Performance Assessment of Plant-Wide Control Systems

Document by: Bharadwaj Visit my website

www.engineeringpapers.blogspot.com
More papers and Presentations available on above site
Abstract: This paper addresses performance assessment of plant-wide control (PWC) systems, which is one of the important areas of PWC of industrial processes but has received very little attention in the past. Some of the performance measures in the literature are the steady-state disturbance sensitivity and the capacity-based economic approach. However, there are some limitations in the measures that have been proposed for performance assessment, from the plant-wide perspective. In this study, several measures based on plant dynamics are described and discussed for evaluating the performance of different PWC structures. These include overall process settling time based on absolute accumulation in the system, dynamic disturbance sensitivity, net variation from the nominal operating profit and deviation from the production target. These measures are then applied to two alternative control structures for a styrene monomer plant, in order to test their applicability. The results indicate that some of the presented measures are indeed effective for evaluating and comparing PWC structures. Keywords: Plant-wide control, Performance assessment, Settling time, Operating profit, Dynamic disturbance sensitivity, Deviation from the production target.

Introduction PWC design is essential because of complex, integrated nature of many chemical plants. The development of a PWC system does not stop with the generation of the control structures. The dynamic performance of the alternative control structures generated must be evaluated in order to make the final control system selection. Thus, performance assessment of PWC systems is important but has not received much attention in the past, especially in the context of dynamic simulations.

The presence of numerous combinations of controlled and manipulated variables in a process leads to many alternative control structures. In the preliminary steps of the control system development, steady-state and dynamic controllability measures such as relative gain array (RGA), condition number (CN), Neiderlinski Index (NI), singular value decomposition (SVD), etc. can be used for initial control structure screening. However, a few alternative control structures may still remain in the end, and more rigorous analysis is needed for the final selection to be made. Also, there is a possibility that measures like RGA and NI may fail, and result in poor closed loop performance, as has been proven in several case studies [1, 2]. These issues have led to the proposition of other performance measures that give more consistent results even for highly complex processes. Yi and Luyben [3] proposed a measure named steady-state disturbance sensitivity to screen alternative control structures. Control structures that require a large change in manipulated variables to handle the disturbances are not recommended, as they are more prone to hit constraints and valve saturation limits. This measure can give us desirable results under normal circumstances, but the decision would be difficult to make in cases where the manipulated variable changes more for some disturbances and less for some other disturbances in one control structure as opposed to the other control structures. Also, the steady-state disturbance sensitivity does not consider the changes during the transient phase. Elliott and Luyben [4, 5] and Elliott et al. [6] devised a generic methodology called the capacity-based economic approach to compare and screen preliminary plant designs by quantifying both steady-state economics and dynamic controllability. They basically proposed a measure based on product quality regulation to measure the dynamic performance of alternative control structures. Specifically, they calculate the loss in plant capacity due to off-spec production. As with the previous measure, this approach is useful under some situations. However, it cannot be always applied as the off-spec product is assumed to be disposed, whereas the normal practice in industry is to recycle the off-spec product in order to avoid yield losses and additional cost of disposal. Also, product quality though important cannot be the only measure for evaluating the control system performance. The dynamic performance of all control loops in the whole plant has to be considered too. Keeping the above measures and limitations in mind, a new dynamic performance index called dynamic disturbance sensitivity (DDS) has been recently developed by Konda and Rangaiah [7]. DDS is equal to the sum of absolute accumulation of each and every component in the process since the onset of the disturbance(s). Konda and Rangaiah [7] have applied DDS to three different PWC structures of the toluene hydrodealkylation

(HDA) process and proven its effectiveness for PWC performance assessment and comparison. DDS measure offers several advantages over the other methods discussed above. However, one major drawback of it and many other measures is that they do not include the economic quantification of the dynamic performance. Hence, a new economic measure based on deviation from the production target (DPT) of the main product during the transient period is proposed in this study. In addition, two more performance measures are discussed. These are the net variation in the plant operating profit and the process settling time based on overall absolute component accumulation. The basic idea behind these measures and their development are discussed, together with the procedure for their computation. These measures are then applied to two different control structures of the styrene monomer plant in order to test their applicability and usefulness. The rest of the paper is organized as follows: the next section discusses the various performance measures, namely, the process settling time based on overall absolute component accumulation, DDS, the net variation in the plant operating profit and DPT. The subsequent section discusses the application of these measures to the styrene plant together with an analysis of the results. The conclusions are finally presented in the last section. Performance Measures for PWC Systems A few performance measures based on plant dynamics are presented in this section. Not much has been said or discussed on what constitutes a good performance measure. A performance measure must be comprehensive (to include transients and steady-state performance), consistent and robust (i.e., less affected by controller fine-tuning), able to differentiate control structures, simple (to define and compute), and also give some indication of the economics and control effort of the control system. A good measure should meet most or all of these features. Process Settling Time Based on Overall Absolute Component Accumulation Settling time is defined as the time required for the process output to reach and remain within certain (5% or 1%) of the final steady-state value of the process variable [8]. While this definition pertains to a single control loop, the question in a plant-wide context is how to define the settling time of a highly integrated chemical process plant with many controlled variables and controllers. We propose to calculate settling time based on transient profile of absolute accumulation of all components in the plant, which is defined as follows:

