A new methodology to diagnose multiple faults in dynamic hybrid systems considering both continuous and discrete behaviors

In this paper, a general multiple faults diagnosis methodology for hybrid dynamic systems is proposed. The considered hybrid systems exhibit both continuous and discrete behaviors without an emphasis of a particular behavior. To diagnose such systems, we rely on bond graph and Petri net modeling. Then, the proposed diagnosis approach deals with the particularity of each system's variables: diagnosing continuous variables using an evaluation of Analytical Redundancy Relations (ARR) and diagnosing discrete ones using a combination of coverability and invariant (T-invariant and P-invariant) notions. The evaluation of residuals (ARR) is based on the combination of adaptive thresholding and fuzzy logic reasoning optimized by Particle Swarm Optimization (PSO). The results of the diagnosis module are finally displayed as a colored causal graph indicating the status of each system variable.