Representation and Use of Knowledge in Automatic Fault Diagnosis

The report relates activities performed within the ongoing project on Integrated Diagnosis Systems (IDS). A unifying aspect of the activities is representation of knowledge applied in diagnosis. New ways of representing such knowledge can improve the diagnoses, enable reuse, and facilitate consistent integration with other operator support systems. The tasks of the diagnostic process, and the roles of domain knowledge, are discussed in relation to different methods of diagnosis. Two primary methods of diagnosis are recognised in the report, model-based and association-based. Distinct differences of these two methods are identified as focus for integration. A methodology for specifying the design of diagnosis systems is reviewed. This methodology seems to provide a good theoretical basis for understanding problems of fault diagnosis. Qualitative and functional modelling methods are studied by application to a common example domain. The two specific techniques are found to be promising in relation to diagnosis. A software setup for simulated diagnosis is presented. This setup is to be used in the activity on knowledge representation, where a blackboard system is the central module of the setup. Presentations of process domain knowledge show the capabilities of the blackboard architecture and suggest schemes for integrated use of the information. The object-oriented architecture is also shown to serve the needs for presentation of diagnostic reasoning, which is a vital aspect when integrating different diagnosis methods. (author)