Rule reduction of a neuro-fuzzy PI controller with real-time implementation on a speed control process

Abstract

This study is an attempt for neuro-fuzzy implementation of a prior-designed fuzzy PI controller (FPIC) with reduced number of rules but without sacrificing the controller performance up to a certain extent. To accomplish the goal, backpropagation-based learning algorithm is used to model a connectionist fuzzy controller based on an input–output data set. The resultant fuzzy controllers with reduced rule sets are much faster in operation and cheaper due to lesser memory space requirement. Effectiveness of the designed fuzzy controllers is studied through simulation as well as real-time experimentation on a servo speed control application. Both the simulation and experimental results substantiate the efficacy of the designed neuro-fuzzy controllers with lesser number of rules for approximating the behaviour of a nonlinear fuzzy controller with considerably larger rule base.