Вы находитесь на странице: 1из 18

I NTERNATIONAL J OURNAL OF C HEMICAL R EACTOR E NGINEERING

Volume 5 2007 Article A42

Nonlinear State Estimation and Generic Model Control of a Continuous Stirred Tank Reactor
Amiya K. Jana

Indian Institute of Technology - Kharagpur, amiya jana@yahoo.co.in ISSN 1542-6580

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Nonlinear State Estimation and Generic Model Control of a Continuous Stirred Tank Reactor
Amiya K. Jana

Abstract
The goal of this paper is to develop a nonlinear observer-based control strategy for a jacketed continuous stirred tank reactor (CSTR). The nonlinear adaptive state estimator/observer (ASE/ASO) is designed based on a model structure that mainly consists of an energy balance equation. In this observation approach, reactor concentration is considered as an imprecisely known extra state (augmented state) with no dynamics. Despite signicant process/model discrepancy, the proposed state observer estimated adequately the states of the simulated reactor. Mainly due to the design simplicity, negligible computational effort and fast convergence, the observer is recommended for online implementation. The generic model controller (GMC) has also been synthesized for the example reactor. The nonlinear GMC scheme receives the required information about the reactor concentration from the ASE for calculating the controller responses. Simulation experiments have been carried out to investigate the superior performance provided by the proposed GMC-ASE algorithm compared to the conventional proportional integral (PI) controller. KEYWORDS: adaptive state estimator, nonlinear control, GMC-ASE, PI, CSTR

Tel.: +91-03222-283918; Fax: +91-03222-255303; E-mail: amiya jana@yahoo.co.in.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Jana: Nonlinear State Estimation and Control of CSTR

1. INTRODUCTION
Severe nonlinearities of chemical processes influence the selection of control schemes for efficient control of a process. In recent years, a number of nonlinear control strategies have been proposed. Among them, most popular are the differential geometric based globally linearizing control (GLC) (Isidori, 1989; Kravaris and Chung, 1987; Jana et al., 2005a), nonlinear model predictive control (NLMPC) (Rawlings et al., 1994; Henson, 1998; Magni et al., 2001), and generic model control (GMC) (Lee and Sullivan, 1988; Guo et al., 2001). A drawback of GLC is that an exact knowledge of the system parameters is required (Henson and Seborg, 1991; Gonzalez-Trejo et al., 1999). This information is needed because the GLC control law is based on nonlinear coordinate transformations and cancellation of nonlinearities. If there is any model/plant mismatch, feedback control design is based on nominal coordinates transformations so that nonlinearities are no longer canceled. In such case, closed-loop stability can not be guaranteed anymore. Even if closed-loop instability problem does not exist, the GLC controller performance is degraded drastically. Again for NLMPC, it is required to solve a nonlinear programming problem online at each sampling period, which often makes NLMPC computationally expensive and numerically complex as the nonlinear program may converge to a local minimum or even sometimes diverge. The generic model control is a model-based control and it is closely related with the internal model control (IMC) (Lee and Sullivan, 1988). It is true that the performance of the GMC scheme is also degraded due to the inaccurate feedback of the state information to the controller. The combination of GMC law and an efficient state estimator may improve the closed-loop performance. The state observation scheme is coupled with the model-based controller, including the GLC and GMC, since all the required states are not practically measurable for most of the chemical processes. The state observers can be broadly classified into two categories: open-loop observer and closed-loop observer. The open-loop observers, in general, do not have any corrector part to reduce the process/predictor mismatch and are not updated by the true process outputs. For open-loop unstable processes (where the error of an open-loop observer may grow without bound) or open-loop stable processes with slow dynamics (where a significant initial state error may decay quite slowly) or open-loop stable processes under uncertainty, the open-loop observers fail to estimate the states precisely and consequently, the corresponding observerbased controller performance has been degraded drastically. Accordingly, the research attention has been paid to develop the efficient closed-loop observers those have both predictor and corrector parts. So far, different nonlinear closed-loop state estimators are reported in scientific literature (Kreisselmeier, 1977; Krener and Isidori, 1983; Krener and Respondek, 1985; Zeitz, 1987; Kantor, 1989; Chen, 1990; Kim and Choi, 1991; Deza et al., 1992; Ciccarella et al., 1993; Gibon-Fargeot et al., 1994; Kazantzis and Kravaris, 1995; Bullinger and Allgower, 1997; Farza et al., 1998; 1999; Dochain, 2003). The main purpose of this work is to develop a high-quality and readily implementable nonlinear control scheme for a variety of chemical processes. It is recognized (Hua and Jutan, 2000) that among the available nonlinear control methodologies the GMC is relatively simple to apply, and it allows us to develop a control algorithm by updating parameters in the process model. In general, the degree of accuracy in the estimation of the unmeasured states (including augmented states) restricts the success of the nonlinear controller over a broad range of conditions (Clarke-Pringle and MacGregor, 1997). Farza et al. (1999) proposed a simple nonlinear observer for the online estimation of the reaction rates in chemical and biochemical reactors. A main characteristic of this estimator lies in the easiness of its implementation and in their calibration method. It also offers transparent tuning and exponential convergence of the estimation error towards zero. This nonlinear adaptive state estimator/observer (ASE/ASO) is extended in the present study and is employed for state estimation of a continuous stirred tank reactor (CSTR). An interesting part of the present study is that the designed observation scheme estimates a true state variable (reactor concentration) accurately without having the proper information on concentration dynamics. Only a limited number of papers have appeared in literature dealing with the observer proposed by Farza et al. (1998). Moreover, in all those papers, the estimation of the partially known parameter(s) along with the measured state(s) has been performed. On the other hand, in the present work, the imprecisely known true state (reactor concentration) has been estimated along with the measured state (reactor temperature). This article is arranged as follows. First, the preliminaries on the generic model controller and the adaptive state estimator are outlined. Then the reactor model is presented. Most interesting feature of this CSTR is its multiple steady states. Among the two state variables, reactor temperature is assumed as measured state and the concentration of reactant in the product stream is considered as an imprecisely known extra state. Based on the energy balance

