ABSTRACT
Design of an efficient fuzzy logic controller involves the optimization of parameters of fuzzy sets and proper choice of rule base. There are several techniques reported in recent literature that use neural network architecture and genetic algorithms to learn and optimize a fuzzy logic controller. This paper develops methodologies to learn and optimize fuzzy logic controller parameters based on neural network and genetic algorithm. The strategies developed have been applied to control an inverted pendulum and results have been compared for three different fuzzy logic controllers developed with the help of iterative learning from operator experience, genetic algorithm and neural network. The results show that Genetic-
Fuzzy and Neuro-Fuzzy approaches were able to learn rule base and identify membership function parameters accurately.
INTRODUCTION
Control of complex and non linear systems is an important and challenging task and various strategies have appeared in recent literature to deal with nonlinearity and strong coupling of dynamic systems. PID is a popular control method extensively used in an industrial set up. The advantages of a PID controller
include its simple structure alongwith robust performance in a wide range of operating conditions. A lot of research has been done on PID control scheme (see for example, references [1, 2]) and the available methods for tuning PID gains are advanced and accurate. This makes the PID control as one of the most favored control strategies. However, the design of a PID controller is generally based on the assumption of exact knowledge about the system. This assumption is often not valid since the development of model of any practical system may not include precise information of factors such as friction, backlash, unmodeled dynamics and uncertainty arising from any of the sources.
In recent years, there has been an increasing interest in the utilization of unconventional control strategies such as neural networks (NN), fuzzy logic, and genetic algorithm (GA) etc. These control methods derive their advantages from the fact that they do not use any mathematical model of the system. Instead they use input-output relations (Neural Network) or heuristic knowledge (Fuzzy Logic) about the system. This paper investigates the use of fuzzy logic to control a single link manipulator robot. Performance of fuzzy controllers derived from three different methods has been compared in this paper. The first (fuzzy) controller has been designed based on the operator experience and trial and error iteration. The second controller has been optimized with the help of GA, while the third controller has its parameters obtained with the help of NN.
Manish Kumar Devendra P. Garg
Research Assistant Professor and ASME Life Fellow
E mail: manish@duke.edu E-mail: dpgarg@duke.edu
Dept. of Mechanical Engineering and Materials Science
Duke University, Box 90300
DURHAM, NC 27708-0300
No comments:
Post a Comment