time to develop two different models. A new product soon to be released by Hyprotech, Ltd. will combine steady state and dynamic modeling in one such package.

3 Our design procedure can be best thought of as general approach rather than a single detailed procedure that covers all cases. The procedure must be adapted to each problem because there are many different types of distillation and almost every industrial problem usually has some unique requirement.

4 In this article, we begin by presenting some background material on distillation column control and the use of steady state models. Next, we describe our design procedure in detail. We then conclude by illustrating the design procedure with an actual applied industrial example. BACKGROUND Distillation Column Control Fundamentals - Figure 1 illustrates a schematic of a simple distillation column. This figure identifies the nomenclature used in this paper and points out the five valves available to control the column. There are five degrees of freedom in a typical binary distillation column which are represented by the feed valve, the steam valve, the reflux valve, the distillate valve, and the bottoms valve. Figure 1 - Schematic of Simple Distillation Column The five valves are used as follows. First, either the feed, the bottoms, or the distillate rate is set independently to define the production rate of the column, thereby eliminating one valve. We call this the demand stream. The reflux drum and the column bottoms level must be controlled, requiring two more valves. This leaves us with two compositions to be controlled with two valves. Traditionally, simple distillation is viewed as a 2x2 control problem because the remaining two composition control loops have strong interactions. No matter what valves we use for composition control or how we use them, fundamentally there are two things that we can manipulate: the feed split and the fractionation. An overall material balance for a column tells us that the distillate flow plus the bottoms flow must equal the feed flow. The feed split is simply the amount of feed that leaves as distillate versus the amount that leaves as bottoms. The other fundamental manipulative variable is the fractionation which is the amount of separation that occurs per stage. The overall fractionation in a column depends on the number of stages, the energy input, and the difficulty of the separation. In order to explain how we pick the single point control schemes it is necessary to show the relative effect of the feed split and fractionation on product compositions. The assumption is that the control objective is to produce high purity products in both ends of the column. This is the objective for the vast majority of cases that we work on. In Figure 2, we use some numerical examples to show the relative importance of the two manipulative variables.

5 50 pph 49.5A 0.5B 40 pph 39.7A 0.3B 100 pph 50A 50B 100 pph 50A 50B 50 pph Base Case 0.5A 49.5B 60 pph 20% Feed Split Change 10.3A 49.7B 50 pph 49.6A 0.4B 40 pph 39.8A 0.2B 100 pph 50A 50B 50 pph 0.4A 49.6B 20% Fractionation Change 100 pph 50A 50B 60 pph 10.2A 49.8B 20% Feed Split Change 200% Fractionation Change Figure 2 - Numerical Examples Showing Importance of Feed Split and Fractionation In the base case (top left), 100 pph is fed to the column made up of 50 pph of A and 50 pph of B. The feed split is such that 50 pph leaves as distillate and 50 pph leaves as bottoms. The column has sufficient heat input to produce enough fractionation to obtain 99% purity in the top and bottom of the column. In the second case (top right), the feed split is changed by 20% so that 40 pph leaves as distillate and 60 pph leaves as bottoms. The feed rate and composition are the same. In this case, we obtain 99.2% purity in the distillate but only 82.8% purity in the bottom. The explanation is that only 40 lb of the 50 lb of A fed to the column, is being allowed to leave in the distillate stream. The remaining 10 pph of A has to leave the column and does so by forcing its way down the column and going out in the bottoms stream. This substantially reduces the bottoms purity. In the third case (bottom left), the feed conditions are the same, but now the fractionation is increased by 20%. In this case, we obtain slightly higher purities in both the top and the bottom of the column. In the last case (bottom right), a 20% feed split change is made simultaneously with a 200% fractionation increase. In this case, we obtain very high purity A in the overhead, but only 83% purity in the bottom. Again, the 50 lb of A fed to the column must leave the column and does so in part by forcing its way down the column and leaving in the bottoms stream, again drastically reducing the bottoms purity. This leads us to a very important concept in distillation column control. The feed split to the column is the most important variable to control; it must be right in order to achieve high purities in both the top and bottom of the column. While fractionation must be great enough to obtain the desired purity, it is only used to fine tune composition control. From this development it should be clear that adjusting feed split is equally important when feed composition and feed rate changes hit the column. When we select the manipulative variable for composition control, we must make sure that it is able to adjust the feed split. In this paper, we are not going to discuss pressure control. Pressure control is a subject in itself, and it is independent of our main topic of discussion. In almost all columns, we are able to achieve very tight and responsive pressure control so that it can be considered a constant. This is one way in which our approach differs from a popular academic approach which includes pressure control as one of the things to be controlled and coolant flow as one of the things to be manipulated. The academic approach always assumes the feed is the demand stream.

