Simple Regulatory Control of the Eastman Process - American

3280

Ind. Eng. Chem. Res. 1996, 35, 3280-3289

KINETICS, CATALYSIS, AND REACTION ENGINEERING Simple Regulatory Control of the Eastman Process William L. Luyben* Department of Chemical Engineering, Iacocca Hall, Lehigh University, Bethlehem, Pennsylvania 18015

A simple regulatory control structure is proposed for the Eastman process that provides more effective control of the production rate than more complex strategies previously proposed in the literature. Stripper bottoms is simply flow controlled, and level controls are arranged in a direction opposite to the process flows. Common heuristics are shown to provide an effective, stable control system. Drastic disturbances are easily handled by overrides. An important feature of the structure is the use of proportional-only controllers on all loops in this integrating system. Introduction Downs and Vogel (1993) provided an important service to the process control community by presenting a simulation program of a realistically complex reactor/ separation process. Several academic workers, those with interests in practical problems, have explored this process using a variety of approaches. A detailed description of the process in this paper is unnecessary since it has been provided by the originators and by other authors. We will only summarize some of the essential and unusual dynamic features. The reactor, which is open-loop unstable, contains both liquid and vapor phases, but there is no liquid stream leaving the reactor. Both the gas pressure and the liquid level in the reactor are integrating phenomena, and the choice of what manipulated variables to use to control them is somewhat clouded. The temperature, pressure, and liquid level in the reactor are all interacting and nonlinear. The gas purge stream from the process is very small, so its effectiveness in controlling pressure is doubtful (although some authors propose its use for this purpose). There are four fresh feed makeup streams, which must be managed in an appropriate way to satisfy the overall component balances. Fortunately, composition analyzers are available. Figure 1 gives a sketch of the process and the values of the flow rates, compositions, temperatures, and pressure at the initial steady state (mode 1). The nomenclature used in this paper is also shown. The purpose of this paper is to demonstrate the application of a general plantwide control procedure recently developed by Luyben et al. (1996) to the Eastman process. The procedure first considers where to set the production rate. Then, product quality, safety, and environmental issues and constraints are addressed, and manipulated variables are selected that permit tight control (minimum variability) of these important controlled variables. Next, manipulated variables are selected to control all inventories (liquid levels and gas pressures) in the process. Then, manipulated variables are chosen to satisfy the component balances. The remaining manipulated variables can be used either to improve the dynamics of the process or * E-mail: [email protected].

S0888-5885(96)00162-5 CCC: $12.00

to achieve optimal economic operation of the process. We show that the application of this method yields a simple but effective control structure for the Eastman process. Previous Work There have been several papers published that study the Eastman process. We summarize below the unique features of these and try to contrast the basic strategies. All structures control the reactor temperature with cooling water to the reactor, but some structures change the setpoint of this reactor temperature controller in a cascade configuration. All structures use the fresh feed FoA to control the composition of component A in the system. The structures differ greatly in how the production rate is set and how the liquid level and pressure in the reactor are held. All of the published structures control the liquid levels in the separator and stripper by manipulating the liquid streams leaving the vessels. Buckley (1973) called this liquid level control in the direction of flow. Note that the diagrams of the control structures proposed by other authors have been simplified by not showing all the flow cascade loops. 1. LG Structure. Lyman and Georgakis (1995) presented four control structures but recommended the scheme shown in Figure 2. The two unique features are (a) production rate is set by condenser cooling and (b) reactor pressure is not controlled. The reactor level is controlled by fresh feed FoC. The compositions of components A, D, and E in the reactor feed are controlled by fresh feed rates FoA, FoD, and FoE. The composition of inert component B in the system is controlled by manipulating the purge rate. The concentration of component E in the product stream B from the bottom of the stripper is controlled by manipulating reboiler steam through a composition/temperature cascade loop. 2. MY Structure. McAvoy and co-workers have published two papers (McAvoy and Ye, 1994; McAvoy et al., 1996). Figure 3 gives their most recent recommendation. The two unique features of this scheme are (a) pressure is controlled by changing the setpoint of the reactor temperature controller and (b) production rate is set by manipulating the fresh feed FoC. The © 1996 American Chemical Society

Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996 3281

Figure 1. Eastman process steady-state parameter values.

