Model-based diagnosis directed by heuristic search

The diagnostic methodology presented is motivated by an application related to offline diagnosis of a rocket engine. The problem of diagnosing devices offline with a fixed set of measurements is addressed. A diagnostic algorithm is presented which combines the advantages of model-based techniques based on constraint suspension and of search directed by heuristic expert knowledge. The algorithm enumerates all faults and fault combinations which are possible given a fixed set of measurement data in order of their likelihood. Expert knowledge is incorporated into a qualitative model-based framework in order to compensate for the lack of sufficient data. The resulting diagnoses are therefore uncertain but represent a best guess consistent with all available data