Analysis of two probabilistic models of fault diagnosis

Computer aided fault diagnosis has been a subject for research since the early 1960s. The potential applications to medicine, process control, quality assurance, and other fields make it an attractive area of applied research. Many attempts have been made to use statistical information for solving such problems. Two such models, those of Peng and Reggia [Peng 1987a] [Peng 1987b] [Peng 1989] and of Liepins and Potter [Liepins 1990] [Potter 1990], are distinguished by the fact that they both formulate the diagnosis problem as an optimization problem. It is the purpose of this work to comparatively evaluate these models in terms of accuracy and sensitivity. The models are described as originally published, with notations altered for consistency. Results of experiments comparing model accuracy are then presented, as well as experimental results from evaluating model sensitivity. The results from each of these efforts are then summarized in the context of evaluating which model to use under what circumstances. Suggestions for future work are also presented.