Study of Learning Fuzzy Controllers

Abstract

This paper compares two types of learning fuzzy controllers, the self‐organizing fuzzy (SOF) controller and the hybrid self‐organizing fuzzy proportional–integral–derivative (SOF‐PID) controller. The SOF is an extension of the rule‐based fuzzy controller, with additional rule creation and rule modification mechanisms. The hybrid SOF‐PID comprises the SOF as a learning supervisory controller readjusting the proportional gain of the PID controller at the actuator section, when the system is on line. The structures of the SOF controller and the hybrid SOF‐PID controller are studied. The performances of the SOF controller and the hybrid SOF‐PID controller are compared by applying them to a two‐link non‐linear revolute‐joint robot arm. For the path tracking experiments, the hybrid SOF‐PID controller followed the required path more closely and smoothly than the SOF controller. The results of the experiments for the SOF controller and the hybrid SOF‐PID controller are also compared with those obtained with a conventional PID controller, using the same values supplied at the setpoint.