On-line trained neural controllers for nonlinear hydraulic system

Abstract

This paper studies the problem to control the nonlinear dynamic model of an hydraulic servo system using neural networks. Hydraulic position servo systems are commonly used in industry. This kind of systems is nonlinear in nature and generally difficult to control. The ordinary linear constant gain controller will cause overshoot or even loss of system stability. Artificial neural networks (ANNs) are essentially adaptive systems able to learn how to perform complex tasks, and therefore, neurocontrol techniques are able to overcome many of the difficulties that conventional control techniques suffer when dealing with nonlinear plants or plants with unknown structure. The Direct Adaptive Neurocontrol approach is used in this application where learning is essentially performed on-line using the plant output error. Simulation results show that the proposed approach provides more accurate, robust, and efficient adaptive control for the nonlinear hydraulic system when compared to the traditionally used control methods.