An intelligent diagnostic system based on Petri nets

Diagnosis methods and diagnostic systems based thereon have practical significance and strong traditions in the process system engineering literature. There are a number of methodologies (see eg Finch et al., 1990 or Catino and Ungar, 1995) applying different AI techniques for model-based diagnosis, fault, failure or hazard detection and identification. AI methods offer the possibility to use heuristic as well as model-based process information but may run to computational complexity problems with increasing process size. Coloured Petri nets (CPNs) (Jensen and Rosenberg, 1991) belong to the area of discrete event system methodology. They are capable of describing both the dynamic qualitative behaviour of process systems and operating and diagnostic procedures, therefore, CPNs are powerful knowledge representation tools for intelligent diagnostic systems. These methods can be the basis of both diagnostic systems for batch process plants (Srinivasan and Venkatasubramanian, 1996) and control system design (Capkovic, 1995). Majority of the methods available apply fixed and in some sense complete process model to support the model-based diagnosis. The possibility to use measured data to identify at least part of the model during the diagnosis has not yet been fully exploited. The theoretical problems in synthesizing a Petri net model are investigated in Desel and Reisig (1996) while a practical attempt of learning control of a disassembly Petri net is reported in Suzuki et al. (1996). Therefore, the aim of this paper is to propose a methodology for an intelligent Petri net-based diagnostic system which is able to refine its uncomplete process model by learning from measured data.