Al-Khwarizmi Engineering Journal Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) Implementation of Neural Control for Continuous Stirred Tank Reactor (CSTR) Karima M. Putrus Department of Chemical Engineering/ University of Technology Email: Karima_cheng@yahoo.com (Received 29 April 2010; Accepted 20 March 2011) Abstract In this paper a dynamic behavior and control of a jacketed continuous stirred tank reactor (CSTR) is developed using different control strategies, conventional feedback control (PI and PID), and neural network (NARMA-L2, and NN Predictive) control. The dynamic model for CSTR process is described by a first order lag system with dead time. The optimum tuning of control parameters are found by two different methods; Frequency Analysis Curve method (Bode diagram) and Process Reaction Curve using the mean of Square Error (MSE) method. It is found that the Process Reaction Curve method is better than the Frequency Analysis Curve method and PID feedback controller is better than PI feedback controller. The results show that the artificial neural network is the best method to control the CSTR process and it is better than the conventional method because it has smaller value of mean square error (MSE). MATLAB program is used as a tool of solution for all cases used in the present work. Keywords: Predictive control, PID control, neural network, nonlinear control, continuous stirred tank reactor. 1. Introduction Chemical reactors are the most influential and therefore the important units that a chemical engineer will encounter. To ensure the successful operation of a continuous stirred tank reactor (CSTR), it is necessary to understand their dynamic characteristics. A good understanding will ultimately enable effective control systems design. The aim of these notes is to introduce some basic concepts of chemical reaction systems modeling and develop simulation models for CSTR's. Non-linear and linear systems descriptions are derived [1]. Chemical process control requires intelligent monitoring due to the dynamic nature of the chemical reactions and non- linear functional relationships between the input and output variables are involved. CSTR is one of the major processing units in chemical engineering; such a problem remains too complex to be solved by the known techniques; therefore, neural networks models can provide good optimal solutions for many applications [2]. Control and optimization problems are some of the more difficult applications for ANN [3]. The mapping functions that must be learned are generally very complex in nature and the problem constraints that must be satisfied are often conflicting (Control problems, typically, require nonlinear time dependent mapping of input signals). Neural Networks based on adaptive resonance theory are equipped with unique computational abilities that are needed to function autonomously in a changing environment [4]. Carpenter et al. implemented a self-organizing and adaptive neural network system in the monitoring and control of the behavior of an industrial / platinum flotation plant (Hydrometallurgical process). Other network formalisms, namely radial basis function (RBF) and adaptive resonance theory-2 (ART2) networks have also been employed for fault This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 40 detection, diagnosis and process monitoring task [5, 6]. LeonardJ et al., Whitley et al. and Krishnaveni et al. used ANN based systems to control patterns estimation for UPFC in power flow problem [7, 8, 9]. Zhang et al used a locally recurrent Neural Network to model the pH dynamics in a CSTR reactor [10]. Petia et al applied a feed forward network in the modeling and control of a fed- batch crystallization process [11]. 