Absolute

n accumulati on = Inflow Outflow + Generation 1

Consumptio

(1) where n is the total number of components in the system. The inflow and outflow refers to the component flows in the input and output streams of the plant at any time. The settling time thus calculated indicates the time taken by the PWC system to bring the overall process to the steady state after the onset of the disturbance(s). Dynamic Disturbance Sensitivity (DDS) DDS is a new dynamic performance index developed by Konda and Rangaiah [7]. The basic idea is to make effective use of rigorous process simulators in the performance assessment of PWC systems. DDS makes use of the strong correlation between the overall control system performance and the sum of the individual component accumulations. When a process is affected by a disturbance, all the process variables go though different transients and ultimately reach steady state only when the overall accumulation of all the components in the plant becomes negligible. The time integral of sum of absolute accumulation of all components in the plant is a measure of the PWC system performance. Accordingly, DDS is defined as
DDS =
ts

t =0

( absolute
i =1to n

accumulati on of component i ) dt

(2)

where ts is the time taken for the process to reach steady state. Obviously, a smaller value of DDS indicates better control. Net Variation in the Plant Operating Profit A profit function has been used in PWC methodologies for steady-state optimization. An example is the self-optimizing control procedure of Skogestad [9]. We propose to compute the operating profit during the transient state in the presence of disturbances. As it is possible that the operating profit may settle at a different value due to changes in the production rate, we propose using profit per unit production rate of the main product. This goes back to almost the same value at the final steady state conditions (see Figure 1). The net area gives an indication of the net variation in the plant operating profit. A positive number indicates an overall increase in the plant operating profit whereas a negative number indicates otherwise. A control system can be taken to be performing well if the computed area is either equal to or greater than zero.

Positive Area US$/kg Product

Steady-State Profit

Negative Area

Time

Fig. 1. Transient profile of profit per unit production rate due to a disturbance Deviation from the Production Target (DPT) There could be one possible drawback with the computation of the net variation in plant operating profit. It is possible to obtain a higher value of the computed area for an inferior control structure primarily due to the large magnitude of the initial production rate transient. This issue is discussed in detail later in the next section. Hence, to overcome this shortcoming, we propose an indirect economic measure based on the production rate. The idea is to compute the DPT (of the main product) during the transient state by computing the net area as shown in Figure 2. DPT is defined as
DPT =
ts

t =0

( actual

production rate t arg et production rate ) dt

(3)

When the plant management wants to increase (decrease) the production rate, the new production target should be achieved at the earliest and any deviation from it is undesirable. So, total deviation from the production rate during the transient can be used as a performance measure of the control system. This can be calculated by the area under the transient in Figure 2 for a production rate change. On the other hand, when the plant is subjected to other unexpected disturbances, the computation of area is done based on the original production rate. The effectiveness of this measure in indicating relative control system performance is illustrated in the next section.

Initial production rate

Production Rate

Over production Under Production Production rate target

Time
Fig. 2. Production rate transient in the presence of disturbance(s) Application of the Performance Measures to the Styrene Monomer Plant Process Description and Simulation In the styrene process (Figure 3), fresh ethyl benzene (EB) and a part of the low-pressure steam (LPS) are mixed and then pre-heated in a feed-effluent heat exchanger (FEHE) using the reactor effluent stream. The remaining LPS is superheated in a furnace to 700-850C, and then mixed with the pre-heated mixture to attain a temperature of around 650C. It is then fed to two adiabatic plug-flow reactors (PFRs), in series with a heater in between, for the production of styrene. The six main reactions that occur in the reactors are as follows:
C 6 H 5 CH 2 CH 3 C 6 H 5 CHCH
2

+H2

(4) (5) (6) (7) (8) (9)

