PARTICLE SWARM OPTIMIZATION TUNED FUZZY CONTROLLER FOR ROBOT MANIPULATOR

Abstract

The robot arm is highly coupled, non linear and dynamic system which has large number of uncertainties both parametric and dynamic. Hence designing a mathematical model for such system is a complex task. Even if the model is designed it would not be so accurate to control the robot arm using conventional control technique. Knowing certain parameters of system, one can design a fuzzy logic controller which is a more robust technique than proportional derivative (PD) controller and use human expert's knowledge to design it's rule base. In absence of expert's knowledge one can design fuzzy logic controller (FLC) by his observation or using trial and error method. Whenever a fuzzy logic controller is designed using trial and error method, one cannot be sure about optimality of controller, as one cannot try all possible solution lying in particular range. In order to make FLC optimum, it is required to use some optimization method so that best suitable parameters for FLC can be found out. Whenever optimization techniques are used, first requirement is to define an objective function. As, objective function can be evaluated only by observing the output of a system, objective function cannot be defmed in terms of parameter to be tuned. Therefore direct search method or analytical method cannot be used. The only choice left is random search technique which starts from random initial position lying in particular range and converge towards optimum solution point. One random search technique is particle swarm optimization (PSO) which has fast convergence rate and much simple in its working. In this thesis, a mathematical model for PUMA 560 is designed in simulink. An objective function is defmed and it can be obtained by integrating mean root square error over simulation time. Then PUMA 560 model is controlled using proportional derivative (PD) controller, and fuzzy PD controller for trajectory tracking. Disturbance rejection ability of system is checked by adding a disturbance signal with control signal at the input of robot manipulator. Particle swarm optimization is used to tune error and derivative error gain of PD and fuzzy PD controller. PUMA 560 model is again controlled using PSO tuned PD controller and PSO tuned fuzzy PD controller, and for disturbance rejection ability a disturbance signal is added with control signal. After, simulation it is observed that in case of trajectory tracking there is not so much difference in objective function value, but drastic change is observed in disturbance rejection case. Complete simulink diagram is given in appendix C and simulation results are presented in chapter 5