Neural adaptative controller for compound table

Abstract

The present work is about a position neural adaptive controller for a table with two degreed of freedom, composed of two bases. The bases are dislocated in the horizontal plan, and are driven by two DC motors. The basis positions are obtained with two positions sensors. As these systems are used in mechanical systems measurements which need high degree of accuracy, adaptive controllers are indispensable to carry out these tasks. A neural adaptive controller is used for each base and is designed to operate in two steps. In the first step, a three layer neural network with four inputs nodes, eight nodes in the intermediate layer and one output node is trained. The inputs of the first layer are designed by analogy with a PD2 controller. The neural network output is the control variable. In this first step, the Transfer Function of each base is used to obtain the weight values of all layers in the off-line mode. In the second step, the weights obtained are used and also updating in the real time, as function of the base movements, furnishing the control variables. Experimental results are presented.