Research And Implementation Of The Incremental Diagnosis Method Based On Windows Updating In Discrete Event Systems

Given the system model and observations，the target of diagnosis ofDiscrete-Event System (DES) is to find out the faulty events. It is veryimportant to get the diagnosis result correctly and efficiently，because itcan help us realize and locate the abnormal component timely to make sure thesystem is in safe. In DES’s diagnosis，the diagnosis space will increase veryfast when the number of state becomes large. It makes the diagnosis verycomplex and not efficient. Incremental diagnosis is raised to solve thisproblem. The diagnosis process consists of several iterations. In everyiteration, only the observations in the current diagnosis window are concernedto get a partial diagnosis. The diagnosis space is restricted to a small scope,so the diagnosis is more efficient. And it can be used in the situation inwhich observations are not enough. The Non-Exhaustive Diagnosis Engine (NEDE)method is one of the incremental diagnosis methods. It transfers the diagnosisinto a SAT problem. With controlling the size of diagnosis window, it expandsthe paths based on observations and returns a preferred diagnosis. However,NEDE only saves one path in each iteration and abandons all the other paths，which may make some better solution missing. This fact affects the accuracyof NEDE.We analyze the limitation of the current incremental diagnosis in this paper. Then we discuss the use of update in incremental diagnosis based onthe update theory, and improve the current algorithm. We propose an improvedmethod called the Incremental Diagnosis Method Based on Windows Updating. Inevery iteration，it firstly finds a partial diagnosis for the diagnosis window.Then it uses this partial diagnosis to update the old result and get a newone. The goal is to obtain a set of models which are the approximate solutionsof the optimal solution. According to the semantic of update, the keyinformation is kept during diagnosis. We then implement the new algorithm andprove the improvement of accuracy with experiment. There are still someremaining questions in our research. First, the size of diagnosis window iscritical in incremental diagnosis. How to find a sound diagnosis window isstill an unsolved problem. So is the proof of it. Second, how to abstract asystem in real world and describe it with DES is a complex question. It callsfor experience. How to build up such models is also very import in incrementaldiagnosis.