Figure 2. Lyman-Georgakis control structure.

reactor liquid level is controlled by manipulating the fresh feed FoE with fresh feed FoD ratioed to it. This ratio sets the desired product mix between components G and H. The recycle flow rate is controlled by changing the setpoint of a temperature controller holding the temperature of the cooling water leaving the condenser by manipulating the cooling water flow rate. 3. R Structure. The most extensive studies of the

Eastman process have been reported by Ricker and coworkers in a series of papers. Ricker (1993) studied model predictive control. Ricker and Lee (1995a,b) explored modeling, state estimation, and nonlinear model predictive control. In a recent most insightful paper, Ricker (1996) developed a decentralized control structure, which he demonstrated is superior to the much more complex model predictive controllers he

3282 Ind. Eng. Chem. Res., Vol. 35, No. 10, 1996

Figure 3. McAvoy-Ye control structure.

Figure 4. Ricker control structure.

presented in previous papers. He did an excellent job in presenting a lucid discussion of the issues in developing a control structure: determining the degrees of freedom, selecting variables that must be controlled, selecting how to set the production rate, and deciding what to do with the remaining degrees of freedom. Figure 4 shows his recommended scheme. The three

unique features are (a) pressure is controlled by the very small purge flow rate, (b) a production rate controller changes the flow rates of all fresh feeds and liquid streams through ratio controllers, and (c) the reactor liquid level is controlled by changing the setpoint of a separator temperature controller that manipulates the cooling water to the condenser.


Proposed Basic Regulatory Control Structure The plantwide control procedure outlined by Luyben et al. (1996) consists of five parts: (1) establish how the production rate is set, (2) select the best manipulated variables to hold the product quality and constraints, (3) select from the remaining manipulated variables those necessary to control all inventories (liquid levels and gas pressures), (4) check to make sure that the overall component balances can be satisfied for all components (particularly inerts and reactants that are not completely converted in one pass through the reactor), and (5) use the remaining manipulated variables to either optimize some steady-state economic objective or improve dynamic controllability (flexibility, resiliency, load rejection, etc). The application of these steps to the Eastman process is given below. As Ricker (1996) nicely outlines, this process has 12 degrees of freedom. One of these is agitation rate, which we simply hold constant. A second is the valve that controls the flow rate of the gas recycle. We fix this valve also, based on the Douglas heuristic (Fisher et al., 1988) that gas recycle flows should be maximized to improve yields. This leaves 10 degrees of freedom (10 control valves to be set). 1. Set Production Rate. At least 1 degree of freedom must be used to set the production rate. The problem statement specifies that the flow rate of the product stream leaving the stripper should be held as constant as possible because it is fed into downstream processes. Therefore, we set the position of the control valve, XMV(8), on this stream (B). No other authors propose this obvious approach. The rest of the liquid level controls must now be chosen to accommodate this first-priority choice. Note that we could put a flow controller on this stream if necessary, but this was not done in the simulations described later. 2. Control Product Quality and Constraints. There is only one product, and the problem statement says that the only quality specification is that the mole % of component G should not vary more than (5%. In most processes, there would be a specification on the amount of component E that can be allowed to drop out of the bottom of the stripper. However, in the problem statement, Downs and Vogel make no mention of controlling the impurity of E in the product stream. Two manipulated variables have a direct effect on stripper bottoms purity (xB,G): the flow rate of steam to the reboiler (FS) and the flow rate of fresh feed (FoC). The former is somewhat smaller than the latter, so one might be inclined to select FoC to control xB,G, using the heuristic that the largest stream gives the best control. However, this fresh feed makeup stream affects the component balances of A and C in the system, while steam does not. Therefore, we select reboiler steam to control product purity. The stripper temperature is used to infer product composition and does a good job in keeping most of the light components from being lost in the product. There are only small changes in xB,E ((0.5%). The open-loop instability of the reactor acts somewhat like a constraint in the sense that closed-loop control of the reactor temperature is required. The logical manipulated variable to achieve this reactor temperature control is reactor cooling water (CWR). However, it is not necessary to hold a constant value of the reactor temperature, so the setpoint of the reactor temperature controller could be considered as a degree of freedom instead of the reactor cooling water flow rate.

