Abstract AAS 97-662

AAS 97-662

APPLICATIONS OF NEURAL NETWORKS TO THE SATELLITE ATTITUDE CONTROL PROBLEM

G. Riestra and J. Puig-Suari - Arizona State University

Abstract

In this paper an inverse control architecture based on a linear neural network (NN) is applied to the satellite attitude control problem. The adaptive capabilities of the NN allow the controller to handle significant uncertainties in the inertia properties of the spacecraft. The adaptation is based on the LMS rule. In order to control a nonlinear system with a linear NN, the nonlinear angular velocity signals (gyroscopic and square terms) are built externally and then fed to the NN. Under the proper conditions of persistent excitation, the entries of the NN weight matrix provide on-line estimates of the inertia properties of the