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

International Journal of Chemical Reactor Engineering

Vol. 5 [2007], Article A42

equation and extra state equation (with no dynamics), the ASE has been designed. The open-loop performance of the developed observation scheme has been checked with initialization error in reactor concentration, and step changes in heat of reaction and overall heat transfer coefficient. The effect of uncertain activation energy is also studied. Subsequently, the hybrid GMC-ASE control scheme has been derived for the example CSTR. The comparative set point tracking and disturbance rejection performances have been studied between the proposed GMC-ASE and a traditional proportional integral (PI) controller. Although the GMC controller and the ASE estimator are not new, the novelty lies in the development of a hybrid GMC-ASE combined structure, which is rarely reported in open literature.

2. GENERIC MODEL CONTROLLER (GMC)


In order to minimize the process/model discrepancy, it is required to feedback the process information as accurate as possible to the nonlinear controller model. Therefore, the time-varying parameters ( ) are needed to update along with the true states using an observation approach. In the state-space model, the parameter should appear with separate identity and its dynamics is provided under the estimation scheme. Accordingly, the following form can describe a nonlinear system:
& x = f ( x , d ) + g 1 ( u , x , d ), y = cx ,

(1)

where the state x n , the model parameter n , the measurable disturbance d q , the input u m . Moreover, f and g 1 are matrices of nonlinear functions, and c (coefficient matrix) is a unity matrix. From the basic principle of GMC (Lee and Sullivan, 1988), the following control law can be derived as: , f ( x , d ) + g (u , x , d ) e e dt = 0
1

(2)

where e is the error ( = y sp y ) to the controller, y sp is the set point of the output

and K 2 are diagonal n n tuning parameter matrices. Equation (2) shows that the GMC algorithm comprises of dynamic process model, proportional action term and integral action term.
K
1

y,

The values of the elements of tuning parameter matrices can be determined based on the following relationships proposed by Signal and Lee (1992) as:
K K
1 (i, i)

= =

2 1i 2i , 1 2i
2

(3)

2 (i, i)

where 1i and 2 i determine the shape and speed of the desired closed-loop trajectory (the reference trajectory), respectively. The reference trajectory gives pseudo-second order response for a step change in the set point. However, Yamuna and Gangiah (1991) confirmed that the above relationships could be applied to compute the specified response accurately. Once the values of 1i and 2 i are obtained, then K 1 and K 2 can be calculated using Equation (3). This nonlinear control strategy has several advantages, such as straightforward formulation and simple tuning. Importantly, the relationship between feedforward and feedback control is explicitly accounted for in the GMC law.

