Distributed Fault Diagnosis Using Dependency Modeling without Revealing Subsystem Details

In the past decade, researchers have studied how to model complex systems for diagnostics using multi- signal digraphs and many algorithms have been proposed to perform inference on this graphical model. Multi-signal dependency model can be applied under a single agent paradigm or a cooperative multi-agent paradigm. Under the multi-agent paradigm, we have developed a distributed diagnosis algorithm which considered perfect tests and assembled dependency graphs at a central location (by knowing model of each agent). However, each subsystem dependency model may be constructed by an independent modeler who will be unlikely to reveal any proprietary information through the dependency model. So the preferred mode of model development will result in independent subsystem models that need to interact during operation in a distributed framework while preserving their integrity and proprietary nature. To meet these requirements, it is desirable not to force each agent to reveal its dependency graph structure. In addition, due to improper setup, operator error, electromagnetic interference, or aliasing inherent in the signature analysis, the nature of tests may be unreliable (imperfect). In this paper, we will extend our distributed diagnosis algorithms in three areas: 1). Dependency modeling without revealing subsystem dependencies; 2). Handle unreliable tests; 3). Distributed multiple fault diagnosis.