Focusing on Least Cardinality Diagnoses

The main problem with Model-Based Diagnosis is its computational complexity. Each of its fundamental steps, prediction, conflict recognition and candidate generation, exhibits a combinatorial explosion for large devices. The algorithm we present generates diagnoses in an incremental way, starting from all those with the same least cardinality. Provided the reliability of the system’s components, the probability of a diagnosis should decrease with its cardinality. Anytime we’d stop the algorithm we’d had already generated some of the more probable diagnoses for the faulty system. Generating least cardinality diagnoses is much less expensive than generating all the minimal diagnoses, even because we can adopt very efficient Operational Research techniques. Moreover, we show that least cardinality diagnoses are the only ones we have to know for determining the best point in which to make a further measurement in order to reduce the conflict sets number.