Model-based Probabilistic Diagnosis in Large Cyberphysical Systems

Model-based diagnosis is concerned with diagnosing faults or malfunction of real-world physical or cyberphysical systems using a model of the structure and behavior of the systems. As cyberphysical systems can be extremely large and complex, and the associated computational models will be then equally large and complex, they impose a hard to beat challenge on the computational feasibility of reasoning with such models. When such a model is able to handle the uncertainty associated with diagnostics, giving rise to probabilistic model-based diagnostics, the computational feasibility becomes even harder. This paper: (1) proposes a novel graphical method underlying model-based diagnostics; (2) demonstrates experimentally how a novel, by the authors developed architecture of partitioned positive weighted model counting, is able to handle exact inference to answer a variety of probabilistic queries regarding the health status of a cyberphysical system. Results obtained are well within acceptable time bounds.