The only constraint for this process is pressure, which must not exceed the shut-down limit of 3000 kPa. All the gas fresh feed streams, all the cooling water streams, the reboiler steam, and the purge all directly affect pressure, and potentially any of these could be used. Of course, the reaction rate also affects pressure, so anything that changes the reaction rate could potentially be used to indirectly control the pressure. The reactor cooling water flow rate and steam have already been selected for other uses. The condenser cooling rate is smaller than the reactor cooling rate, so it would not be very effective in controlling the pressure. This leaves us with using one of the gas flows. The purge stream is only 15.1 kg-mol/h, while the largest fresh feed makeup stream, FoC, is 417 kg-mol/h. The vapor holdup in the reactor, separator, and stripper is estimated to be about 15 m3. This gives a time constant of about 2 min if FoC is used to control the pressure. If the purge flow is used, the time constant is about 60 min. Thus, the fresh feed FoC is the logical choice to control the pressure. 3. Control of Inventories. There are three liquid levels in the system, so 3 degrees of freedom must be used to control these inventories. We use the Buckley strategy of level control in the reverse direction to flow. The product flow rate B is already fixed by the production rate. Therefore, flow from the separator (L) is chosen to control the level in the stripper. To control the level in the separator, it is logical to select the cooling water flow rate to the condenser (CWC). Now we are left with deciding how to control the liquid level in the reactor. This liquid consists of mostly the heavy products, components G and H. The more fresh feed of components D and E fed into the process, the more products will be produced. So we select the two fresh feed flow rates FoD and FoE to control the reactor liquid level. We ratio one to the other depending on the desired split between components G and H in the final product. Simple flow ratios should be accurate enough to maintain the desired product distribution without any feedback of product compositions. So online analyzers on the product streams should not be required. 4. Overall Component Balances. A light inert component B enters in one of the feed streams, and the only place it can leave the process is in the purge stream. Therefore, the purge flow rate is used to control the composition yB in the purge gas stream. The other component that must be considered is component A. There must be some feedback mechanism to guarantee that precisely the correct numbers of molecules of this component are fed into the system to react with the numbers of molecules of component C. The component balance for C is satisfied by bringing it in on pressure control. The only stream available to satisfy the component balance for A is the fresh feed stream FoA. So we select this flow to control the composition of A in the purge gas stream, yA. Note that the compositions of either the purge gas or the reactor feed could be used, but both are not necessary. Only two analyzers are required to run this process: one measuring the amount of A in the system and one measuring the amount of B. The latter could be eliminated if the amount of inert B coming into the system does not vary drastically. The purge stream is very small, and variations in the concentration of B in the system should have only a minor effect on controllability. However, Ricker (1996) has shown that the


Figure 5. Proposed control structure.

Figure 6. Structure with overrides.

purge does have a significant economic impact. All of the other published control structures depend on three or more on-line analyzers being operable. Some other structures can operate for a limited time period without analyzers, but not for extended periods of time. 5. Remaining Control Loops. Of the original 10 degrees of freedom, we have used up 1 for production rate, 1 for product quality, 1 for the pressure constraint, 3 for liquid levels, and 2 for component balances. An additional one was used to set the G/H ratio. This

leaves just 1 degree of freedom to be specified, which is the setpoint of the reactor temperature controller. The best way to manage the reactor temperature setpoint is not immediately obvious. It might be used in conjunction with the production rate controller; i.e., higher temperatures may be needed to increase the throughputs. It might be adjusted to maximize yields and surpress undesired byproducts. However, after making some simulation runs with several of the disturbances suggested by Downs and


Figure 7. Fifteen percent decrease in production rate.

Vogel, it became apparent that the temperature in the separator was changing quite substantially and adversely affecting the stripper. A low separator temperature drops too many light component into the stripper, and the reboiler steam has trouble maintaining product quality. Therefore, a separator temperature controller was added, whose output signal was the setpoint of the reactor temperature controller. The final basic regulatory control structure is shown in Figure 5. It is simple, effective, and easily understood by operating personnel. Override Controls The basic regulatory control structure outlined above was able to hold the process at the desired operating point for most of the disturbances. However, when manipulated variables hit constraints, it was unable to prevent a unit shutdown. Disturbance IDV(6) is probably the most drastic: the fresh feed flow rate FoA is shut off. The resulting imbalance in the stoichiometric amounts of components A and C drives the concentration yA down quite rapidly. The reaction rate slows up, reactor temperature drops, and the process shuts down on high pressure. Since 1 degree of freedom has been removed by this disturbance, the control structure must be modified to handle the component balances. The FoC stream contains more C than A, so the excess C must be removed from the system. The only place available is the purge stream. Therefore, a low FoA flow override controller is used to open the purge valve. See Figure 6. The other action that must be taken is to prevent the concentration of component A in the system from dropping down too low and reducing the reaction rates. This is achieved by using a low yA concentration override controller to pinch the fresh feed flow rate FoD to slow the rate of consumption of A. Of course, FoE is also reduced through the ratio. Now the liquid level loops must also be modified since we no longer can specify the production rate and the

