A Model-Based Diagnosis Framework for Distributed Systems

Abstract : We present a distributed model-based diagnostics architecture for embedded diagnostics. We extend the traditional model-based definition of diagnosis to a distributed diagnosis definition, in which we have a collection of distributed components whose interconnectivity is described by a directed graph. Assuming that each component can compute a local minimal diagnosis based only on sensors internal to that component and knowledge only of its own system description, we describe an algorithm that guarantees a globally sound, complete and minimal diagnosis for the complete system. By compiling diagnoses for groups of components based on the interconnectivity graph, the algorithm efficiently synthesizes the local diagnoses computed in distributed components into a globally-sound system diagnosis using a graph-based message-passing approach.