1.1. Types of Feedback Controller The block diagram of feedback controller is shown in Fig (1). The most important types of industrial feedback controllers include [13]: Fig.1. Block Diagram of Feedback Controller. A) Proportional Control sc ctEKtc  )()( …(1) B) Proportional – Integral Controller Most controller loops use (PI) controller; the advantage of this controller is that it has quick response for large error and does not have set point offset. The measured variable can be returned to the set point without excessive oscillation. sc ctEKtc t dttEc K I  0 )()()(  …(2) C) Proportional – Integral - Derivative Controller (PID) The adjustment of the input variable is accomplished using all three methods: proportional, integral, and derivative. The advantage of (PID) action is that it gives response. The disadvantage is that you have to tune three parameters. sc ctEKtc t dttEc K I  0 )()()(  …(3) 1.2. Description of the Artificial Neural Networks Model Referring to Figs. (2 and 3), in the network functions, each neuron receives a signal from the neurons in the previous layer, and each of those signals is multiplied by a separate weight value. The weighted inputs are summed, and passed through a limiting function which scales the output to a fixed range of values. The output of the limiter is then broadcast to all of the neurons in the next layer. So, to use the network to solve a problem, we apply the input values to the inputs of the first layer, allow the signals to propagate through the network, and read the output values [14]. There are two types of artificial neural network: + - E M + +C ym ysp Controller Final Control Element Process Measuring Device y Disturbance Variable This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 41 Output YK X1 W1K X1 W1K Activation Function F (.) Bias bK X1 W1K Synaptic Weights Fig.2. Nonlinear Model of Neuron Fig.3. A Generalized Network. A) NN Predictive Control The model predictive control method is based on the receding horizon technique. The neural network model predicts the plant response over a specified time horizon. The predictions are used by a numerical optimization program to determine the control signal that minimizes the following performance criterion over the specified horizon [3].                     21 1 1 2 2 jtujtup jtyjty N Nu j mpN J …(4) Where N1, N2, and Nu define the horizons over which the tracking error and the control increments are evaluated. The u variable is the tentative control signal; is the desired response, and ym is the network model response. The p value determines the contribution that the sum of the squares of the control increments has to the performance index. The block diagram in Figs (4 and 5) illustrates the model predictive control process. The controller consists of the neural network plant model and the optimization block. The optimization block determines the values of u' that minimize J, and then the optimal u is the input to the plant. The controller block is implemented in stimulant; the neural network plant model uses previous inputs and previous plant outputs to predict future values of the plant output [15]. This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 42 Fig.4. NN Predictive Control. Fig.5. Block Diagram of the NN Predictive Control. B) Narma-L2 Control Using the NARMA-L2 model, you can obtain the controller [14] ...(5) Which is realizable for (d ≥2). The following figure is a block diagram of the NARMA-L2 controller. This controller can be implemented with the previously identified NARMA-L2 plant model, as shown in Figs (6 and 7). Therefore, the aim of the present work is to propose (NARMA-L2 and NN Predictive) network, which is used to model the dynamics of the CSTR problem and a typical problem that was solved. Neural networks are well known for their ability to imitate the skill of experts by capturing knowledge, generalizing non-linear functional relationship between input- output variable, and they provide a flexible way of handling complex and intelligent information processing. Artificial Neural Networks (ANNs) have been shown to be effective as computational processors for various tasks including data compression, classification, combinatorial optimization problem solving, modeling and forecasting, and adaptive control. The neural network predictive controller developed in this paper uses a neural network model of a nonlinear plant to predict future plant performance. The controller calculates the control input that will optimize plant performance over a specified future time horizon. Simulation of the neural network based predictive control of the continuous stirred tank reactor is presented. This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 43 Fig.6. Block Diagram of the NARMA-L2 Control. Fig.7. Narma L2 Control Structure. 