Modeling Cyber–Physical Systems for Fault Diagnosis

Existing algorithms for consistency-based fault diagnosis are sound and complete according to some correct logical model. But obtaining a good model is the crucial and difficult part. Originally, the classical diagnosis algorithms were developed to analyze Boolean circuits, such that simple propositional or predicate logic models were sufficient to express the structure and behavior. Cyber–physical systems, however, exhibit hybrid behavior, meaning they generate continuous and discrete values that need to be interpreted. This adds significant complexity. Furthermore, modern cyber–physical systems may change their structure over their lifetime and thus require model adaptations. This article presents a novel formalism to model cyber–physical systems for consistency-based fault diagnosis. Drawing from the research fields of artificial intelligence and control theory, the approach models system structure and behavior through the use of satisfiability modulo nonlinear arithmetic. The approach proves advantageous compared to previous modeling techniques through its integration of nonlinear behavior models, implicit computation of residual values to compute fault symptoms, its representation of different operating modes of the system, and its integration with existing sound and complete fault diagnosis algorithms. The approach was validated empirically using two benchmarks from the process industry, a simulation of battery packs, and Boolean standard circuits. Throughout all experiments, an accuracy of 97% was achieved.