6 Specialized and Multicomponent Distillation Columns We have applied this general procedure to many different types of specialized columns including homogeneous and heterogeneous azeotropes, extractive distillation, strippers and absorbers and multicomponent columns. We have also used this procedure for many different column configurations including columns with either liquid or vapor side draws, columns with partial condensers and with both packed and tray columns. Because we often encounter columns with multiple components in the feed, a little more should be said about these cases. In multicomponent columns, unlike binary columns, fixing temperature and pressure does not fix composition. In spite of this limitation, temperature control can still be used to meet many composition specifications. Often this results in larger yield losses or higher energy consumptions than if an on-line analyzer was available for control. This is where steady state models can be very helpful to us because we can use them to quantify the incremental benefit of on-line analyzers versus temperature control. In one case, we used this technique to document the yield improvement to be gained from the addition of an on-line analyzer. The savings was over two hundred thousand dollars a year. Steady State Distillation Models - Steady state models are easily manipulated and provide robust solutions. In order to make a change to the solution conditions, only a few changes need to be made to the model input file. The model input file is then submitted to the software which finds a new solution. Generally, very little time is spent getting converged solutions, which allows us to efficiently generate the large number of case studies necessary for this design procedure. DESIGN PROCEDURE We have extended the design procedure reported by Tolliver and McCune (1978). The design procedure is composed of the following steps: Step 1 Develop design basis Step 2 Select a candidate control scheme. Step 3 "Open loop" test using model to find a candidate temperature sensor location. Step 4 "Closed loop" test candidate control scheme for feed rate and feed composition disturbances. Step 5 Objectives met? We have included a Step 5 to illustrate that the procedure can be iterative. If the objectives are met, then the procedure is complete. Otherwise, we might return to Step 2 and select a different candidate control scheme or we might return to Step 3 and select a different candidate sensor location. The step that we return to depends on the nature of the problem. The first four steps of the design procedure are explained in detail in the following sections.

7 Step 1 Develop Design Basis - Like any design effort, the ideal first step is to completely define the design basis providing all the information needed to select the best design alternative. This basis should be a contract between the client and the designer. The accuracy of the basis is mainly the client's responsibility. The components of the design basis are summarized in Table 1. The first piece is the product composition specifications for the top and bottom of the column. We need to know if the specifications are one or two sided. A one sided specification means that we need to meet or exceed a product composition specification. A two sided specification means that we need to keep a composition within a certain range. For example, a two sided specification would require that the product composition stay within ppm. A one sided specification is much more common. We have encountered only one column that has a two sided specification. It is also important to know the reasons for the specifications. Occasionally, some specifications are picked arbitrarily simply to have a sizing basis for a column and we find that tight control is not critical. The design basis also includes the economic considerations and the disturbances to the column. As you will see, a good definition of the range of feed rate and feed composition disturbances is required to complete the design procedure. The next element of the design basis is the constraints. We need to know which of the streams will be the demand stream. If the design is a retrofit of an existing column, we need to know how the column is currently controlled and why. This is important because if we see a need to change the control strategy, we need to make sure that the change will not upset the overall control strategy for the process. The reasons for a given control strategy can be very subtle, particularly if there are recycle streams in the process. Other constraints include those imposed by the upstream and downstream equipment and recycle streams involving the products from the distillation column. If the process has recycle streams, the approach is to draw a box around the process so that the recycle stream remains inside the box and then analyze that part of the process as a system. Although the design procedure for recycle systems is not covered here, it is just an extension of this procedure. The last part of the design basis is simply the base case that was originally used to size the column. Step 2 Select a Candidate Control Scheme - The second step of the design procedure is to select a candidate control strategy. In a 5x5 system, like simple binary distillation, there are 120 possible single input, single output control combinations. And this is without considering combinations of process outputs as variables. Fortunately, in most situations, only a few combinations are left after everything (constraints, economics, etc.) is considered. The control strategy selection procedure is as follows. It is very similar to the procedure outlined by Buckley, et. al. (1985). First, we need to determine which of the feed, bottoms or distillate streams will be the demand stream. Generally, the feed is set as the demand stream. Second, we need to determine how the column base and reflux drum level will be controlled. This is done by comparing the relative magnitudes of the reflux flow versus the distillate flow and the boilup flow versus the bottoms flow. If there is a 10 to 1 or greater difference, then the level needs to be controlled by manipulating the larger stream. One common situation where this

