Comparison of fuzzy and neural network adaptive methods for the position control of a pneumatic system

This paper reports on a study whose objective is to explore the potential and compare the performance of intelligent adaptive control methods. Specifically, the performance of PID plus an adaptive neural network compensator (ANNC) is compared with the performance of a fuzzy adaptive PID controller. The application is position control of a pneumatic gantry robot. Both controllers were carefully tuned to provide a fair comparison. Experimental results were collected for the tracking of a sine wave. Both adaptive controllers were found to improve tracking performance over fixed gain PID by upwards of 70%. However, the tuning procedure for the fuzzy controller is judged to be more intuitive in nature and hence, more practical than that for ANNC. The fuzzy adaptive controller uses a novel rule set that is reduced in size from that used in previous studies.

1. Introduction

This paper compares the performance of a PID controller with an adaptive neural network compensator and fuzzy adaptive PID controller, as benchmarked against a fixed gain PID. Pneumatic servosystems are attractive due to their lower installed cost, relative to hydraulic and electric servosystems. But the problem with pneumatics is their accuracy is low with conventional control. A large body of research is devoted to the application of advanced control techniques to servo pneumatics in order to improve their performance [1]. So-called “intelligent” algorithms such as neural networks and fuzzy rule based controllers are attractive as they don't require a system model. Adding an adaptive feature can improve performance still further, where the controller gains are constantly updated to account for changes in operating conditions and system parameters.