Efficient Generation of Plausible Diagnoses in a Model-Based Diagnostic Engine

An important way in order to reduce the complexity of model-based diagnosis is to focus the attention on a subset of ”plausible” diagnoses. Two important methods of focusing on plausible diagnoses have been reported: (i) search for the most probable diagnoses (de Kleer); and (ii) search for the preferred diagnoses (Dressler and Struss). The first method has the nice property that it allows to focus only on a small number of candidates. The second one allows to structure the candidate space such that the discovery of a conflict prunes a whole subspace of candidates. We propose a focusing method which combines the advantages of the above mentioned approaches, i.e. we search for the most probable preferred diagnoses. Our algorithm relies more on the explicit construction of the focus diagnoses rather than on a merely sequential test of all candidates. The effect of using a small focus combined with pruning the candidate space provides dramatic reductions of the time required for diagnosis. Content areas: diagnosis.