An architecture for time-limited model based diagnosis

The second generation of knowledge based diagnostic systems is characterised by the model-based diagnostic system (MBDS) paradigm. The aim is to find a solution based on an ideal simulation of the real world. If the output of the simulation does not match with the output of the real system, a failure exists and it must be found. However, MBDS are in general slow, and in particular cannot adjust their processing to enable deadlines dictated by external events to be met. The authors´ research is concentrated on a time-limited diagnosis. The consequence has been Tilde (time-limited diagnostic environment). In the Tilde architecture, it is planned to combine the speed of the traditional rule-based systems with the completeness and detail of the MBDS. Furthermore the real system is modelled in several hierarchical levels with different degrees of detail. Consequently Tilde starts by localising the fault at the simplest model, the resulting candidates are then focused upon, and more detailed models are used to refine the diagnosis