Neuro-fuzzy adaptive control of a revolute stewart platform carrying payloads of unknown inertia

Abstract

SUMMARYIn this research, a Stewart parallel platform with rotary actuators is simulated and a prototype is tested under different operative conditions. The purpose is to make the robot robust against inertia variations considering the fact that different payloads of unknown size may be transported. Due to the complexity issued by expressing the equations of motion with independent variables, the governing equations are derived by Lagrange's method using Lagrange multipliers for imposing the kinematic constraints imposed on this parallel robot. Eliminating Lagrange multipliers by projecting the equations onto the orthogonal complement of the space of constraints, the equations of motion are transformed to a reduced form suitable for the purpose of controller design. The control approach considered here is based on a neuro-fuzzy interference method. As a first step, each revolute arm link are individually trained under different loadings and diverse maneuvers. It is purposed that once employed together, the links will have learned how to collaborate with each others for performing a common task. Training data are divided to several clusters by using a subtractive clustering algorithm. For every cluster, a fuzzy rule is derived so that the output follows the desired trajectory. In the last stage, these rules are employed by utilizing back propagation algorithms and the effectiveness of the neuro-fuzzy system becomes approved by performing multiple maneuvers and its robustness is checked under various inertia loads. The controller has ultimately been implemented on a prototype of the Stewart mechanism in order to analyze the reliability and feasibility of the method.