** R.G.Jamkar. II. Description of flow control system. *J.V.Kul karni

Size: px

Start display at page:

Download "** R.G.Jamkar. II. Description of flow control system. *J.V.Kul karni"

Rosa Eaton
5 years ago
Views:

1 Proceedings of the 2000 EEE nternational Conference on Control Applications MM5-5 2:20 Anchorage, Alaska, USA September 25-27,2000 NEURAL NETWORK BASED FLOW CONTROLLER *J.V.Kul karni ** R.G.Jamkar *Lecturer, Dept. of Electronics, KBP, Kopergaon **Lecturer, Dept. of nstrumentation, SGGS college of Engg. and Tech., Nanded (NDA), d. : rajujamkara hotmail.com. Abstract This paper presents an application of backpropogation neural network to control flow rate. A back propogation neural network is trained to learn inverse dynamics model of a flow control system and then configured as a direct controller to the process. The ability of neural network to learn inverse,dynamics model of the process plant is based on input vectors with a no a priori knowledge regarding process dynamics. For interfacing flow transmitter to personal computer & also for giving controller output to the valve from pgsonal computer, Add on card and signal conditioning cards are designed. Process is tested by using neural network controller. Experimental results shows that neural network gives satisfactory results to different set points & also exhibits better performance for load disturbance. 1. ntroduction The applications of neural networks to control systems have become increasingly important. The massive parallel processing and self-learning abilities of neural networks have been motivating for development of intelligent control. Without knowing the prior knowledge regarding the system dynamics, the neural network can be used as a direct controller instead of P, P, PD controllers. t is found that tracking performance, load disturbance rejection of neural network based controller [l] is best as compared to fuzzy logic controller, generalised predictive controller, P controller. The principal of neural network found applications in process control with increase of complexity of process. The Sideris and Yamamura [2] proposed a method, which they referred to as general learning and.specialised learning. n the general learning scheme the network is trained off line to learn a plants inverse dynamics directly. Even though many process exhibits significant non-linear behaviour, most controller design techniques are based on linear model [3]. The prevalence of linear control strategies is primarily due to two reasons. First they are well-established methods for the development of linear model from input/output data [4], while practical identification techniques for non-linear model are not still fully developed. Further more controller designed for nonlinear models is considerably more difficult than for linear models. The ability of Neural network to model almost any non linear functionality in a noisy environment with a priori knowledge suggest that they may provide a promising note for modelling non linear process. Neural network have been successfully used for a number of applications including sincere data analysis, fault detection & process modelling. One of the application of Neural network was developed by Heng Ming & others [5] for tracking controller for the industrial drives system. Maria C. Palancar & others [6] Proposed a control system based on two artificial neural networks for the ph neutralisation to an acidic liquid stream. Naveen Bhat & others [7] used back propagation net in developing non linear model of chemical systems. Marzuki Khalid and Sigemu Omatu [8] used BPNN to develop a temperature control system. Neural network is used as a direct controller Bhupindar S. Dayal & others [9] discussed the issues relating to the information contents of the data used to train the neural network components of these non linear predictive controls schemes. This paper gives an example where multilayered backpropogartion neural network is trained offline to perform as a controller for flow control system with no a priori knowledge regarding its dynamics.. Description of flow control system Fig.(l) shows a detailed circuit diagram of the hardware of the system. The system can be divided into following components. 1. Pump, 2. Orifice plate, 3. DPT, 4.Current to voltage converter, 5. Analog to digital converter, 6. Add on card, 7. Signal conditioning card. We have used a flow process set up installed in our laboratory. Photograph of the set-up is shown in fig.(2). A brief description and design of each unit is as follows. 1. Pump :- t is used to lift the water from a tank, to pass a water through a pipeline, which is to be controlled. 2. Orifice : t is restriction type flow measuring device having following specifications /00$ EEE 208

