Abstract

In this thesis, we have proposed a novel approach which strengthens existing efficient fault-detection mechanisms with an additional ability to classify different types of faults to effectively determine potential causes of failure in a subsystem. This extra capability ensures a quick and efficient recovery/reconfiguration from disruptions. Our developed diagnosis/analysis procedure exploits a widely used qualitative technique called fault-tree analysis for failure analysis in the Attitude Control Subsystem (ACS) of a spacecraft. The proposed fault-tree synthesis algorithm utilizes machine-learning techniques to classify and rank primitive events in terms of their severity for a particular system failure. The effectiveness of the fault-tree synthesis algorithm presented in this thesis has been demonstrated under different simulated ACS failure scenarios. Constructed fault-trees have been able to represent combinations of events leading to different failures resulting due to artificially injected faults in a Simulink model of ACS. It is important to emphasize that proposed technique has potentials for being integrated in an on-board spacecraft health monitoring and diagnosis tool. (Abstract shortened by UMI.)