2. Theory 2.1. Modeling Process Scheme A jacketed continuous stirred tank reactor of Butru [16] is taken as an example process in the present work which has been used to carry out the exothermic chemical reaction of acetic anhydride with water. A dynamic model for first order CSTR is developed to predict the transient responses to normal and abnormal (upsets) conditions using servo technique. The mathematical model obtained using mass and energy balances is used to develop a model for controlling, feed rate, the jacket rate and both feed rate and jacket rate using several control strategies, feedback and neural network (NARMA-L2, and NN Predictive) control for two different scheme. The chemical reaction is: COOHCHO 322HO2CO)3(CH  …(6) Figure (8) shows the schematic diagram of jacketed reactor. The anhydride (feed) enters the reactor through control valve (CV1 ) with flow rate (F1) and temperature (T1). The concentration of anhydride in feed stream and reactor are (CNo and CN) respectively; while the concentration of the product (acetic acid) is (CA) and the acidity of solution is (pH). The temperature and volume of This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 44 reactants in the reactor are (T and V) respectively . The reactor is cooled by water-jacket through control valve (CV2) with hold up of Vj. The inlet temperature, outlet temperature and flow rate of cooled water are (Tj, Tjo and Fj) respectively. The inner and surface areas of the reactor are (Ai and Ao) respectively. The experimental results of this work are taken from the same reference [16]. The properties of the feed solutions and parameters of the reactor are presented in Tables (1 and 2) and the approximate correlation of the above reaction is given by: Ca = 198.182 EXP (-1.849 pH) Total Mass Balance 021  dt dV FFF …(7) 21 FFF  …(8) Mass Balance on component (A) dt AdCVRT E eACoVKAFCAoCF   1 …(9) Heat Balance on component (A) dt dT CVTTUA eCVKHTCFTCFTCF PjH RT E AorPPP      )( 2211 …(10) Heat Balance on the jacket dt jdT PjCjjVTjTHUAjTjoTjPCjjF   )()( …(11) Fig.8. Schematic Diagram of Jacketed CSTR Process. Table 1, Properties of the Feed Solutions at the 250C. Material Density Kg/m3 Concen- tration mole/L Ionization constant Molecular weight Acetic acid 1528 19.1 Completel y ionized 40 Distillate water 1050 14 1.75*10-5 60.05 Table 2, Parameters of the Reactor of Butrus [16]. Martial Property Area for heat exchange 5.33x10-2 (m3) Ideal gas constant 8.314 (kJ/kmol K) Heat of reaction - 45000 (kJ/ kmol) Activation energy 36700 (kJ/kmol) Overall heat transfer coefficient 1480 ( w/ m2 K) Heat capacity for jacket 4.186 (kJ/kg K) Density of acetic acid 1.05 (kg/m3) Density of acetic anhydride 1.08 (kg/m3) Density for jacket 997 (kg/m3) Feed flow rate 1.83x10-2 (m3/min) Distillate water flow rate 1.714x10-5 (m3/min) Jacket feed flow rate 0.0235 (m3/min) Jacket output flow rate 0.0235 (m3/min) Inlet temperature of reactant 298 (K) Jacket feed temperature 298 (K) Jacket outlet temperature 299 (K) Reaction rate constant 9.36 (min-1) Concentration inlet 0.999 (kmol/m3) Volume of reactor 2x10-3 (m3) Volume of jacket 2x10-3 (m3) Height of reactor 0.5 (m) This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 45 2.2. Control Schemes 2.2.1 Scheme A/ Control of the Reactant Concentration The acetic acid concentration CA was controlled by manipulating the reactant flow rate (F1) as shown in Fig.(9). The holdup of reactor was adjusted by using a level controller; hence, the reactant flow rate is kept constant. The transfer function for Scheme A is: 14.15 15.0129 )(1    s e sG … (12) The approximate correlation of this relation was established by least squares method which gives the experimental form. Fig.9. Scheme A of Control of Reactant Concentration. Fig.10. Scheme B of Control of Reactor Temperature. 