3. ADAPTIVE STATE ESTIMATOR (ASE)


It is assumed that the parameter dynamics in the nonlinear system (Equation (1)) obey the following first-order equation:
& = g 2 ( u , x , d ) + ,

(4)

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Jana: Nonlinear State Estimation and Control of CSTR

where

g2

is a nonlinear function and is an unknown function that may depend on x, , u , d , noise, and so on. The

assumptions that have been made are: is an unknown but bounded function and the disturbance d with its time derivative are also bounded. The nonlinear system Equations ((1) and (4)) can be expressed in the following condensed form:
& Z = F ( x , d ) Z + G (u , x , d ) + , y = C Z

(5)

x 0 where Z = , F ( x , d ) =

In

0 g 1 (u , x , d ) f ( x, d ) , = and C = [I n , 0], with , G (u , x , d ) = g 2 (u , x , d ) 0 0 the n n identity matrix. f is an n n matrix which is differentiable and the corresponding partial derivative

is continuous. According to Farza et al. (1998), the nonlinear adaptive observer can be used to track the vector Z as follows:
& Z = F ( y , d ) Z + G (u , y , d ) -
1

( y, d ) S

(C Z - y )

(6)

where
y n (i) Z = 2 n , are the estimated vectors of state and parameter, respectively.

(ii) ( y , d ) = I n 0

0 . f ( y, d )

(iii) S is the unique symmetric positive-definite matrix which satisfies the algebraic Lyapunov equation. Now, the gain of the estimator is:

2 I n ( y , d ) S 1 C T = 2 f 1 ( y , d )

(7)

where, > 0 is a design parameter (Gauthier et al., 1992). It is obvious from Equation (7) that only a single tuning parameter is involved in the estimator. When = 0 , the convergence of the observer error is an exponential one. In the case where 0 , the asymptotic error can be made arbitrarily small by choosing a sufficiently large value of . However, a very large value of may make the observer sensitive to noise. Thus, the choice of (Farza et al., 1999) is a compromise between fast convergence and sensitivity to noise. This adaptive observation scheme (Farza et al., 1998) offers several advantages, such as ease of construction and online implementation, negligible computational load, transparent tuning and more importantly, fast convergence. The GMC-ASE structure that will be designed for a CSTR is consisted with the nonlinear generic model controller and the adaptive state estimator. Indeed, the poorly known extra state along with the measured state is estimated in the nonlinear closed-loop observation scheme. The measured state is also estimated for calculating the residual in the corrector part. But note that the estimated extra state and measured true state will be used to compute the GMC responses.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

International Journal of Chemical Reactor Engineering

Vol. 5 [2007], Article A42

4. SIMULATION EXPERIMENTS
To study the performance of the adaptive observation scheme and the GMC as well as PI control law, a jacketed continuous stirred tank reactor (Jana et al., 2005b) is taken as an example process. An irreversible, exothermic, firstorder chemical reaction (AB) occurs in the perfectly mixed CSTR. The schematic representation of the example reactor is shown in Figure 1. The design parameters are reported in Table 1. Most interesting feature of this reactor that creates control difficulty is its multiple steady states. Figure 2 illustrates this feature clearly. As given in Table 2, there are three steady state operating points: SS1, SS2 and SS3. The steady state operating points SS1 and SS3 are stable, whereas the steady state operating point SS2 is not stable. It should be better to operate the reactor at high temperature steady state operating point (SS3) for high-conversion. But the high temperature steady state may be very high, causing unsafe conditions, destroying the catalyst, degrading the product, and so on (Stephanopoulos, 1998). Therefore, sometimes the reactor is operated at unstable steady state operating point (SS2) with medium-conversion.
F
C Af

