Online Fault Diagnosis of Discrete Event Systems. A Petri Net-Based Approach

This paper is concerned with an online model-based fault diagnosis of discrete event systems. The model of the system is built using the interpreted Petri nets (IPN) formalism. The model includes the normal system states as well as all possible faulty states. Moreover, it assumes the general case when events and states are partially observed. One of the contributions of this work is a bottom-up modeling methodology. It describes the behavior of system elements using the required states variables and assigning a range to each state variable. Then, each state variable is represented by an IPN model, herein named module. Afterwards, using two composition operators over all the modules, a monolithic model for the whole system is derived. It is a very general modeling methodology that avoids tuning phases and the state combinatory found in finite state automata (FSA) approaches. Another contribution is a definition of diagnosability for IPN models built with the above methodology and a structural characterization of this property; polynomial algorithms for checking diagnosability of IPN are proposed, avoiding the reachability analysis of other approaches. The last contribution is a scheme for online diagnosis; it is based on the IPN model of the system and an efficient algorithm to detect and locate the faulty state. Note to Practitioners-The results proposed in this paper allow: 1) building discrete event system models in which faults may arise; 2) testing the diagnosability of the model; and 3) implementing an online diagnoser. The modeling methodology helps to conceive in a natural way the model from the description of the system´s components leading to modules that are easily interconnected. The diagnosability test is stated as a linear programming problem which can be straightforward programmed. Finally, the algorithm for online diagnosis leads to an efficient procedure that monitors the system´s outputs and handles the normal behavior model. This provides an oppo- rtune detection and location of faults occurring within the system