reactor level control cannot use FoD. This is easily accomplished by using low-level override controllers on each of the three levels. A low stripper level pinches the product flow rate B. A low separator level pinches the separator liquid flow rate L. A low reactor level pinches the condenser cooling water flow rate CWC. In an override situation, the level control structure has been reversed from the basic structure, and now levels are held in the direction of flow. Simulation Results Simulations were made with the proposed control structure for all disturbances proposed by Downs and Vogel. Only the mode 1 operation was studied. Figures 7-12 gives the results for several disturbances. Figure 7 shows how a 15% reduction in production rate is handled. Product stream B immediately changes, and the rest of the process adjusts feed flows, lining out in about 2 h. The Lyman and Georgakis structure takes 2 h to adjust the production rate. Ricker’s control structure takes 15 h. If the product stream is cut off completely, this control structure successfully and automatically shuts down the process (all feeds gradually go to zero, and the process just sits there). Figure 8 shows how very drastic changes in the production rate can be handled quickly. At time equal 1 h, the valve position on the production rate (B) is dropped 50%. At time equal 10 h, it is increased back to its base case value. The process follows these changes in B by gradually and smoothly reducing fresh feeds through the level controllers. Figure 9 gives the results for changing the G/H split in the product. The two fresh feed streams FoD and FoE are immediately changed to the appropriate new values. Note that the cascade flow controllers used are not shown in Figure 5 for simplicity. The reactor level controller output signal is sent to a flow controller on


Figure 8. Drastic changes in production rate.

Figure 9. Change in G/H split.

FoD, and the bias value on this flow controller is changed to the desired value. At the same time, the ratio between the two flow rates is set to the new desired number. Product compositions xB,G and xB,H change to the new levels in about 4 h. The Lyman and Georgakis structure takes 10 h to accomplish this transition. Ricker’s control structure takes 7 h.

Figure 10 gives the response to a pressure setpoint change from 2705 to 2645 kPa. Figure 11 shows how the process rides through the loss of FoA disturbance, IMV(6). The override controller takes action almost immediately, and the production rate is reduced after about 5 h when liquid levels drop. Figure 12 shows the responses to IMV(1), a change


Figure 10. Change in pressure setpoint.

Figure 11. Loss of FoAsIDV(6).

in the composition of A and C in the FoC stream. As the amount of A in the system drops, the override controller cuts feed streams FoD and FoE. When the reactor holdup drops, the override controller cuts condenser cooling. When the stripper level drops, the override controller cuts the stripper liquid L. Finally,

after about 25 h, the low level in the stripper cuts back slightly on production rate B. Control Criteria In fairness to other authors of alternative control structures, it is important to point out that different


Figure 12. Change in C/A ratio in feedsIDV(1).

workers have interpreted the control objectives outlined by Downs and Vogel (1993) in different ways. McAvoy et al. (1996) attempt to reduce the variability in the feed streams to the process so as to not upset upstream units. Ricker (1996) attempts to satisfy economic objectives. In this paper, the control criterion is rapid changes in the production rate of the product stream from the bottom of the stripper. This diversity of structures is a very nice example of one of the basic process control principles that says that the “best” control structure depends on the control objectives. You cannot solve the problem unless you define what the problem is. The only universal criterion for any structure is that it must yield a closed-loop system that is stable. Controller Tuning A final word needs to be said about controller type and controller tuning. Although process design and control structure rank ahead of controller algorithm selection and tuning, these last items are important to the success of a control system. Two features should be recognized about the Eastman process. First, it is an integrating process in terms of pressure, liquid levels, and chemical components with little self-regulation. Second, there are no tight specifications on any variables. The integrating nature of this process makes it difficult to tune controllers with integral action. Two integrators in series present a challenging control problem because 180° of the phase angle is lost. The absense of tight specifications implies that steady-state offset or error is not a problem. Thus, both of these features lead us to use simple proportional-only controllers on all loops. Both the basic regulatory controllers and the override controllers were P controllers.

