Parallel Model-Based Diagnosis Using PVM

Extended Abstract The present work concerns the PVM implementation of a parallel approach to model-based diagnosis, using a Petri net model 4] of the behavior of the system to be diagnosed. Solving a diagnostic problem usually involves nding the primary causes that produce a particular malfunction, whose eeects are observed in the behavior of a given system. In the model-based approach to diagnosis, a model of the structure and/or of the behavior of the system to be diagnosed is supposed to be available; the solution to a diagnostic problem is obtained by tracking back the observed misbehavior on the available model, looking for a possible explanation of such a misbehavior (see 3] for a complete survey). In the present work we assume that the we can model the faulty behavior of a system by means of a set of states related through cause-eeect relationships. In such a case a diagnosis can be represented through a set of initial or primary causes such that their presence is able to predict the observed parameters (i.e. the symptoms characterizing the diagnostic problem to be solved). This is a quite common approach in model-based diagnosis (see 2]); the novelty is that we rely on a Petri net representation of such a causal behavior by adopting a particular net model called Behavioral Petri Net (BPN) 1, 6]. The basic advantages of such an approach is that classical reachability analysis techniques of Petri nets can be fruitfully exploited in performing diagnostic problem solving. In particular, thanks to some particular features of the BPN model, reachability graphs analysis can be suitably adopted in the search for the solutions to a diagnostic problem (see 5]). Since diagnosis is a problem of post-diction rather than prediction, backward reachability analysis on the state space of the net is adopted, in order to explain the misbehavior of the modeled system. The observed information about the state of the system under examination (i.e. the symptoms of a particular diagnostic problem) will be modeled by means of a marking from which backward reachability analysis will start. Such an analysis is performed by deen-ing a backward ring rule for each kind of transitions of the BPN formalism, and by applying these rules, starting from the marking corresponding to the observed symptoms, in order to generate the reachability graph in a backward