C 6 H 5 CH 2 CH 3 C 6 H 6 + C 2 H 4 C 6 H 5 CH 2 CH 3+H 2 C 6 H 5 CH 3 + CH 4

2 H 2 O + C 2 H 4 2CO + 4 H 2 H 2 O + CH 4 CO + 3H 2 H 2 O + CO CO 2 + H 2

The reactor effluent is cooled in the FEHE and then in a cooler before being sent to the three-phase separator, where the light gases are removed as the light products and water is removed as the heavy product. The intermediate organic layer is sent to two distillation columns for styrene separation from the other components. In the first column (i.e., product column), operating under vacuum to prevent styrene polymerization, styrene is removed as the bottom product, and the top product is sent to a second column (i.e., recycle column) to

separate un-reacted EB from the by-products, toluene and benzene. The un-reacted EB is then recycled back. For simulating the styrene plant in HYSYS, Peng-Robinson equation of state is chosen as it is very reliable for predicting the properties of hydrocarbon components over a wide range of conditions and appropriate for the components in the styrene process. The steady-state model of the styrene process is developed and optimized in HYSYS. The optimal conversion is 66.6%. The distillation columns are modeled by rigorous tray-by-tray calculations. Preliminary estimates of the number of trays and feed tray location for each column in the process are obtained by using the shortcut column in HYSYS, and then optimized using the rigorous calculations. The final steady-state flowsheet developed is shown in Figure 3. Dynamic Simulation of the Selected Control Structures Four different control structures have been recently developed for the styrene monomer plant [10]. We consider two of the control structures here; these are summarized in Table 1. HS is the control structure developed using the heuristics procedure of Luyben et al. [11]. One of the characteristic features of this procedure is to control a flow somewhere in the recycle loop supposedly to overcome the snowball effect. IF is the control structure developed using the integrated framework of simulation and heuristics [12]. One of the main features of this procedure is the detailed analysis of the effects of recycle in order to overcome its negative impact on control system performance. There are several main differences between the two control structures in Table 1. Firstly, throughput manipulation is achieved by changing EB feed flow in IF, while it is achieved by changing total EB flow (feed plus recycle) in HS. Secondly, the recycle flow is explicitly fixed in HS; this is achieved by controlling the total feed flow. Thirdly, IF utilizes the conversion controller as a result of the detailed analysis of the effects of recycle. Finally, the number of control loops is different: HS has 20 feedback control loops, while IF has 21 control loops, 2 of which form a cascade loop. The two control structures are implemented in HYSYS dynamic mode; the controller tuning parameters are also given in Table 1.

Fig. 3. Steady-state HYSYS flowsheet of the styrene monomer process. Results and Discussion Both the control systems are analyzed for their relative performance for production rate and feed composition disturbances. The dynamic performance measures discussed in the previous section are employed to analyze their performance. Overall Process Settling Time Settling times based on overall absolute component accumulation are computed for both the control systems and presented in Table 2. The criterion for computing the settling time is that the overall absolute component accumulation should be below 1 kmol/h (= 0.84% of the styrene production rate of 120 kmol/h). The results in Table 2 show that IF performs better in terms of the settling times, for the disturbances studied. This means that this structure is able to attenuate the effect of disturbances quickly with smaller settling time for the disturbances considered. The overall absolute component accumulation is a good basis for computing settling time as it takes into account the entire plant performance during the transient state and not just individual process variables or control loops.

Table 1. HS and IF Control Structures for Styrene Plant

Controlled Variable (See Figure 3 for Abbreviations) Reaction Section Total EB Flow EB Feed Flow PFR1 Inlet Steam/EB Ratio LP1 Flow EB Conversion PFR1 Inlet Temperature PFR2 Inlet Temperature Phase Sep Temperature Phase Sep Pressure Phase Sep Liquid % Level Phase Sep Aqueous % Level Product Column Condenser Pressure Condenser Level Reboiler Level Top Styrene Composition Reflux Flow Bottoms EB Composition Vent Flow Recycle Column Condenser Pressure Condenser Level Reboiler Level Top EB Composition Bottoms Toluene Composition Manipulated Variable with Controller Parameters: Kc (%/%), Ti (min) in Brackets HS IF EB Feed Flow (0.5, 0.3) EB Feed Flow (0.5, 0.3) Steam Feed Flow (0.36, 0.035) LP1 Flow (0.5, 0.3) PFR1 Inlet T SP (0.1, 0.5) Furnace Duty (0.11, 0.088) Intermediate Heater Duty (0.54, 0.087) Cooling Water Flow (0.13, 0.14) Lights Flow (2, 10) Organic Flow (18.8, 0.45) Water Flow (1.31, 0.12) Condenser Duty (2, 10) Distillate Flow (0.8) Distillate Flow (2) Bottoms Flow (0.5) Bottoms Flow (1.5) Reflux Flow (0.5, 27.3) Reflux Flow (0.5, 0.3) Reboiler Duty (0.12, 108) Reboiler Duty (0.23, 54) Compressor Duty (0.5, 0.3) Condenser Duty (1.5, 30) Condenser Duty (2, 20) Reflux Flow (1.2) Reflux Flow (2) Bottoms Flow (2) Distillate Flow (0.43, 73.5) Reboiler Duty (6.54, 1.05)

