Model Abstractions for Diagnosing Hybrid Systems ?

Hybrid systems models are a powerful tool for representing systems with both discrete and continuous dynamics. However, computationally these models are challenging to perform most classes of inference on. In this article we focus on diagnosing hybrid systems. Rather than work on the full hybrid model, for which diagnosis is undecidable in the general case, we abstract the model to a propositional model-based diagnosis (MBD) model. We describe how we can translate a reference hybrid systems model into a propositional diagnosis model, which involves translating the model itself, as well as a sequence of observed events, or trace. We provide an illustrative example of the process, and outline how this process guarantees that any diagnoses in the hybrid systems model will be preserved in the MBD model.