Fj T CA
T
j

F Fj T
jin

V T
j

CA
T

Figure 1. Schematic representation of a jacketed CSTR 8.5615 10.0 1.0 500.0 150.0 38.2 25.0 25.0 11843.0 5960.0 0.005 34930800.0 1.987 kmol/m3 kmol/m3 hr-1 kcal/m3 0C kcal/m3 0C hr 0 C 0 C
0

CA C Af F /V C p U 0A / V T T
f

Tj E
t

( H )
K0 R

kcal/kmol kcal/kmol hr hr-1 kcal/kmol K

Table 1. Steady-state design parameters of the CSTR

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Jana: Nonlinear State Estimation and Control of CSTR

108

99

SS3

90

81

72 T ( C)
0

63

SS2

54

45

36

SS1

27

12

15

18

21

24
0

27

30

33

36

39

42

45

48

T j ( C)

Figure 2. Steady state reactor temperature vs. steady state jacket temperature

Steady-State Operating Point SS1 (stable) SS2 (unstable) SS3 (stable)

(kmol.m -3) 8.5615 5.518 2.359

C Ass

(0C) 38.2 66.1 95.1

T ss

Table 2. Steady-state operating points

4.1 Mathematical process model


The following model structure can be developed for the prescribed reactor performing the material balance and energy balance:
dC dt
A

F (C V

Af

E C K 0 exp RT

(8)

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

International Journal of Chemical Reactor Engineering

Vol. 5 [2007], Article A42

H dT F E U A = (T f T ) + K 0 exp RT C A V 0 (T T dt V Cp Cp

).

(9)

In the above model, A is the heat transfer area, C A the concentration of reactant A in the reactor (or product stream), C Af the concentration of A in the feed stream, C p the heat capacity, F the volumetric flow rate of the feed as well as the product stream, K 0 the pre-exponential factor, R the universal gas constant, t the time, T the reactor (or product stream) temperature, T f the feed temperature, T j the jacket temperature, U 0 the overall heat transfer coefficient, V the reactor volume, E the activation energy,

( H )

the heat of reaction, and the

density. In the present study, the developed reactor model has been simulated using the fourth-order Runge-Kutta integration approach.

4.2 Controller synthesis


The control objective of the example reaction system is to maintain the reactor temperature ( T ) at its desired value by manipulating the coolant flow rate. However, for the ease of GMC controller synthesis, the coolant jacket temperature ( T j ) was used as manipulated input instead of flow rate. A simple steady state correlation can easily be developed between the coolant flow rate and temperature using the energy balance equation around the jacket (Luyben, 1990). Moreover, to investigate the high-quality of the hybrid GMC-ASE controller, the complex CSTR has been chosen due to mainly the existence of multiple steady states. Figure 2 shows that the output multiplicities exist with respect to the coolant temperature; so, T j is directly considered as the manipulated input. To meet the operational objective we will design two controllers, namely GMC-ASE and conventional PI. 4.2.1 GMC-ASE structure As noted previously, the GMC-ASE structure is formulated combining the GMC controller with the ASE estimator. The hybrid GMC-ASE structure is shown in Figure 3. In the following, the design of each individual component of the control structure is given. 4.2.1.1 GMC controller It is easy to calculate the relative order of process output with respect to input as unity. Therefore, to derive the GMC equation for the concerned CSTR, we will follow the theory as reported in Section 2. Accordingly,
F f ( x , d ) + g 1 ( u , x , d ) = T& = V

(T

)+

H K C p

E exp C RT

U 0A (T T VC p

(10)

Substituting the above equation in Equation (2) and rearranging, one finally obtains the GMC controller equation:
T = T V C U
0 p F (T A V

T ) +

( H ) K C p

exp

( E ) C RT

K 1e K

e dt

(11)

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Jana: Nonlinear State Estimation and Control of CSTR

