Ordered Diagnosis

摘要

We propose to regard a diagnostic system as an ordered logic theory, i.e. a partially ordered set of clauses where smaller rules carry more preference. This view leads to a hierarchy of the form observations < system description < fault model, between the various knowledge sources. It turns out that the semantics for ordered logic programming nicely fits this intuition: if the observations contradict the normal system behavior, then the semantics will provide an explanation from the fault rules. The above model can be refined, without adding additional machinery, to support e.g. problems where there is a clear preference among possible explanations or where the system model itself has a complex structure. Interestingly, these extensions do not increase the complexity of the relevance or necessity decision problems. Finally, the mapping to ordered logic programs also provides a convenient implementation vehicle.