Simulation (model) based fault detection and diagnosis of a spacecraft electrical power system

Model-based artificial intelligence approaches to diagnosis require encoding a reasonable facsimile of the problem domain. The model is encoded as a classical engineering simulation of the domain, a spacecraft electrical power system (EPS). Portions of the reasoning system thus become comparators between the expected behavior and the EPS. The diagnostic problem is partitioned into six discrete steps including: fault detection, diagnosis and hypothesis generation, hypothesis space pruning, validation of the hypothesis, test case selection, and verification of the diagnosis. The system performs a simplified form of learning by injecting diagnosed faults into the model thereby maintaining its consistency with the EPS and reducing unnecessary future computations. The authors demonstrate that shallow reasoning systems performing a comparative analysis of the data are sufficient for a complex application