GMC-ASE Controller
y sp

GMC +

Process

Output Map

x ASE

Figure 3. GMC-ASE control structure where, is the error to the controller and T sp is the set point value of T . In the above Equation (11), K 1 and K 2 are the GMC tuning parameters. The values of these parameters have been determined based on the guidelines reported in Section 2 as: K 1 = 520.0 and K 2 = 478.0 . In the present case, the reactor temperature (true state) is assumed as measured variable, whereas the reactor concentration is considered as poorly known extra state. It is obvious in the above controller Equation (11) that the both states T and C A are required for the implementation of the GMC controller. Since C A is imprecisely known so we have to estimate it. The estimation of the reactor concentration is mostly preferred when reasonably good estimation is possible than the direct measurement of the same. The reasons are: the component analyzers are difficult to maintain and that also entail significant measurement delays. The key purpose to incorporate the state estimator within the GMC structure is to estimate the poorly known reactor concentration based on the measured temperature.
e (= T
sp

T)

4.2.1.2 ASE estimator The basic model structure that has been used to design the ASE scheme is given as:
dT F = (T dt V dC A = 0 .0 dt H T )+ C p E K 0 exp C RT U 0A (T T V C p

(12)

As stated earlier, the above predictor model comprises of the energy balance equation and the extra state equation that has no dynamics. Obviously, here is the reactor concentration. Combining Equations (6), (7) and (12), we have the following adaptive state estimator:

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

International Journal of Chemical Reactor Engineering

Vol. 5 [2007], Article A42

& T 0 = & C A 0

- H C p

E K 0 exp RT 0

F T (T + V C A

T )

U0A (T T V C p 0

(13)

T T H E C K 0 exp RT p 2
2

where T and C A are estimates of T and C A , respectively. In the above estimator structure, is the tuning parameter. The value of this parameter is obtained based on the technique presented by Farza et al. (1999) as: = 36 .

4.2.2 Conventional PI A proportional integral controller, which has been employed to keep the reactor temperature at a desired value T sp , is written as:
T j=T
js

1t + K c e + e dt 0

(14)

where, T js is the bias signal, K c the proportional gain and the integral time constant. The tuning parameters are chosen based on the integral square error (ISE) performance criteria as: K c = 100.0 and = 20.0. The jacket temperature is bounded as: 15 T j 66 0C.

5. RESULTS AND DISCUSSION


The present work provides the comparative performance study of the proposed control scheme and the conventional PI law on a simulated reactor. The control objective in this simulation-based work is to maintain the CSTR at the high conversion, high temperature steady state operating point. For the example reactor, this operating point (SS3) is stable and therefore, there is no such control difficulty. The main difficulty arises when the operating point is unstable. Sometimes the reactor is preferred to operate at the unstable operating point (SS2), although medium conversion is achieved, due to the following reasons: the low temperature steady state (SS1) causes very low yields; the high temperature steady state (SS3) may be very high, causing unsafe conditions, destroying the catalyst, degrading the product quality, and so on. In this challenging situation, it is quite difficult to achieve satisfactory closed-loop performance. However, in order to ensure a high-quality set point tracking performance by the proposed controller, the closed-loop simulation experiment has been performed running the reactor at all three steady state operating points. In the following, the convergence capability of the designed ASE estimator has been tested first. In the subsequent part, the comparative performance study has been carried out between the proposed GMC-ASE and traditional PI controller.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Jana: Nonlinear State Estimation and Control of CSTR

5.1 Open-loop performance of ASE


5.1.1 Rejection of initialization error In Figure 4, the initialization error rejection performance of the ASE scheme has been depicted. In the present study, +10% error in the reactor concentration (8.56159.42) is introduced at the beginning of the open-loop simulation experiment. It is obvious in the figure that very fast convergence of the estimation error towards zero has been achieved by the proposed observation approach. The estimation of the reactor temperature is also slightly affected by the initialization error.
38.30 P rocess O utput E stim ate d O utput 38.25

T ( C)
3

38.20

38.15

38.10 10.0 9.5 9.0 8.5 8.0 7.5

CA (kmol/m )