8 consideration is important, is in a tar still. In a tar still, we are usually trying to remove a small quantity of a high boiling material. The boilup can be 100 times that of the bottoms stream. The bottoms stream is too small to compensate for common disturbances, therefor, the base level must be controlled by manipulating the steam flow, rather than the very small bottoms stream. In most cases, we use a column tray temperature to infer composition. We use temperatures because the measurement is inexpensive, highly reliable, repeatable, has a high degree of resolution, is continuous, and is generally an excellent indicator of column product compositions. The third step in the control strategy selection process is to consider the constraints, the economics, and dynamics all simultaneously and pick the best feed split control scheme. If the feed to the column is the demand stream and we do not have a tar still, the control strategy choice usually comes down to two very common choices illustrated in Figures 3 and 4. Figure 3 - Direct Feed Split Control Scheme In Figure 3, we have a direct feed split control scheme. Product compositions are controlled by fixing a column temperature. The temperature controller manipulates the distillate flow. This control scheme is often selected when the heat input is limited or must be fixed. This is a direct feed split control scheme because the distillate is manipulated directly to control composition.

9 Figure 4 - Indirect Feed Split Control Scheme The second common choice is illustrated in Figure 4. The compositions are controlled by a temperature controller which manipulates the steam flow. This is an indirect feed split scheme because the distillate flow is increased indirectly by increasing the steam flow. This control strategy alternative has two advantages. One is that it generally has faster closed loop response (i.e., a shorter natural period) and therefore provides better disturbance rejection. The second is that since the reflux drum level sets the distillate flow, the reflux drum can be used to smooth flow disturbances to other downstream unit operations. To achieve flow smoothing the level controller must have averaging level controller tuning. The last part of the control strategy selection process is to select a ratio-control alternative that might use less energy than the primary alternative. One example of a ratio control alternative for the scheme illustrated in Figure 4 would be a controller that keeps a constant reflux to feed flow ratio. This scheme will likely consume less energy than the non-ratio alternative because as the feed flow to the column decreases, the amount of reflux will decrease. Less reflux will require less heat input. Step 3 "Open Loop" Test Using Model To Find Candidate Temperature Sensor Location - The third step of the design procedure involves what we have termed "open loop" testing. The purpose of the "open loop" testing is to use the steady state model to identify a candidate temperature sensor location for inferred composition control. This is accomplished by changing the temperature control manipulative variable up and down from the base case value. Figure 5 provides a plot of three steady state runs where temperature is plotted versus tray number. Figure 5 - Steady-State Temperature Profile Sensitivity to Changes in Distillate (Fixed Heat Input) In this example, the candidate control strategy is to fix the heat input and vary the distillate flow to control composition (i.e., the Figure 3 scheme). The three temperature profiles shown in Figure 5 are a base case profile, one where the distillate flow is increased by 1%, and one where the distillate flow is decreased by 1%. As you can see, when the distillate flow is increased, some tray temperatures increase. This is because the concentration of the high boiling point material is higher in the distillate which shifts the composition profile up, producing higher temperatures in the top trays. When we decrease the distillate, the opposite effect occurs. The temperature sensor location we picked was tray 38. In this particular case, we looked for a location where the temperature change was significant and nearly equal in both directions. Tray 10 would be a poor location because there is no temperature sensitivity to negative changes in the distillate flow. These temperature profiles are taken from the design procedure used for an actual column that is currently operating and giving very good control performance. The performance is reported as example number two in Chien and Fruehauf (1990). The amount to change the temperature control manipulative variable varies from case to case. We usually start with +/- 1,2,5 & 10% and observe how much the temperature profiles move.