Table 1. Controller Tuning Constants basic loops levels pressure temperature compositions

reactor separator stripper reactor reactor separator stripper yB yA

Kc

transmitter span

2 4 2 100 3 0.15 2 16 10

100% 100% 100% 3000 kPa 100 °C 100 °C 100 °C 100 mol % 100 mol %

override controllers levels composition flow

reactor separator stripper yA FoA

Kc

transmitter span

1 2 2 1 100

100% 100% 100% 100 mol % 100%

Table 1 give values for the controller gains used and the transmitter spans used. All the valves are spanned in the program provided by Downs and Vogel between 0 and 100%. The level controller gains ranged from 1 to 4 and required little tuning. The only loops that required a little empirical tuning were the three temperatures, the pressure, and the two compositions. The reactor temperature was tuned first, followed by pressure, separator temperature, stripper temperature, component A composition, and component B composition. No claim is made that these are the best settings, but they give adequate control and required little time to tune. Conclusions A plantwide control design procedure was used to develop a simple but effective regulatory control system for the Eastman process. Control of the production rate is essentially instantaneous. Drastic upsets and dis-


turbances are handled by simple proportional-only overrides. Nomenclature B ) flow rate of stripper bottoms CWC ) flow rate of cooling water to condenser CWR ) flow rate of cooling water to reactor Foj ) flow rate of fresh feed, j ) A, C, D, E FS ) flow rate of steam to stripper HR ) reactor holdup, percent level Hsep ) separator holdup, percent level Hstrip ) stripper holdup, percent level L ) flow rate of liquid from separator to stripper P ) pressure Purge ) purge gas flow rate Recycle ) flow rate of gas recycle to reactor TR ) reactor temperature Tsep ) separator temperature Tstrip ) stripper temperature xBj ) composition of stripper bottoms, mole fraction component j yj ) composition of purge gas, mole fraction component j

Literature Cited Buckley, P. S. Technique of Process Control; Wiley: New York, 1964. Downs, J. J.; Vogel, E. F. A Plant-Wide Industrial Process Control Problem. Comp. Chem. Eng. 1993, 17 (3), 245-255.

Fisher, W. R.; Doherty, M. F.; Douglas, J. M. Ind. Eng. Chem. Res. 1988, 27, 611. Luyben, M. L.; Tyreus, B. D.; Luyben, W. L. Plantwide Control of the Vinyl Acetate Process. Paper submitted for presentation at the Chicago AIChE Meeting, Nov 1996. Lyman, P. R.; Georgakis, C. Plantwide Control of the Tennessee Eastman Problem. Comp. Chem. Eng. 1995, 19 (3), 321-331. McAvoy, T. J.; Ye, N. Base Control for the Tennessee Eastman Problem. Comp. Chem. Eng. 1994, 18, 383-413. McAvoy, T. J.; Ye, N.; Gang, C. Nonlinear Inferential Parallel Cascade Control. Ind. Eng. Chem. Res. 1996, 35, 130-137. Ricker, N. L. Model Predictive Control of a Continuous, Nonlinear, Two-Phase Reactor. J. Proc. Cont. 1993, 3 (2), 109-123. Ricker, N. L. Decentralized Control of the Tennessee Eastman Challenge Process. J. Proc. Cont. 1966, 6 (4), 205-221. Ricker, N. L.; Lee, J. H. Nonlinear Modeling and State Estimation for the Tennessee Eastman Challenge Process. Comp. Chem. Eng. 1995a, 19 (9), 983-1005. Ricker, N. L.; Lee, J. H. Nonlinear Model Predictive Control of the Tennessee Eastman Challenge Process. Comp. Chem. Eng. 1995b, 19 (9), 961-981.

Received for review March 22, 1996 Revised manuscript received April 26, 1996 Accepted May 6, 1996X IE960162X

X Abstract published in Advance ACS Abstracts, June 15, 1996.

Simple Regulatory Control of the Eastman Process - American

Recommend Documents