A Model-Decomposition Approach to Computing Most-Likely Diagnostics of Dynamic Systems

Accurate and efficient fault identification is a necessary task for system reconfiguration, fault prognostics, and fault adaptive control in complex dynamic systems. However, timely on-line fault identification for large systems can be computationally expensive. In this paper, we show how we can decompose a system model into sub-models, diagnose each sub-model independently by specifying the most-likely probabilistic diagnoses, and then integrate the diagnoses of the sub-models. We use the possible conflicts (PCs) to find the set of minimally redundant subsystems that can be used to decompose the global LYDIA-NG model into minimal sub-models, and then use Bayesian networks to compute a system-level diagnosis. We demonstrate the feasibility of this method by running experiments on a simulated model of a multi-tank system.