CI-based design of a satellite attitude controller

Abstract

This work presents the embedding of Computational Intelligence (CI) mechanisms in well-known techniques available for control system design, applied to satellite attitude control by using a reaction wheel. It is shown that effective search and scoring procedures can replace human-performed trial-and-improvement actions for gain computation of four types of controllers (including linear-quadratic and H2), and produces performance indexes and torque levels compatible with real world specifications. The CI mechanisms are: (i) a genetic algorithm, which generates, combines and selects controllers candidates, and (ii) a fuzzy system, for scoring performance indexes and torque levels of the controller candidates, which are subsequently used by the genetic algorithm. Not only the controller gains their selves are used as candidates, but also weighting matrices (for the linear quadratic technique) and weighting signals (for the H2 technique). Finally, a comparison of the controllers found by each technique is provided, according the hypothetical example FireSat.