Fault diagnosis using timed automata

Abstract

Fault diagnosis consists in observing behaviours of systems, and in detecting online whether an error has occurred or not. In the context of discrete event systems this problem has been well-studied, but much less work has been done in the timed framework. In this paper, we consider the problem of diagnosing faults in behaviours of timed plants. We focus on the problem of synthesizing fault diagnosers which are realizable as deterministic timed automata, with the motivation that such diagnosers would function as efficient online fault detectors. We study two classes of such mechanisms, the class of deterministic timed automata (DTA) and the class of event-recording timed automata (ERA). We show that the problem of synthesizing diagnosers in each of these classes is decidable, provided we are given a bound on the resources available to the diagnoser. We prove that under this assumption diagnosability is 2EXPTIME-complete in the case of DTA's whereas it becomes PSPACE-complete for ERA's. © Springer-Verlag Berlin Heidelberg 2005.