Attitude fault-tolerant control applied to microsatellite

Abstract

Spacecraft play a pivotal role in various aspects of our daily lives, including mapping, disaster monitoring, and telecommunications. However, akin to any technical apparatus or industrial system, they are susceptible to faults and failures throughout their operational life. Moreover, the repair of damaged components is a rare option in certain missions, such as those involving the Hubble telescope and the International Space Station. To address these challenges, researchers have embraced two key paradigms: fault diagnosis and fault-tolerant control. Both rely on the availability of a physical model that comprehensively captures system dynamics. Our research in this thesis revolves around two primary axes. Firstly, we employ a hybrid approach, combin- ing a model-based method (Kalman filter) with a data-driven method to enhance gyro fault assessment. This is followed by the reconfiguration of satellite attitude control parameters. Secondly, we delve into the application of reinforcement learn- ing, a cutting-edge artificial intelligence method, to optimize attitude fault-tolerant control. It is noteworthy that our scientific contribution, particularly in the second as- pect, focuses on refining the reward function by incorporating the similarity be- tween the torques generated by the reinforcement learning agent and the conven- tional control system. Simulation results underscore the efficacy of the proposed methods in this thesis, substantiated through comparisons with the latest advance- ments documented in scientific publications.