Qualitative, model-based diagnosis of complex physical devices

We describe the basic paradigm that is commonly used for model-based anomaly detection and isolation. In particular, we are interested in techniques that use an abstract, qualitative representation as opposed to a precise numerical specification to describe the relevant structural and behavioral properties of the system. In general, these techniques describe the system using a structural model defining the qualitative relationships between a finite set of components coupled with a declarative specification of the valid behaviors for each component. The diagnostic engine reasons about the interaction between components to derive a description of the expected system behavior which in turn is used to detect discrepancies by comparing the expected behavior against the observations. Once a discrepancy is detected, the model is used to identify components whose failure could result in the observed behavior. This basic approach can be used when reasoning about either the steady state or the time varying behavior of both discrete and continuous systems.