Table 2. Process Settling Times for HS and IF Control Structures based on Overall Absolute Accumulation Profile
No. d1 d2 d3 d4 Disturbance Magnitude -5% +5% -20% -2% Settling Time (minutes) for Overall Absolute Accumulation HS IF 710 345 700 335 865 455 595 245

Production Rate Feed Composition

Dynamic Disturbance Sensitivity (DDS) Since the computation of DDS is also based on overall component accumulation as discussed earlier, it is imperative that the results based on both DDS and settling time based on accumulation be compared. Accordingly, DDS values are computed and presented in Table 3. Again, IF shows better performance than HS in terms of DDS. However, one major difference between settling time and DDS is the ability of the latter to track the transient behavior of the process variables. On the other hand, settling time ignores the magnitude of the absolute component accumulation during the transient state, and hence is not a comprehensive measure compared to DDS. Table 3. DDS Values for HS and IF Control Structures
No. d1 d2 d3 d4 Disturbance Production Rate Feed Composition Magnitude -5% +5% -20% -2% DDS (kmol) HS IF 43 19 44 18 177 74 21 11

Net Variation in the Plant Operating Profit The data for the profit function per unit production rate is collected and the net area indicated in Figure 1 is computed for both the control structures. This area that gives an indication of the net variation in the plant operating profit, is given in Table 4. For easier understanding, the net variation in profit per tonne of styrene is computed for the duration for which the simulation is run in all cases (2000 minutes, i.e., approximately 1.4 days) by the following equation:
$ Net Variation tonne styrene $. hr = Net Variation kg styrene 60 m in 1000 kg 1 hr 2000 min tonne

(10) These values are also given in Table 4. A positive value indicates profit increase and vice-versa, and so a profit increase indicates better performance. As shown in Table 4, HS shows better performance for d1 and d3, even though the other performance measures applied so far indicate otherwise. This can be attributed to the greater production of styrene during the transient state (as a result of larger fluctuations in the process variables) which is manifested in higher profit. Actually, when the objective of the plant operator is to decrease styrene production in the case of d1 and d3, a higher amount of styrene produced during the

10

transient state is undesirable. The reason for this is that any excess amount of styrene produced cannot be sold easily given the lower demand. Thus, there is a major drawback with this economic measure as it does not take into account the over-produced amount of styrene during the transient state. Hence, as discussed in the previous section, an alternative economic measure (DPT) is evaluated next. Table 4. Net Variation in the Plant Operating Profit with Units of (a) $/(kg of Styrene/hr) and (b) $/(tonne of Styrene), for HS and IF Control Structures
No. d1 d2 d3 d4 Disturbance Production Rate Feed Composition Magnitude -5% +5% -20% -2% Net Variation ($/kg/hr) HS IF 0.20 0.13 -0.21 -0.12 0.76 0.50 -0.013 0.00 Net Variation ($/tonne) HS IF 5.9 4.0 -6.2 -3.7 22.7 15.1 -0.38 0.07

Note: In this table, the largest value of the net variation for each disturbance is shown with grey background.

Deviation from the Production Target (DPT) of Styrene The DPT of styrene over the duration of the total simulation time (2000 min) is computed. The results presented in Table 5 give a clear picture. For d1 to d3, the target production rate in equation 3 is the final production rate at the new steady-state. For d4, the target production rate in equation 3 is the original production rate. A positive value of DPT indicates over production, while a negative value indicates under production. Since both over and under production is undesirable, smaller (absolute) deviation from the desired production rate target indicates better performance. In general, IF shows better performance for all disturbances except d4. This means that a smaller amount of styrene over/under production is achieved with this control system. This translates into a lower loss due to unwanted amount of product being produced. Interestingly, IF shows poorer performance for d4. This could be due to the fixing of EB feed flow in IF, whereas it is allowed to vary in HS. For d4, this results in a 2% decrease in the fresh EB flow in IF as opposed to 0.75% decrease in the case of HS. The greater decrease in the fresh EB flow is manifested as a larger DPT as the styrene production rate is directly proportional to the fresh EB flow rate. General Assessment of the Performance Measures The performance measures presented in the previous section have been evaluated on two different control structures for the styrene plant in this section. In general, control structure IF performs better in terms of most of the measures except for the net variation in the plant