2.2.2 (Scheme B)/ Control of the Reactor Temperature. In this scheme, the reactor temperature T is controlled by manipulating the cooling water flow rate (F1) as shown in Fig.(10). The transfer function for Scheme B is: 16.17 15.0173 )(2    s e sG …(13) This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 46 2.3. Simulation Program Computer simulation was carried out using MATLAB. It starts with the creation of a mathematical model and the obtained equations are solved by using an appropriate calculation method. MATLAB’s Graphical User Interface (GUI) can be used for investigating the static and dynamic behavior and adaptive control of the nonlinear system represented by continuous stirred tank reactor (CSTR). There are many types of models; the main categories are real models and computer models. Based on this division, also simulation can be done for a real model investigation of its behavior as a result of input stimulation [17]. The importance of computer simulation will grow in the future when computers are faster [18]. 3. Results and Discussion The first part is to study the dynamic behavior of the system theoretically and plot the step responses where the transfer functions between the controlled variables and manipulated variables are computed from the experimental work of Butrus [16]. The second part is to study the closed loop system which is the main aim of this work through applying different control strategies; these strategies are: feedback control, NARMA-L2 control and NN Predictive control. 3.1. Dynamic Behavior In this section, the dynamic responses are studied for a different step change in the manipulated variable which is the reactant flow rate F1 in order to study the effect of this change on the controlled variable of the acetic acid concentration CA(in terms of pH (scheme A) and the temperature of reactor T (scheme B)). These changes are: (-50 % and -100%) step change in the feed flow rate F1. The responses are shown in Figs (11 and 12). These results are obtained by using computer simulation programs. From these Figures, it can be seen that the concentration increases and the temperature decreases with the decreasing of the feed flow rate for different step changes. Fig.11. Transient Response of Concentration for Step Change in Feed Flowrate. 0 0.2 0.4 0.6 0.8 0 10 20 30 40 50 60 Time (min) C o n ce n tr a ti o n ( k m o l/ m 3 ) -50% step change -100% step change This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 47 -6 -5 -4 -3 -2 -1 0 0 10 20 30 40 50 60 Time (min) T em p er at u re d if f. ( K ) -50% step change -100% step change Fig.12. Transient Response of Reactor Temperature for Step Change in Feed Flowrate. 3.2. The Closed Loop System Feedback controller is applied using PI and PID controller modes to control the CSTR process; therefore, tuning the control parameters (proportional gain (KC), time integral (τI) and time derivative (τD)) must be done first. The optimum values of the controller parameters (KC, τI, τD) are obtained using computer simulation programs based on mean square error (MSE). The control tuning is found by two different methods which are Process Reaction Curve (PRC) and Frequency Analysis (Bode diagram). a). Results of Control Tuning Using PI, PID Controller. In this section, PI and PID controller is used to control CSTR reactor for the two transfer functions (Eq. 13 and 14) as follows: Figs (13 and 14) show the Bode diagram of the closed loop system for the first and second transfer functions respectively to determine the value of ultimate gain (Ku) and ultimate period of sustained cycling (Pu) in order to tune the adjusted parameters values for the CSTR reactor as shown in Tables (3 and 4). On the other hand, Figs (15 and 16) show the transient response of different control tuning methods with PI controller mode; while Figs (17 and 18) show the (Time  absolute square error) versus time for the first and second transfer functions respectively. The results in Figs (19 and 20) show the transient response of different control tuning methods with PID controller mode; while Figs (21 and 22) show the (Time  absolute square error) versus