Generalized Predictive Control for a Pneumatic System Based on an Optimized ARMAX Model with an Artificial Neural Network

Abstract

Pneumatic systems play an important role in applications of robotics, industrial automation, and manufacturing fields. However, accurate control performance on such systems is very difficult to be achieved due to nonlinearity of the system, dead time and parameter variations in the control process. This paper has developed an effective approach to the precise control on a pneumatic system through the combination of an artificial neural network and generalized predictive control (GPC) algorithm. An ARMAX model of the pneumatic system is derived from the weights of a multilayer feed-forward neural network trained with Levenberg-Marquardt method. Nelder-Mead downhill simplex method was applied in this paper to optimize the built ARMAX model, and the better results were obtained through the generalized predictive control for this pneumatic system. The performance of the designed GPC controller is very impressive for the fast response and high accuracy tracking.