10 In general we prefer to measure temperature for control purposes nearer to the end of the column with the most important purity specification. Additionally, the temperature at the sensor location ought to be reasonably sensitive to changes in the manipulative variable, and should vary linearly with increasing and decreasing values of the manipulative variable. We have found that in some cases we can control tray temperature to within a half of a degree centigrade. Based on this performance, we consider a plus or minus one degree change at a given location to be sufficient for temperature control. In the literature, there are many different techniques proposed on how to locate a temperature sensor. One of the many techniques is reported by Shunta and Luyben (1975). This is one area where we feel more research work is needed to find a single reliable method. We have found that the equal temperature change rule used in the above example does not always determine the best location. This is one area where steady state techniques are limited. Although it is not fully tested, we believe that a technique that uses dynamic modeling may be superior. The technique would involve plotting temperature profiles at fixed time intervals for step increases and decreases in the temperature control manipulative variable. To do this efficiently, a modeling tool which has both dynamic and steady state modeling will be required. When looking for a candidate temperature sensor location it is important to fix the other flows per the candidate control strategy. If we are fixing a reflux ratio as part of our control strategy, it is important to set the steady state model solution condition so that the reflux ratio is fixed. It is also important to fix mass flows and not molar flows. All previous reports of design procedures using steady state models have incorrectly set molar flows. If we set molar flows, we can get incorrect results when the molecular weights of the components are different. In practice, we control mass flows, volumetric flows, or flows measured by the differential pressure generated by an orifice plate. The last two are essentially the same as mass flow and are different from molar flows when the molecular weights are different. One example that illustrates the problem with using molar flows is that the use of molar flows does not predict the existence of multiple steady states that have recently been shown to exist in some columns where mass flows are controlled. After we complete the design procedure and finalize the temperature sensor location, we routinely specify three temperature nozzles. In addition to the primary nozzle, we install two extra nozzles, one theoretical stage above and one below the primary nozzle. These are added to accommodate for any small inaccuracies in the model-predicted results. Adding a nozzle to a column after installation can be very expensive. One column we worked on had a Hastelloy C liner. Adding a nozzle required welding inside the column to retain the integrity of the lining. This is very costly because it requires a vessel entry as well as packing removal and reinstallation. Step 4 "Closed Loop" Test Candidate Control Scheme for Known Disturbances - The last step in the design procedure is to perform what we have termed "closed loop" testing. In this step, we use the steady model to simulate the candidate control strategy and test it for feed composition and feed flow changes. The control strategy is simulated by setting the model solution specifications to mimic the way the column is controlled. For example, if we are considering a control strategy where the reflux flow is fixed and a tray temperature is controlled with steam flow, then the steady state model solution conditions would be to set the mass reflux flow and tray temperature. The first step is to find a set of operating conditions that meet or exceed the product composition specifications for all expected feed rates and compositions. This can be an iterative process, however in many situations we can use our knowledge of distillation to help find these conditions. Again, if we are fixing the reflux flow, we need to find the feed conditions that require the highest reflux flow to satisfy the product composition specifications. The reflux flow will then be set at this value for all other feed rate and feed composition combinations. We also must find a

11 temperature setpoint that will keep the top and bottom product compositions at or below specifications for all expected conditions. Next we test the model with these operating conditions for various feed composition and feed rate combinations. We recommend that a minimum, a middle, and a maximum feed rate and feed composition be selected which defines nine cases to be simulated. For some multicomponent cases, more than one feed composition may need to be varied independently. If two compositions are varied then 27 cases must be simulated. The case study features of the steady state models make it easy to run the large number of cases needed. If the top and bottom product specifications are met or exceeded for all cases, then we have found an acceptable control strategy for the column. The last step in the closed loop testing procedure is to compare the energy consumption of the candidate scheme, the a ratio alternative, and the minimum. The minimum energy consumption is found by setting the top and bottom product purities exactly at the specifications. If the energy cost difference is large between the minimum, the candidate control scheme and the ratio alternative, then another candidate control scheme should be evaluated to see if a more energy efficient scheme exists. INDUSTRIAL APPLICATION OF THE DESIGN PROCEDURE Example - Methanol-Water Column - Figure 6 - Methanol-Water Distillation Column Schematic To illustrate the design procedure, we have selected an example of an actual column from one of our plants. This particular column, used to separate methanol and water, is illustrated in Figure 6. The split-column arrangement exists to accommodate another process which happens to share the equipment. For our analysis, however, the system may be thought of as one large column. Step 1 is to define the design basis including product specifications, economic considerations, and typical disturbances and constraints. This column has the following one sided product specifications: less than 100 ppm water in distillate less than 100 ppm methanol in bottoms

