Injecting Semantics into Diagnosis of Discrete-Event Systems

Abstract

Most modern approaches to diagnosis of discrete- event systems (DESs) are syntax-oriented. DESs are modeled as networks of communicating automata, where each automaton defines both normal and faulty behavior of one component: the syntax of its regular language. Faulty behavior is associated with a subset of transitions (or events) of the automaton. An evolution of the system (conforming to the observation) is classified as faulty when it involves at least one of such faulty transitions (or events). Consequently, the nature of the system behavior (normal or faulty) strictly conforms to the nature of the behavior of its components. This paper claims that syntax- oriented diagnosis suffers from limited expressiveness when applied to complex DESs. Since a complex DES is topologically organized in a hierarchy of subsystems, different (possibly independent) abstraction levels of diagnosis are required. To overcome the limitations of syntax- oriented diagnosis, a new approach is proposed, based on semantics. A set of semantic rules is specified on a semantic domain (the set of subsystems relevant to diagnosis). Each rule defines the faulty behavior of a subsystem, possibly depending on the behavior of other subsystems. The diagnostic output is a set of candidate diagnoses, which account for the faults of every subsystem in the semantic domain.