Efficient diagnosis of hybrid systems using models of the supervisory controller

This paper presents a model-based approach to diagnosis of dynamic systems that combine continuous and discrete behaviors, i.e., hybrid systems. Typically these systems have been abstracted into discrete event models or approximated by continuous models with steep slopes so that existing algorithms can be applied for fault isolation tasks. To overcome the limitations of these approaches, we have developed combined qualitativequantitative schemes for diagnosis that use hybrid models of the system and the model of the supervisory controller. The model of the supervisory controller is used to significantly cut down complexity in tracking behaviors, and generating and refining hypotheses across mode changes. In this paper we present the algorithms for the main tasks involved in hybrid system diagnosis, namely, quantitative tracking with an observer, hypotheses generation by back propagation, hypotheses refinement by forward propagation, progressive monitoring and parameter estimation. Example scenarios are presented to demonstrate the effectiveness of this approach.