A domain-independent theory for testing fault hypotheses

Advantages of model-based reasoning in diagnosis include efficient detection of faults (manifested as discrepancies between the observed behaviour of a system and its behaviour as predicted by the model) and generation of fault hypotheses which could account for such discrepancies. Intelligent strategies for testing fault hypotheses, however, must often rely on probabilistic reasoning to take the uncertainty inherent in the relationships between faults and their manifestations. Empirical studies in medical diagnosis have shown that human diagnosticians use hypothetico-deductive reasoning, selecting measurements or tests on the basis of their usefulness for confirming one hypothesis or ruling out another. The author describes a probabilistic model of hypothetico-deductive reasoning which includes strategies for confirming the likeliest hypothesis, disconfirming alternative hypotheses, and discriminating competing hypotheses. Based on a corollary of Bayes´ theorem, the model provides a domain-independent theory for testing fault hypotheses within the framework of a differential diagnosis. A Prolog implementation of the model is described