Research On Fuzzy Clustering Algorithms Based Fault Diagnosis

Fault diagnosis technology based on fuzzy clustering is an important area of fault diagnosistechnology. It is useful to deal with diagnosis of huge or complex systems and does not need too manypriori knowledge or precise mathematical modes of these systems. This technology can obtaininformation of pattern from a large number of monitoring data of systems. With the complexity ofindustry process and equipment increase, research on fault diagnosis technology based on fuzzyclustering has very important theoretical significance and engineering application value.Fault diagnosis technology based on fuzzy clustering has been a hot area in related research fields,and there have been lots of different clustering algorithms and diagnostic method since thistechnology come into being. But the research on problems of diagnosis on unknown faults is alwaysslightly weak, and there is no mature method and consensus in this area. So this article studiedquestions of how to isolate the unknown fault with known faults and how to isolate faults of differentstrength in same type. A lot of deep theoretical analysis and experimental verification were carried outon both fuzzy clustering algorithm and online diagnosis scheme, and an improved possibilistic GKclustering (IPGK) algorithm and a new online fault diagnosis scheme based on fault vectors wereproposed. Problems in diagnosis of unknown faults were solved with combination of these two newmethods. The main contents of this paper are as follows:1. Review the development of fault diagnosis technology and introduce some common diagnosismethods based on category. These research results, research literatures, theoretical basis andtechnical methods related to fault diagnosis technology based on fuzzy clustering were summed.The relationship between fuzzy clustering, pattern recognition and fault diagnosis was describedand a few key words involved by fault diagnosis technology based on fuzzy clustering wereexplained.2. The development of fuzzy mathematics was simply introduced, as well as the developmenthistory and theoretical background of fuzzy clustering techniques. Theoretical analysis andexperimental verification were carried out on fuzzy c-means clustering (FCM) algorithm whichhas been widely used. Results showed that diagnosis of unknown faults needs that clusteringalgorithm is able to detect data in the distribution of hyperellipsoid or superlinear and is suitablefor the detection of outlier. According to these specific needs, research and improvement onalgorithm were carried out based on characteristics and shortcomings of three mature algorithm: FCM algorithm, possibilistic c-means clustering (PCM) algorithm, improved possibilistic c-means clustering (IPCM) algorithm. The improved possibilistic GK clustering (IPGK) algorithmbased on Mahalanobis distance was proposed and it can deal with data in distribution of ellipsoidand superlinear data and is suitable for the detection of outlier and diagnosis of unknown faults insimulation experiment.3. Faults in different strength of same type are difficult to isolate with using common onlinediagnosis methods, it’s found by theoretical analysis and simulation experiments and will lead towrong diagnosis. One single categorical data clustering algorithm was proposed with simplemodifications on IPGK, so clusters of online data can be calculated now. Calculation method offault strength and principles in distribution of clusters of fault data were found by using concept of directional residual, and the directional residual was translated to fault vectors in the field offault diagnosis based on fuzzy clustering. Then a new online diagnosis method based on faultvectors was proposed and this method was proved to be able to isolate fault in different strengthof same type by simulation experiment.4. A fault diagnosis process based on fuzzy clustering was proposed by combining both IPGK andthe new online diagnosis method. Then simulation experiments on TE benchmark and NSVmodel were carried out. With results of experiments, the new method proposed by this article wasproved to be able to deal with fault diagnosis of unknown faults very well, isolate faults indifferent strength of same type and has certain anti-interference ability.