time for the first and second transfer functions respectively. The control parameters of PI and PID for first transfer function are listed in Tables (3 and 4) while the same parameters for the second transfer function are listed in Tables (5 and 6). From the Comparison of the Process Reaction Curve method with Frequency Analysis Curve method, it is concluded that the tuning by using Frequency Analysis Curve method is worse than Process Reaction Curve method because Frequency Analysis Curve method depends on closed loop system; while Process Reaction Curve method depends on open loop system and also the proportional gains are larger for the Process Reaction Curve method. It is clear that PID mode is better than PI mode because of the good tuning of adjusted parameters values in PID mode which gives the smaller overshoot and makes the system with smaller oscillation and reaches the new steady state value in shorter time and reaches the new steady state value in shorter time. Also the area under the curve of the Process Reaction Curve method is lower than the area under the curve of the Frequency Analysis Curve method and values of the MSE in the first method are less than those in the second method. The results agree with the work of Derar [19]. This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 48 Fig.13. Bode Diagram of the CSTR Process. Fig.14. Bode Diagram of the CSTR Process. Fig.15. Transient Response of the CSTR Process with PI Feedback Controller. -50 0 50 M ag ni tu de ( dB ) 10 -4 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 -90 0 90 180 270 360 P ha se ( d eg ) Bode Diagram Frequency (rad/sec) 0 2 4 6 8 10 12 -0.5 0 0.5 1 1.5 2 2.5 Time (min) p H Bode Digram Process Reaction Curve -50 -40 -30 -20 -10 0 10 20 30 40 50 M ag ni tu de ( dB ) 10 -4 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 -90 0 90 180 270 360 P ha se ( de g) Bode Diagram Frequency (rad/sec) This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 49 0 2 4 6 8 10 12 -0.025 -0.02 -0.015 -0.01 -0.005 0 0.005 0.01 0.015 0.02 Time (min) T im e * m ea n s q u ar e e rr o r Bode Digram Process Reaction Curve Fig.16. Transient Response of the CSTR Process with PI Feedback Controller. Fig.17. The Relationship between the )Time  mean Square Error) and Time with PI Feedback Controller. Fig.18. Time  mean Square Error versus Time with PI Feedback Controller. 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 -0.5 0 0.5 1 1.5 2 Time (min) p H Bode Digram Process Reaction Curve 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 Time (min) T im e* m ea n sq ua re e rr or Bode Digram Process Reaction Curve This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 50 Fig.19. Transient Response of the CSTR Process with PID Feedback Controller. Fig.20. Transient Response of the CSTR Process with PID Feedback Controller. Fig.21. The Relationship between the (Time  mean Square Error) and Time with PID Feedback Controller . 0 0.5 1 1.5 2 2.5 3 3.5 -0.5 0 0.5 1 1.5 2 2.5 Time (min) p H Bode Digram Process Reaction Curve 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 -0.5 0 0.5 1 1.5 2 Time (min) p H Bode Digram Process Reaction Curve 0 0.5 1 1.5 2 2.5 3 3.5 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 Time (min) T im e* m ea n sq ua re e rr or Bode Digram Process Reaction Curve This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 51 Fig.22. Time  mean Square Error versus Time with PID Feedback Controller. Table 3, Control Parameters of PI Controller for First Transfer Function. Control Tuning Methods Controller Parameters MSE Kc τI τD Frequency Analysis Curve 0.9396 0.4943 - 2.5353*10 ^-4 Process Reaction Curve 0.9971 0.4899 - 1.0721*10 ^-4 Table 4, Control Parameters of PID Controller for First Transfer Function. Control Tuning Methods Controller Parameters MSE Kc τI τD Frequency Analysis Curve 1.2159 0.2966 0.0741 1.1071*10 ^-5 Process Reaction Curve 1.7665 0.3677 0.0544 0.3295 *10 ^-5 Table 5, Control Parameters of PI Controller for 2nd Transfer Function. Control Tuning Methods Controller Parameters MSE Kc τI τD Frequency Analysis Curve 0.3045 0.4945 - 3.6336*10 ^-4 