New Techniques in Model-Based Diagnosis

For about a decade, model-based reasoning has been investigated by AI research. In particular diagnosis of technical devices requires explicitly representing principal knowledge about the structure and behavior of the device to be diagnosed and its components. The General Diagnostic Engine (GDE) introduced by J. de Kleer and B. Williams is probably the most promising approach to model-based diagnosis. This paper discusses problems emerging from case studies based on GDE carried out by the Advanced Reasoning Methods group at SIEMENS. In particular, questions of handling devices with many components, reasoning about device behavior over time, exploiting multiple tests and knowledge about faults, and dealing with uncertain measurements are raised. Based on extensions to GDE, at least partial solutions to these problems have been achieved, making real applications of this approach become more feasible.