An intelligent method to estimate the inertia matrix of a robot arm for active force control using on-line neural network training scheme

Abstract

This paper presents a new intelligent controller algorithm comprising an on-line multi-layer artificial neural network (ANN) training scheme to estimate the inertia matrix of the robot arm to enhance the performance of the active force control (AFC) scheme. The robot under study is a planar two-link rigid robot which is subjected to a non-linear disturbance torques acting at the robot joints. The algorithm has two stages, namely the ANN training stage and the implementation stage. During the training stage, the proposed ANN scheme trains the ANN parameters (weights and biases) for a period of time by utilising the back-propagation (BP) learning method. After a sufficient training period, the training session is switched off, and the ANN is reay to be used in the implementation stage of the intelligent AFC-ANN controller scheme. The results of the training and implementation stages are shown and discussed. It is shown that the proposed controller scheme is very effective and robust. The simulation is accomplished using MATLAB(R) software.