Position Control of a Pneumatic Servoactuator With Time Delay Using a Neural Network Predictor

Abstract

ABASTRACTConsideration is given to data-based dynamic modelling and positioning of a pneumatic servo-actuator with time delay due to friction and air compressibility. A Smith predictor-type control scheme is considered to improve the control behaviour. System non-linearity is verified by the inability of linear modelling to provide satisfactory prediction. Data based dynamic modelling is undertaken using the group method of data handling(GMDH) neural network, a feedforward type neural network, and the techniques show potential for prediction of non-linear behaviour. Hence the time delay effect to the closed-loop control can be removed by incorporating the trained nonlinear model as a predictor in the feedback path. Conventional feedback control with/without the predictor are compared, and it is shown that the time delay system stability is significantly improved by using the non-linear predictor. The effect of time delay is most significant at small stroke changes of fast response demand and experiments were undertaken around the mid-stroke, which is less damped than any other position.