Identification of Discrete Event Systems for Fault Detection Purposes

To increase the availability of automated production systems, it is crucial to avoid unexpected downtimes.Model-based fault detection where the current evolution of the system is compared with its “ideal”evolution represented by a model should enable detecting faulty modes that can lead to a failure.This thesis is concerned with the construction of a fault-free behavioral model of discrete event systems(DES) using an identification technique. The raw material for the identification is a set of production cyclesthat are series of the input/output signals of the system’s controller.Two main problems characterize the identification of DES. First, the explosion of the number of possiblebehaviors makes it impossible to observe all of them in a finite time. Second, the current techniques fortranslating the observed behaviors into a finite state machine introduce exceeding behaviors that have notbeen observed on the system. These behaviors represents the set of non-detectable faults and must thereforebe reduced.To identify the behavior of DES that behave like a spontaneous event generators, we defined a newautomaton type that is autonomous, non-deterministic and has an output function. We also developed apassive, sequential and constructive identification method. In order to state about the quality of the model, abehavioral accuracy criterion as well as two structural complexity criteria have been defined.A parametrized identification algorithm that allows setting a goal for the behavioral accuracy has beendeveloped. We proved that the identified automaton is complete.Experimenting our theoritecal results on the identification of a large industrial DES (336 inputs/outputs)demonstrates the usability of the method as well as its efficiency in terms of computational effort. Key wordsDiscrete event systems, fault detection, identification, finite state machines, behavioral accuracy, structuralcomplexity