Process Reaction Curve 0.6126 0.4911 - 1.7094*10 ^-4 Table 6, Control Parameters of PID Controller for 2nd Transfer Function. Control Tuning Methods Controller Parameters MSE Kc τI τD Frequency Analysis Curve 0.3941 0.2967 0.0742 5.0057*10 ^-5 Process Reaction Curve 0.9083 0.3679 0.0545 3.4917*10 ^-5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 Time (min) T im e* m ea n sq ua re e rr or Bode Digram Process Reaction Curve This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 52 b). Results of Control Tuning Using NARMA-L2 and NN Predictive Controller. NARMA-L2 algorithm and NN Predictive control are implemented using back-propagation networks in this work, which depends on:  Changing the number of neurons in the hidden layer can represent the degree of the complexity of the system.  The ability of input layer to store information was used to represent the dynamic behavior of system by using the tapping delay lines for input/output signals. Figs. (23 and 24) show the transient response of NARMA-L2 and NN Predictive control respectively for the first transfer function only. Comparing NARMA-L2 with NN Predictive control, it can be seen that the NARMA-L2 control is better than NN Predictive control because the values of the MSE in the first method are less than those in the second method. This comparison is listed in Table (7). Finally the comparisons among feedback control, NARMA-L2 and NN Predictive control for 1st transfer function is listed in Table (8). Many authors have reported the same results [12, 19]. Fig.23. The Transient Response of NARMA-L2 Control. Fig.24. The Performance of NN Predictive Control. 0 1 2 3 4 5 6 7 8 9 10 0 0.2 0.4 0.6 0.8 1 1.2 1.4 time (sec) p H NARMA-L2 disired output This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 53 Table 7, Comparisons between NARMA-L2 and NN Predictive Control Transfer Function. Criteria NARMA-L2 NN Predictive MSE 3.4429*10^-6 4.2112*10^-6 Training 6.43*10^-3 6.99*10^-3 Validation 7.21*10^-4 6.91*10^-3 Test 8.45*10^-4 9.87*10^-4 Table 8, Comparisons among Feedback Control, NARMA- L2 and NN Predictive Control for 1st Transfer Function. Controllers MSE PID 0.3295 *10 ^-5 PI 1.0721*10 ^-4 NARMA-L2 3.4429*10^-6 NN Predictive 4.2112*10^-6 4. Conclusion 1. The Process Reaction Curve method is better than the Frequency Analysis Curve method. 2. PID feedback controller is better than PI feedback controller. 3. Implementation of artificial neural network is the best method to control the continuous stirred tank reactor (CSTR). 4. The NARMA-L2 control is better than NN Predictive control 5. Finally the network is successfully used to model and solve the CSTR problem keeping the system at its optimum. Notation A Magnitude of change AH Area of heat transfer CA0 Inlet concentration of acetic anhydride CA Outlet concentration of acetic anhydride d Modeling error( ) F Volumetric flow rate of distillate water F2 Volumetric flow rate of acetic anhydride Gc(s) Transfer function of controller Gd(s) Transfer function of disturbance H Nonlinear activation function L Linear activation function m Number of previous input M Model n Number of the previous output Neto The weighted sum of the inputs P Plant k Discrete time instant K Proportional gain K The integral decreasing factor Kp Gain of the process Q Number of patterns in training set Reactor output Neural model output Desired output of the plant Set point The output of network N1 [.] The output of network N2 [.] y Output variable V Volume of tank sgn Sigmund function s Laplacian variable t time Tcl(s) Transfer function of close loop T Temperature of reactor T1 Inlet temperature of reactant T2 Outlet temperature of reactant Tj0 Inlet temperature of coolant Tjj Outlet temperature of coolant td Time delay 5. References [1] Willis M.J., Dept. of Chemical and Process Engineering and Process Engineering, University of Newcastle, March, (2000). [2] Himmelblau D. M. (1978) Fault detection and diagnosis in chemical and Journal of Theoretical and Applied Information Technology © 2005 - 2008 JATIT. All rights reserved.www.jatit.org, [3] Petrochemical processes, Elsevier, Amsterdam. Patterson