12 The economic considerations amount to 8000 pph of steam that the process consumes during 1/3 of the year; not a large incentive for minimizing energy usage. The disturbances and constraints are: Feed Rate: 4 to 26 Kpph Feed Composition: 15 to 60% methanol plus some salts Demand stream: Column Feed Upstream/Downstream Equipment: Feed Tanks No recycle streams Having defined the design basis, Step 2 is to select a candidate control strategy. In this example, the feed is the demand stream. Comparing the relative magnitude of reflux versus distillate (L/D = 2.0) and boilup versus bottoms (V/B =.5) shows that we have no restrictions on the choice of level control variables because the ratios are significantly less than 10. The candidate control strategy selected is the same as that illustrated in Figure 4. Here, the feed and reflux streams are flow controlled, the reflux drum level is controlled by manipulating distillate flow, and the base level is controlled by manipulating the bottoms flow. Composition is regulated by a temperature controller that manipulates steam flow % Temperature [C] Base Case +1% -5% +0.5% Control Location Top Bottom Theoretical Stage Feed Figure 7 - Temperature Profile Sensitivity to Changes in Steam Rate Having selected a candidate control scheme, Step 3 is to evaluate the open loop temperature sensitivity to the manipulative variable, and to find an appropriate temperature sensor location. Figure 7 shows the temperature sensitivity to changes in steam rate around the base case profile. Temperatures below the feed appear to be reasonably sensitive to steam rate changes. Additionally, positive changes are nearly equal in magnitude to negative changes. Normally, we would select a temperature location in this region based on the above characteristics. However, the presence of variable amounts of salts which are high boiling components make temperature sensing below the feed unreliable for inferring composition. Temperature no longer uniquely defines composition under such conditions. We are therefor left with the region above the feed. For this example, we selected stage 34. Notice that while there is sufficient sensitivity in this region of the column, the changes are not symmetrical about the base case profile. We have developed a technique to deal with unequal temperature sensitivity like this by redefining the base case so that the temperature at the location selected is equally spaced between the temperature extremes. While the new base case will have a slightly higher bottoms purity and steam flow rate, the result is a more linear temperature variation in the region of our temperature measurement. Case 1 Case 2 Case 3 Case 4 Case 5

13 Tray 34 Temp [C] Reflux [pph] Top Water [ppm] Bottom Methanol [ppm] Table 2 - Steady-State Cases Used To Establish New Base Case (Tray 34 Temp Centered Between Extreme Profiles) Table 2 illustrates the steady state cases run to select the new base case. Cases 1 to 3 are experiments in moving the temperature profile up the column by increasing the base-case steam flow. Notice in Case 3 that the top water specification is significantly exceeded. To correct this, Case 4 was defined with a lower temperature and higher reflux rate. In Case 5, the reflux rate was further increased to meet the distillate specification. Case 5 is the new base case where tray 34 is set at 75.5 degrees C and the reflux flow is set at 7700 pph. We decided to recheck the temperature sensitivity at the sensor location because we have changed the base case conditions. Figure 8 indicates that we do have significant temperature sensitivity, and that the variations in the positive and negative directions are nearly equal Temperature [C] % Control Location Base Case -5% Top Bottom Theoretical Stage Feed Figure 8 - Temperature Profile Sensitivity With New Base Case Having completed the open-loop testing to determine sensitivity, and identified an appropriate temperature sensor location, we proceed to Step 4: closed-loop testing. The closed loop performance is tested by fixing reflux flow at 7700 pph, forcing tray 34 to be 75.5 degrees C, and then subjecting the column model to the extremes of feed rate and feed composition. Table 3 shows that for the expected range of feed rate and composition disturbances, the selected control strategy does indeed maintain the top and bottom compositions below the specifications. Case 1 Case 2 Case 3 Case 4 Feed [pph] 4,200 26,000 4,200 26,000 Feed - % Methanol Top Water [ppm] Trace* 0.5 Trace* Trace* Bottoms Methanol Trace* 10 Trace* 18 [ppm] * Trace indicates compositions less than 1 ppm

14 Table 3 - "Closed Loop" Testing of Control Strategy This completes the design procedure for this column. A fixed reflux-ratio alternative scheme was also analyzed, however it was found to have significantly higher energy consumption.

