Behavior Graphs for Hybrid Systems Monitoring

Abstract

Hybrid Dynamical Systems (HDS) constitute a wide class of common industrial applications, where the behavior is determined by the interaction between continuous and discrete dynamics, i.e. behavioral modes succession. The general principle of model-based Fault Detection and Isolation (FDI) algorithms is to compare the expected behavior of the system, given by a model, with its actual behavior, known through on-line observations. Faults in HDS may corrupt the two dynamics. In that paper, we propose to limit the set of possible mode candidates by using a priori information on the discrete evolution under normal and faulty hypothesis. Two kinds of graphs are derived from the initial hybrid model, namely Normal Behavior Graphs (NBG), Faulty Behavior Graphs (FBG). Using these graphs allows us not only to identify efficiently the actual mode but also to directly interpret (diagnose) the discrete faulty evolution in terms of faults. The whole FDI methodology is described and applied to a two tanks system example.