11

operating profit, where a clear picture does not emerge from the results. However, as discussed earlier, this measure does not indicate the superior performance of a control structure. Of all the measures presented, we recommend DDS (as it gives an indication of the ability of the control system in handling disturbances dynamically) and DPT (as it gives an indication of the economic performance of the control system). Table 5. DPT of Styrene for HS and IF Control Structures
No. d1 d2 d3 d4 Disturbance Production Rate Feed Composition Magnitude -5% +5% -20% -2% DPT of Styrene (kg) HS IF 2329 1558 -2275 -1376 8844 5656 -2558 -5947

Note: In this table, the smallest DPT of styrene for each disturbance is shown with grey background.

A typical plant with a control system has many process and operating variables that have to be monitored, and it can be quite time-consuming and difficult to analyze and compare the numerous profiles of the different alternative control systems. Besides, such qualitative analysis is subjective. Hence, quantitative measures based on process dynamics such as presented here enable effective and easy analysis with minimal computational effort. All the calculations presented can be easily automated and done using a spreadsheet. Conclusions Several performance measures based on plant dynamics for assessing PWC systems have been presented in this work. The main aim of this is to present easier and more reliable quantitative tools for assessing different PWC structures. The feasibility of using these measures for performance assessment of PWC systems has been illustrated on two alternative control structures for the styrene plant. In particular, DDS and DPT are recommended in order to get an overall picture of PWC system performance covering various aspects. References 1. He, M.J. and Cai, W.J., New Criterion for Control-Loop Configuration of Multivariable Processes, Ind. Eng. Chem. Res., 43, pp. 7057-7064 (2004). 2. Xiong, Q., Cai, W.J. and He, M.J., A Practical Loop Pairing Criterion for Multivariable Processes, J. Proc. Cont., 15, pp. 741-747 (2005).

12

3. Yi, C.K. and Luyben, W.L., Evaluation of Plant-Wide Control Structures by SteadyState Disturbance Sensitivity Analysis, Ind. Eng. Chem. Res., 34, pp. 2393-2405 (1995). 4. Elliott, T.R. and Luyben, W.L., Capacity-Based Economic Approach for the Quantitative Assessment of Process Controllability during the Conceptual Design Stage, Ind. Eng. Chem. Res., 34, pp. 3907-3915 (1995). 5. Elliott, T.R. and Luyben, W.L., Quantitative Assessment of Controllability during the Design of a Ternary System with Two Recycle Streams, Ind. Eng. Chem. Res., 35, pp. 3470-3479 (1996). 6. Elliott, T.R., Luyben, W.L. and Luyben, M.L., Application of the Capacity-Based Economic Approach to an Industrial-Scale Process, Ind. Eng. Chem. Res., 36, pp. 17271737 (1997). 7. Konda, N.V.S.N.M. and Rangaiah, G.P., Performance Assessment of Plant-Wide Control Systems of Industrial Processes, Ind. Eng. Chem. Res., 46, pp. 1220-1231 (2007). 8. Seborg, E., Edgar, T.F. and Mellichamp, D.A., Process Dynamics and Control, Wiley, New Jersey (2004). 9. Skogestad, S., Control Structure Design for Complete Chemical Plants, Comput. Chem. Eng., 28, pp. 219-234 (2004). 10. Vasudevan, S., Rangaiah, G.P., Konda, N.V.S.N.M. and Tay, W.H., Application and Evaluation of Three Methodologies for Plant-Wide Control of the Styrene Monomer Plant, Ind. Eng. Chem. Res., In press (2009). 11. Luyben, W.L., Tyreus, B.D. and Luyben, M.L., Plant-Wide Process Control, McGrawHill, New York (1998). 12. Konda, N.V.S.N.M., Rangaiah, G.P. and Krishnaswamy, P.R., Plant-Wide Control of Industrial Processes: An Integrated Framework of Simulation and Heuristics, Ind. Eng. Chem. Res., 44, pp. 8300-8313 (2005).

13