An abductive reasoner for diagnosis of continuous processes

Summary form only given. The author describes the approach taken to diagnosis by a prototype system that has had some success in discovering faults in continuous processes. The prototype has been tested on the feedwater system of the Bruce B nuclear power plant in the context of the APACS project, a research project headed by Ontario Hydro. The APACS diagnosis prototype is an abductive system. The system attempts to discover one or more faults that entail the observations in the context of a set of causal relationships. The system is novel in its representation of causal knowledge using event classes and in its ability to handle some of the real time issues. The prototype system contains a model of the Bruce B feedwater system and makes diagnoses using data generated by a high-fidelity simulation of the plant. The prototype system has successfully diagnosed instances of several classes of faults generated by the simulation. A diagnosis is displayed to the operator as a graph showing the causal relationships between events