Optimizing the system observability level for diagnosability

Abstract

A system model is diagnosable when every fault can be unambiguously detected from its observable events. Diagnosability is a desirable system property, enabling large and complex systems to be designed with automatic fault detection and isolation mechanisms. In this paper we study the relation between a system's level of observability and its diagnosability. We provide both necessary and sufficient conditions on the observable events maintained by the system in order to be diagnosable. We concentrate on two problems: First, we show how to transform a diagnosable system into another one which is still diagnosable but also has a minimal level of observability. Second, we show how to transform a non-diagnosable system into a diagnosable by subsequently increasing the level of observability. Finally, we expand our framework with several extensions, dealing with distinguishability, predictability and extended fault models. © 2008 Springer-Verlag.