2 line size - 2, Material used - stainless steel with 16 diameter. t is square edge concentric fype orifice plate. The flanged taps measures the differential pressure proportional to the flow rate. This differential pressure is applied as input to differential pressure transmitter. 3. Differential pressure transmitter (DPT) : n our set up a rosemount 115 idpt is used. The delta cell is a variable capacitance sensing module. The differential pressure from orifice is applied to DPT. DPT converts this pressure into equivalent current signal which is given to current to voltage converter. 4. Current to voltage converter (CVC) : The circuit diagram of current to voltage converter is as shown in the figure (1). By using CVC the incoming current form DPT (4-2OmA) is converted to 0 to 2, V. This output is then given to analog to digital converter. 5. Analog to digital converter : As PC accepts signals in digital nature hence the output of CVC is given to analog to digital converter. CL 7109 CPL is uied as a analogue to digital converter, which is a 12-bit dual slope integrating type. 6. Design of Add-on card : Add-on card is used to interface ADCDAC to Personal computer which is as shown in fig. (l).the photograph of the same is shown in fig.(2). Add-on card consists of fgllowing components, i) PP 8255, ii) 74Ls138 decoder, and iii) Bi-directional Buffer Add on card is fixed into the PC input, output slots. Power supply for Add-on card is taken from the Personal computer s switch mode power supply. Two 8255s are used to interface ADC as well as DAC. Port A and Port C (upper) of st.8255 are used to interface analog to digital converter to personal computer. Port B and Port C lower of 1st 8255 is used to interface digital to analog converter to personal computer. Output of decoder is used to select at a time one 8255 from two 8255s. 7) Design of signal conditioning card : To convert measured flow rate signal into digital form & also to convert controller output to analog form, signal conditioning card is designed. The detail circuit diagram is shown in fig (1). The photograph of the same is shown in fig (3). This card contains A. Current to voltage converter (CVC) B. Analog to digital converter (ADC) C. Digital to analog converter (DAC) D. Voltage of current converter (VCC) The details of CVC and ADC are already explianed. The output of ADC is given to PC through This hexvalue i.e. flow is compared with the set point. Error between set point and measured flow rate is calculated and depending on this value of error neural network calculates controller output. This controller output is in digital nature so it is necessary to convert it into analog form so that stem position of the valve can be varied to control the flow rate. Hence the controller output is given to digital to analog convertor Here we have used CL 7541 digital to analog converter, which is a 12 bit DAC. Output of digital to analog converter is between 0 to 2V.To vary the position of stem of valve the controller output should be converted into current so that it can be given to E to P converter as shown in Fig (1). Hence VCC is designed to convert 0-2 V into 4-20 ma signal using opamps. The output of VCC is then given to E to P converter which in turn controls the position of stem of the valve and ultimately flow rate can be controlled Structure of neural network n the system which is developed to control flow rate, we used neural network having following structure. A three layered Neural Network having 6 neurone in the hidden layer and 12 neurones at the input layer and one neurone at the output layer. The activation function used is a sigmoidal function. The weight between input and hidden layer are adjusted by using delta rule. V. Design of software of system t is well known that for a personal computer based hardware to work, we have to develop appropriate software. Modular programming is used for software development. There are four subroutines called by main program as follows : 1) ADC interface program 2) Back propagation algorithm/dac routine 3) Online graph of process response program 4) Online block diagram program representation 1) ADC interface n this module, program asks user to enter set point between 0 to 2400 LPH and then it reads data i.e. flow rate from ADC. The measured flow rate is then subtracted from set point to find the error. 2) BPNN/DAC routine Neural network gives output corresponding to the error between set point and measured flow rate. This output is given to digital to analog converter.. This voltage is then given to VCC which converts voltage into current in between 4-20 ma 3) Online graph of process response (OGPR) routine n this module, program is written for plotting graph of process output i.e. flow rate with respect to time i.e. real time response. From this graph we can easily analyse the performance to the system. 4) Online Block diagram representation (OBDR) routine : This routine gives the online details of the various values of ADC output, controller output, set point etc. at the respective positions in the block diagram. 209

3 Main program calls all these routines and 161. displays graph of flow rate w.r.t. time also calculates error, controller output etc The flow chart of the same is shown in fig(5). [71. V. Result and conclusion Having trained the network for all the cases of input vectors,the neural network is configured as direct controller. Three sets of experimentation were conducted on the flow control system. n the first set of experiments the tracking performance of controller with respect to set point changes are studied. We have given set points as 1850 LPH, 100 LPH to the developed set up & obtained the results. The tracking performance of the neural network controller is as shown in fig (6) and (7) for two set points. t can be observed that the neural network controller tracked well for all set points. n addition, the controller operates extremely faster in achieving set points. The second set of experiment was carried out with the purpose of studying controller when load disturbance is imposed on the process. n order to do this the control valve was suddenly turned off and then turned on. Pcrforniancc of neural controller when load disturbance is applied is as shown in fig (8). The neural network controller showed very fast recovery under the effect of the load disturbance. Hence it can be seen that developed system behaves in a satisfactorily manner. As per the procedure outlined above, the experiment for flow control using neural controller is carried out. t can be seen that the tracking performance of neural controller is good, settling time required is also less. Neural controller takes very little time to get settled when load disturbance is applied. Hence it can be seen that direct inverse model of neural network can be used for controlling flow rate [91. Palancar Maria C.& others, "ph system based on neural networks.", nd. Engg. Chem. Res, 1998, , Naveen Bhat & others, "Modelling chemical processes systems via neural computation" EEE control system magazine Marzuki Khalid and Sigem Omatu," A Neural controller for a temperature controlled system " EEE Control System, Bhupinder S. Dayal & others, "The design of experiments,training and implementation of nonlinear controllers based on neural networks" The Canadian journal of chemical Engg., vo1.27, Dec.94, pp1066. V. References Khalid Marzuki,Sigeru Omatu and Rubiyah Yusof," Temperature regulation with neural network and alternative control Scheme " - EEE transactions on neural networks, Vo1.6, 1995,pp D.Tsaltis,A.Sideris and A.Yamamura,"A multilayered neural network controller",eee control system magzine,vol. 10,pp 44-48,April 89. Psciogios D. C. & Unger L.H., "Direct & indirect model based control using ANN" nd, Engg. Chem. Rec Ljung L.," system identification theory for user ",PH Englewood Cliffs N.J Heng Ming Tai and others, "A neural network based tracking control system." EEE trans. On ndustrial electronics, ~01.30, Dec. 1992, 210

4 Fig(2) Hardware of the set up (7 START 7 ENTER SET L= a CALL ADC BPNNDAC Fig(3) Photograph of Add-on card - CALLOGPR Pressed? No $ > CALL OBDR.. Fig. (4) Photograph of signal conditioning card Fig(5): Flow chart of Main Program 21 1

5 21 2

6 0 0 f c\1 i 480. j P60.

Embedded based Automation System for Industrial Process Parameters

Embedded based Automation System for Industrial Process Parameters Godhini Prathyusha 1 Lecturer, Department of Physics (P.G), Govt.Degree College, Anantapur, Andhra Pradesh, India 1 ABSTRACT: Automation