D. W, "Artificial Neural Networks Theory and Applications", Prentice Hall Simon & Schuster (Asia), Singapore pg 220, (1996). This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ Karima M. Putrus Al-Khwarizmi Engineering Journal, Vol. 7, No. 1, PP 39 - 55 (2011) 54 [4] Carpenter G. A and Grossberg S.),"The ART of Adaptive Pattern Recognition by Self- Organizing Neural Networks, IEEE Computer, pp 77-88, (1988). [5] Carpenter G. A., Grossberg S. and Rosen D. B," Fuzzy ART: Fast Stable Learning and categorization of Analog Input Pattern by an Adaptive Resonance System", Neural Networks, Vol. 4, pp 59-71, (1992). [6] Aldrich C., Moolman D. W and Van Deventer J. S; "Monitoring and Control of Hydrometallurgical Process with Self- Organizing and Adaptive Neural Net Systems", Comput. Chem. Engrg, Vol. 19 pp S803-S808, (1995). [7] Leonard J. A and Kramer M. A," Radial Basis Function Networks for Classifying",(1991). [8] Whitley J. R. and Davis J. F., A," Similarity- Based Approach to Interpretation of Sensor data Using Adaptive Resonance Theory", Comput. Chem. Engrg. Vol. 18, pp 637-661, (1994). [9] Krishnaveni K and Tulasi R.D, "ANN Based Control Patterns Estimator for UPFC Used in Power Flow Power", Journal of Theoretical and Applied Information Technology, Pg 45- 49, (2007). [10] Zhang J. and Julian A.,"A Sequential training strategy for locally recurrent Neural Networks", (1994). [11] Petia G. and Sebastiao F., "Application of Feed Forward Neural Network in Modeling and Control of a Fed-Batch Crystallization Process", Proceedings of World Academy of Science, Engineering and Technology, Vol. 12 Pg 65-70, ISSN 1307-6884, (2006). [12] Anna Vasičkaninová and Monika Bakošová, "Neural Network Predictive Control of a Chemical Reactor", Acta Chimica Slovaca, Vol.2, No.2, 21 – 36, (2009). [13] Luyben M. L. and Lnybem W. L.," Essentials of Process Control", McGraw Hill, New York. (1997). [14] Haykin .S. ,"Neural Network A comprehensive foundation" second edition prentice Hall International .Inc Hamilton .Ontario .Canada, (1999). [ 1 5 ] Montandon A. G., Borges R. M. and Henrique H M., "Constrained Nonlinear Model Predictive Control for Practical Application". School of chemical engineering /federal university of Uberlandia. MG. Brazil, (2007). [16] Butrus K. M., "On-Line Computer Control of A Continuous Stirred Tank Reactor", Ph. D., Thesis, University of Technology, (1999). [17] Jiří Vojtěšek, Petr Dostál, "Program for Simulation of a Continuous Stirred Tank Reactor in MATLAB’S GUI", 22nd European Conference on Modeling and Simulation, ECMS, (2008). [18] Ingham, J., Dunn, I. J., Heinzle, E., Přenosil, J. E., "Chemical Engineering Dynamics. An Introduction to Modeling and Computer Simulation", Second Completely Revised Edition, VCH Verlagsgesellshaft, Wilhelm, (2000). [ 1 9 ] Derar I. A., "Implementation of Neural Control for Neutralization Process", Ms. Thesis, university of technology, (2009). This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/ - 55 ، صفحة1، العدد 7لخوارزمي الھندسیة المجلد مجلة ا طرسبكریمھ مروكي 39 )2011( 55 استخدام نظام السیطره بالشبكھ العصبیھ للسیطره على مفاعل ذو الخلط المستمر طرسبكریمھ مروكي الحامعھ التكنولوجیھ /قسم الھندسھ الكیمیاویھ Karima_cheng@yahoo.com : االلكتروني البرید الخالصة ن CSTR)( كیمیاوي مستمر جید الخلطالسلوك الدینامیكي وطرق السیطرة لمفاعل البحث ھو دراسة إن الھدف من ھذا ة م واع مختلف وذلك باستخدام أن دي وعین و )PIDو PI ( المسیطرات مثل المسیطر التقلی بیة بن بكة العص یطر الش ا ( مس ؤي (و )نارم زان ذو ). التنب ي للخ ل الریاض ل المودی م تمثی د ت وق . لخلط المستمر بدالة تحویل من الدرجة األولى مع تأخیرا رات تم توصیف ر متغی یطر بط ا یالمس ین ھم اد Process Reaction Curve methodو Frequency Analysis Curve methodقتین مختلفت إلیج , KCأفضل قیم للمعامالت I , D .أللخطو تم استخدام معیار متوسط التربیع (MSE) كأساس للمقارنة بین الطریقتین أعاله. ع للخط CSTRالشبكة العصبیة الصناعیة ھي أفضل طریقة للسیطرة على ال ط التربی ار متوس ك الن معی ة و ذل رق التقلیدی ل أوھي أفضل من الط م . اق ت . االتل في جمیع الحفي ھذا البحث كأداة للح MATLABاستخدام برنامج This page was created using Nitro PDF trial software. To purchase, go to http://www.nitropdf.com/ http://www.nitropdf.com/