Improving Model-Based Diagnosis Through Algebraic Analysis: The Petri Net Challenge

Abstract

The present paper describes the empirical evaluation of a linear algebra approach to model-based diagnosis, in case the behavioral model of the device under examination is described through a Petri net model. In particular, we show that algebraic analysis based on Pinvariants of the net model, can significantly improve the performance of a model-based diagnostic system, while keeping the integrity of a general framework defined from a formal logical theory. A system called Invads is described and experimental results, performed on a car fault domain and involving the comparison of different implementations of P-invariant based diagnosis, are then discussed. Introduction In some recent papers (Portinale 1993), we have shown that Petri nets (PNs) (Murata 1989) can be fruitfully employed to face the problem of model-based diagnosis. This is accomplished by taking into account a formal logical framework of reference, defining classical notions (from the AI point of view) concerning the c...