Analysis of Model-Based Diagnosis Using Problem-Reformulation.

摘要

Model-based diagnosis (MBD) is a method to determine the malfunctioning components in a technical system. It requires a model of the system and uses this model and observations on the system to derive diagnostic conclusions. Among the research items in MBD are such diverse things as: fault models, hierarchy, test generation, probabilistic algorithms, abstraction, probe selection, computer algebra, multiple models and many more. One thing which has been lacking, however, is an integrated theory which unites all these research items in a coherent framework. Having a single, integrated framework has the following advantages: it facilitates the understanding of problems and solutions in MBD; it allows the definition of a uniform terminology; and it makes comparison of different approaches possible and is therefore a necessary first step towards finding the best approach. In this report, such a framework, based on reformulations of the diagnostic problem is presented. It is shown how the majority of diagnostic actions can be described as reformulations of the diagnostic problem. A classification of reformulations is given which allows reusing diagnostic conclusions. We then formally describe and classify diagnostic actions according to this classification.