Optimal communication distributed Petri net based diagnosers of Discrete event systems

This paper deals with distributed model-based diagnosers of partially observable Discrete event systems (DES). The problem of partitioning an interpreted Petri net (IPN) diagnoser model into m modules is addressed; in such a manner that minimal communication among modules is guaranteed. Two main contributions cope with this problem. First the communication-dependence fault graph (CDFG) is introduced; it represents both the system faults (edges) and the measurable places associated to each fault (vertices). In the graph, two and more faults incident to a vertex v_k, means that the detection and location of these faults is based on the partial output information represented by v_k. The second contribution is a polynomial algorithm that exploits the CDFG to compute m diagnoser models in such a way that the number of exchanged messages is minimal.