-1

2
T im e (hr)

Figure 4. Comparison of the estimated outputs and process outputs with +10% initialization error in reactor concentration (changed from 8.5615 to 9.42) 5.1.2 Disturbance in heat of reaction Figure 5 compares the estimated outputs and true process outputs with two consecutive step changes, +20% and 16.7%, in the heat of reaction (step increase: 59607157 at time = 5 hr, and step decrease: 71575960 at time = 30 hr). In this case, we have observed very interesting reactor dynamics. Before introducing any step change in the heat of reaction, the reactor was at operating point SS1. As +20% disturbance has been considered, the reactor achieved a new steady state operating point that is not any one among SS1, SS2 and SS3. Again when the disturbance is reduced to its initial constant value (= 5960), the reactor did not come back to the initial steady state SS1, it stayed at SS3. In this diverse situation, the ASE estimator also confirmed its convergence capability.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

10

International Journal of Chemical Reactor Engineering

Vol. 5 [2007], Article A42

160 140 120 T ( C) 100 80 60 40 20 10 8 CA (kmol/m )


3

Process Output Estimated Output

6 4 2 0

10

15

20

25 Time (hr)

30

35

40

45

50

Figure 5. Comparison of the estimated outputs and process outputs with step changes in heat of reaction (changed from 5960 to 7157 at time = 5 hr and then from 7157 to 5960 at time = 30 hr)

5.1.3 Disturbance in overall heat transfer coefficient The performance of the proposed state observer has been tested considering +10% and then 9.1% step changes in the overall heat transfer coefficient (step increase: 150165 at time = 5 hr, and step decrease: 165150 at time = 30 hr). Figure 6 clearly shows an excellent agreement between the process outputs and the estimator outputs. In fact, in the figure the process/estimator discrepancy is so close to zero that no differences are perceptible between the outputs.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Jana: Nonlinear State Estimation and Control of CSTR

11

38.4 38.2 38.0 T ( C) 37.8 37.6 37.4 37.2 8.64 Process Output Estimated Output

0 3

8.62 CA (kmol/m )

8.60

8.58

8.56

10

15

20

25 Time (hr)

30

35

40

45

50

Figure 6. Comparison of the estimated outputs and process outputs with step changes in overall heat transfer coefficient (changed from 150 to 165 at time = 5 hr and then from 165 to 150 at time = 30 hr)

5.1.4 Uncertainty in activation energy Figure 7 illustrates the tracking performance of the ASE scheme under uncertain activation energy. In the present study, the activation energy has been considered with a different value in the estimator than in the theoretical process (step decrease: 1184311643.8 at time = 5 hr, and step increase: 11643.811843 at time = 15 hr). The simulation result shows that the proposed observation approach provided promising convergence of the estimation error under uncertainty.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

12

International Journal of Chemical Reactor Engineering

Vol. 5 [2007], Article A42

160 140 120 100 T ( C) 80 60 40 20 15 12 CA (kmol/m ) 9 6 3 0 Process Output Estimated Output


0

12

15 Tim e (hr)

18

21

24

27

30

Figure 7. Comparison of the estimated outputs and process outputs under uncertainty in activation energy (changed from 11843 to 11643.8 at time = 5 hr and then from 11643.8 to 11843 at time = 15 hr)

5.2 Closed-loop performance of GMC-ASE and PI


5.2.1 Servo problem Figure 8 represents the comparative set point tracking performance between the hybrid GMC-ASE and PI controllers for +73.0% (38.266.1), +43.9% (66.195.1), and subsequently 30.5% (95.166.1) set point step changes in reactor temperature. Under these large step changes, the PI responses provide worse closed-loop performance and a slow approach to reach at target temperature. It is very prominent in the figure that the PI controller provides significantly large overshoot when the set point temperature is changed from 38.2 (stable steady state) to 66.10C (unstable steady state). On the other hand, the GMC control scheme shows relatively improved performance.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Jana: Nonlinear State Estimation and Control of CSTR

