Pneumatic control valve stiction modeling using artificial neural network

One of the common and widely appearing problem in the process industry is stiction in pneumatic control valves. It introduces oscillations in control loops which in turn results into variations in product quality, instability in control loops and increases the wear and tear of the control valve. In this paper, a novel way of modeling of pneumatic control valve stiction using feed forward neural network is presented. Efficiency of the developed model was evaluated by comparing it with the real valve data and the data generated by Choudury valve model. The proposed artificial neural network (ANN) based valve modeling technique revealed an accuracy of 99.9% in both the cases. Based on the presented detailed investigations it can be concluded that ANN based modeling of pneumatic control valve is an effective technique.

I. Introduction

Flow is the most common process which frequently comes across in any process industries. In the flow control loop, pneumatic control valve acts as the final control element and is used to regulate the flow of any fluid in accordance with the input signal. That's why pneumatic control valve is one of the most important final control element used in the process industries. Therefore, measurement and control of flow variables is one of the most significant applications of process control in industry [1].