Opportunistic multi-model diagnosis with imperfect model. Information Sciences.

Any aspect of the world can be modeled in different ways, each suitable for a reasoning activity. We present OMISSYS, a system that can diagnose faults in plants whose components are described by different types of models at different levels of abstraction. Moreover, models and data may be affected by uncertainty and imprecision. OMISSYS applies opportunistically different reasoning mechanisms on the available models to find a set of diagnoses for a given symptom. The adoption of different types of models may help to speed-up the reasoning process by focusing on the most promising aspects of the plant. Moreover, the possibility to exploit the knowledge represented by different models may help to cope with the lack of knowledge of a certain type. OMISSYS has been tested both on a power plant, and on a heating plant; for both of them it has efficiently produced diagnoses accepted by the experts.