13

120 105 90 T ( C) 75 60 45 30 70 60 50 Tj ( C) 40 30 20 10 Set Point PI GMC

10

15

20

25

30 Time (hr)

35

40

45

50

55

60

Figure 8. Comparison of the GMC-ASE and PI controllers for large set point step changes in reactor temperature (changed from 38.2 to 66.1, 66.1 to 95.1, and 95.1 to 66.1)

5.2.2 Regulatory problem The regulatory behavior of the GMC-ASE has been compared with that of the PI controller in Figure 9 for the rejection of 40% step disturbance in feed temperature (25150C) at time = 5 hr. It is obvious from the simulation experiment that the PI controller failed to take immediate necessary actions as the operating temperature of the reactor started to fall down with step decrease in feed temperature. As a result, the reactor starts with a deviation to return towards the target temperature. Whereas, the GMC law shows excellent disturbance rejection performance compared to the traditional PI controller.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

14

International Journal of Chemical Reactor Engineering

Vol. 5 [2007], Article A42

38.4 38.3 38.2 TC 38.1 38.0 37.9 37.8 70 60 50 Tj ( C)


0

Set Point PI GMC

40 30 20

10 Time (hr)

12

14

16

18

20

Figure 9. Comparative regulatory performance of the GMC-ASE and PI controllers for - 40% step change in feed temperature (changed from 25 to 15 at time = 5 hr)

6. CONCLUSIONS
The main aim of the present research program is to develop a high-quality control scheme that can provide satisfactory performance even at critical situations. To meet this objective we have attempted to design a hybrid GMC-ASE control algorithm. The performance of this proposed control strategy has been investigated on a continuous stirred tank reactor that has multiple steady states. First, an adaptive state estimator has been designed for the example CSTR. Then the proposed nonlinear observation scheme showed the effectiveness in tracking time-varying extra state. In the next part, we have developed a hybrid GMC-ASE control structure. The superior performance of the nonlinear GMC-ASE scheme has been observed over the conventional PI controller by performing closed-loop simulation experiments.

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Jana: Nonlinear State Estimation and Control of CSTR

15

REFERENCES
Bullinger, E., Allgower, F., An Adaptive High-gain Observer for Nonlinear Systems, Proc. 36th CDC, San Diego, 4348-4353 (1997). Chen, Y.H., Adaptive Robust Observers for Non-linear Uncertain Systems, Int. J. Systems Sci., Vol. 21, 803-814 (1990). Ciccarella, G., Dalla Mora, M., Germani, A., A Luenberger-like Observer for Nonlinear Systems, Int. J. Control, Vol. 57, 537-556 (1993). Clarke-Pringle, T., MacGregor, J.F., Nonlinear Adaptive Temperature Control of Multi-product, Semi-batch Polymerization Reactors, Comp. Chem. Eng., Vol. 21, 1395-1409 (1997). Deza, F., Busvelle, E., Gauthier, J.P., Rakotopara, D., High Gain Estimation for Nonlinear Systems, Syst. & Cont. Let., Vol. 18, 295-299 (1992). Dochain, D., State and Parameter Estimation in Chemical and Biochemical Processes: A Tutorial, J. Proc. Cont., Vol. 13, 801-818 (2003). Farza, M., Busawon, K., Hammouri, H., Simple Nonlinear Observers for On-line Estimation of Kinetic Rates in Bioreactors, Automatica, Vol. 34, 301-318 (1998). Farza, M., Hammouri, H., Jallut, C., Lieto, J., State Observation of a Nonlinear System: Application to (Bio) chemical Processes, AICHE J., Vol. 45, 93-106 (1999). Gauthier, J.P., Hammouri, H., Othman, S., A Simple Observer for Nonlinear Systems-Application to Bioreactors, IEEE Trans. Automat. Control, Vol. 37, 875-880 (1992). Gibon-Fargeot, A.M., Hammouri, H., Celle, F., Nonlinear Observers for Chemical Reactors, Chem. Eng. Sci., Vol. 49, 2287-2300 (1994). Gonzalez-Trejo, J., Ramirez, J.A., Fernandez, G., Robust Control with Uncertainty Estimation for Feedback Linearizable Systems: Application to Control of Distillation Columns, J. Proc. Cont., Vol. 9, 221-231 (1999). Guo, B., Jiang, A., Hua, X., Jutan, A., Nonlinear Adaptive Control for Multivariable Chemical Processes, Chem. Eng. Sci., Vol. 56, 67816791 (2001). Henson, M.A., Nonlinear Model Predictive Control: Current Status and Future Directions, Comp. Chem. Eng., Vol. 23, 187-202 (1998). Henson, M.A., Seborg, D.E., Critique of Exact Linearization Strategies for Process Control, J. Proc. Cont., Vol. 1, 122-139 (1991). Hua, X., Jutan, A., Nonlinear Inferential Cascade Control of Exothermic Fixed-bed Reactors, AICHE J., Vol. 46, 980-996 (2000). Isidori, A., Nonlinear Control Systems, Springer-Verlag, New York (1989). Jana, A.K., Samanta, A.N., Ganguly, S., Nonlinear Model-based Control Algorithm for a Distillation Column using Software Sensor, ISA Transactions, Vol. 44, 259-271 (2005a). Jana, A.K., Samanta, A.N., Ganguly, S., Globally Linearized Control on Diabatic Continuous Stirred Tank Reactor: A Case Study, ISA Transactions, Vol. 44, 423-444 (2005b).

