Component Group Diagnosability via Directional Resolution

This paper addresses simplification of diagnostic models of systems that can be described in propositional logic. Performing diagnostics on the entire model of a system when only a few variables are expected to be observed, is not efficient. If we know the limited set of variables which might appear in the observation, then we can simplify the diagnostic model before the diagnosis inference takes place. An extended model pruning procedure was proposed which systematically removes parts of a model that do not contribute to the overall system diagnosis. It employs an algorithm deciding component group diagnosability based on directional resolution. The paper analyses behavior of group diagnosability for different component groupings in a model. A set of general rules capturing the diagnosability changes for growing groups is derived. The pruning procedure is modified on the basis of these rules.