15 CONCLUSIONS We have developed and successfully applied an effective and efficient control design procedure for distillation columns using steady state models. We have improved the technique originally proposed by Tolliver and McCune (1978). The steps of the design procedure are: Step 1: Develop design basis Step 2: Select candidate control scheme Step 3: "Open loop" test to find sensor location Step 4: "Closed loop" test candidate control scheme for known disturbances The procedure has been successfully applied to 33 industrial columns. We have found that for the vast majority of cases we can exceed our objectives with single point composition control where temperature is used to infer composition. We have shown that the most important manipulative variable, by far, for composition control is the feed split. We advocate the use of mass flows in the model versus molar flows, and that the proposed control strategy be enforced when performing "open loop" and "closed loop" testing. We have shown that for multicomponent cases, the steady state model allows us to quantify the incremental benefit of on-line analyzers versus temperature controls.

16 Figure 1: Schematic of a simple distillation column showing nomenclature and the five valves available for control. Figure 2: Numerical examples showing the effect of changing the feed split and fractionation on product compositions for distillation columns. Figure 3: A very common control strategy, when column is not a tar still and feed is demand stream. This strategy directly manipulates the feed split. Figure 4: Another very common control strategy, when feed is a demand stream and column has relatively low reflux ratio. This strategy indirectly manipulates the feed split. Figure 5: "Open loop" testing example. Steady state temperature profiles for three different distillate flows used to locate temperature sensor. Figure 6: Process schematic for example No. 1 (i.e., methanol, water separation). A split column configuration is used. Figure 7: "Open loop" testing results for example No. 1. Changes are made to steam flow with mass reflux flow fixed.

17 References Buckley, Luyben and Shunta, Distillation Column Control Design, Thurston C. W., "Computer-Aided Design of Distillation Column Controls", Hydrocarbon Processing, Part 1, July 1982, Part 2, August 1981, Page 5 Tolliver T. L. and McCune L. C., "Distillation Column Control Design Based On Steady State Simulation", ISA Transactions, Vol. 17, No. 3, 1978, Pages 3-10 Roat S. D., Moore C. F., and Downs J. J., "A Steady State Distillation Column Control System Sensitivity Analysis Technique", 1988, Proceedings IEEE Southeast Con, Pages Skogestad S., "Dynamics and Control of Distillation Columns - A Critical Survey", Preprints IFAC Symposium, DYCORD + 92, College Park, MD, USA, Pages 1-25 Luyben W. L., "Steady-State Energy Conservation Aspects of Distillation Column Control Design", I&E.C., Fundam., Vol. 14, No. 4, 1975, Pages Chien I-L and Fruehauf P. S., "Consider IMC Tuning to Improve Controller Performance", C.E.P., Vol. 86, No. 10, October 1990, Pages Shunta J. P. and Luyben W. L., "Dynamic Effects of Temperature Control Tray Location in Distillation Columns", AICHE J, Vol. 17, No. 1, January 1971, Pages Rademaker O., Rijnsdorp J.E. and Maarleveld A., Dynamics and Control of Continuous Distillation Columns, American Elsevier Publishing Company, Inc., New York, New York, 1975

18 Table 2: Steady state runs used to find new base case where Stage 34 temperature is centered between extremes Table 3: Check of temperature sensitivity of Stage 34 at new base case conditions. Table 4: "Closed loop" testing of candidate control strategy for example 1. Steady state results for different feed rates and compositions with reflux fixed at 7700 pph and Stage 34 temperature at 75.5 Degrees C. Table 1. Design Basis Product Composition Specifications - One or Two Sided? - Reason for? Economic Considerations - Product valve - Energy costs - Value of Incremental Production Increase (when sold out) - Waste treatment Costs Disturbances - Feed Rate: Min. and Max. - Feed composition Min and Max for each component Constraints - Demand Stream - Reason for currant control approach ( if retrofit) - Flooding limits, Reboiler and Condenser Capacities - Safety Limits - Upstream and downstream equipment: Recycle streams? Column Design Basis