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

16

International Journal of Chemical Reactor Engineering

Vol. 5 [2007], Article A42

Kantor, J.C., A Finite Dimensional Nonlinear Observer for an Exothermic Stirred-tank Reactor, Chem. Eng. Sci., Vol. 44, 1503-1510 (1989). Kazantzis, N., Kravaris, C., A Nonlinear Luenberger-type Observer with Application to Catalytic Activity Estimation, Proc. Acc, 1756 (1995). Kim, K.J., Choi, K.Y., On-line Estimation and Control of a Continuous Stirred Tank Polymerization Reactor, J. Proc. Cont., Vol. 1, 96-110 (1991). Kravaris, C., Chung, C.B., Nonlinear State Feedback Synthesis by Global Input/Output Linearization, AICHE J., Vol. 33, 592-603 (1987). Kreisselmeier, G., Adaptive Observers with Exponential Rate of Convergence, IEEE Trans. Automat. Control, Vol. 22, 2-8 (1977). Krener, A.J., Isidori, A., Linearization by Output Injection and Non-linear Observers, Syst. & Cont. Let., Vol. 3, 47-52 (1983). Krener, A.J., Respondek, W., Nonlinear Observers with Linearizable Error Dynamics, SIAM J. Control Optim., Vol. 23, 197-216 (1985). Lee, P.L., Sullivan, G.R., Generic Model Control (GMC), Comp. Chem. Eng., Vol. 12, 573-580 (1988). Luyben, W.L., Process Modeling, Simulation, and Control for Chemical Engineers, McGraw-Hill, Singapore (1990). Magni, L., De Nicolao, G., Magnani, L., Scattolini, R., A Stabilizing Model-based Predictive Control Algorithm for Nonlinear Systems, Automatica, Vol. 37, 1351-1362 (2001). Rawlings, J.B., Meadows, E.S., Muske, K.R., Nonlinear Model Predictive Control: A Tutorial and Survey, IFAC Symposium on Advanced Control of Chemical Processes, Kyoto, Japan, 203-214 (1994). Signal, P.D., Lee, P.L., Generic Model Adaptive Control, Chem. Eng. Commun., Vol. 115, 35-52 (1992). Stephanopoulos, G., Chemical Process Control: An Introduction to Theory and Practice, Prentice-Hall of India Private Limited, New Delhi (1998). Yamuna, K.R., Gangiah, K., Adaptive Generic Model Control: Dual Composition Control of Distillation, AICHE J., Vol. 37, 1634-1642 (1991). Zeitz, M., The Extended Luenberger Observer for Nonlinear Systems, Syst. & Cont. Let., Vol. 9, 149-156 (1987).

Brought to you by | University of the Philippines Diliman (University of the Philippines Diliman) Authenticated | 172.16.1.226 Download Date | 3/8/12 3:53 AM

Вам также может понравиться