Distributed Diagnostic Problem Solving With Colored Behavioral Petri Nets

Abstract

With the profound diffusion of technological systems in the real world, fault diagnosis has become a major requirement due to its importance in terms of system reliability, security, and efficiency. In this paper, we focus on the problem of causal model-based diagnosis (MBD) of spatially distributed systems in terms of colored Petri nets (CPNs). First, we introduce the colored behavioral Petri net (CBPN) model as a particular CPN intended for the description of a system's causal behavior where each transition is associated with a matrix describing the possible ways it may fire. On the basis of such matrices, we define a particular technique that we call CW-analysis as a backward reachability analysis of CBPNs. With no need to expressions inversion, which is particularly required for original CPNs, the CW-analysis can be realized by a simple manipulation of the net transition matrices. Second, we propose a new approach for distributed systems, where the diagnostic process is captured within a framework based on the formalism of CBPNs. The systems we consider consist of a set of interacting subsystems. Hence, the overall system model is given as a set of place-bordered CBPNs. The interactions between the subsystems are captured by tokens that may pass from one net model to another via bordered places. The diagnostic system is defined as a multiagent system, each agent is in charge of diagnosing an associated subsystem, by having its local CBPN and receiving only the observations generated by its elements. We show how the formalization of the diagnostic process at the level of each agent can be obtained in terms of reachability in a CBPN and can be implemented by exploiting the CW-analysis technique. Once local diagnoses are obtained by the different agents, a cooperation process should be initiated to ensure global consistency of such diagnoses.