Conflict-based diagnosis: Adding uncertainty to model-based diagnosis

Consistency-based diagnosis concerns using a model of the structure and behaviour of a system in order to analyse whether or not the system is malfunctioning. A well-known limitation of consistency-based diagnosis is that it is unable to cope with uncertainty. Uncertainty reasoning is nowadays done using Bayesian networks. In this field, a conflict measure has been introduced to detect conflicts between a given probability distribution and associated data. In this paper, we use a probabilistic theory to represent logical diagnostic systems and show that in this theory we are able to determine consistent and inconsistent states as traditionally done in consistency-based diagnosis. Furthermore, we analyse how the conflict measure in this theory offers a way to favour particular diagnoses above others. This enables us to add uncertainty reasoning to consistency-based diagnosis in a seamless fashion.