Research Of Fault Diagnosis Method Combined By SVM And FCM

For fault diagnosis of complex system, intelligent information processing technology is current hot spot and inevitable tendency to development of fault diagnosis. The methods of fault diagnosis combined by FCM and SVM are studied and analysed in this paper.This paper mainly studies and discusses on the following aspects:First, fuzzy clusting analysis is one of important method for fault diagnosis.The dissertation studies the whidely applicable fuzzy c-means clustering algorithm.Based on being limitation of"transfer deviation"about transitive closure method and initial sensitivity about fuzzy c-means algorithm, an mixed fuzzy clustering analysis method has been proposed.This method combines the transitive closure method based on fuzzy equivalent relation with the fuzzy c-means algorithm based on fuzzy partition. The experimental results indicate that the mixed fuzzy clustering method can not only diagnose single-faults, but also diagnose multiple-faults effectively. At the same time,it can overcome the missed diagnosis produced by the transitive closure method .Second, as for the fuzziness in fault diagnosis, we process it adopting fuzzy theory is rational extremely. Hence, fuzzy support vector machine(FSVM) is deeply analysed and studied in this paper. A fuzzy membership grades is introduced into SVM adopting fuzzy c-means algorithm,thus the SVM becomes the FSVM, and two different optimal hyperplane are obtained. The FSVM algorithms is used to simulate experiment about the faulty diagnosis methods of turbogenerator.Third, in order to avoiding the refusal classification produced by"one-against-one"and"one-against-all", we analysed and discussed the multi-class algorithm based on hierarchical clustering and decision tree, And concretely studies multilevel binary tree classifier algorithm(FSMBTC) based on FSVM. The FSMBTC algorithm contains two process about crude classification and fine classification. Namely, it first realizes the crude classification adopting FCM algorithm, then achieves the fine classification adopting SVM algorithm. Thoery proved that the FSMBTC algorithm can not only speed up the calculation velocity, but also there is no the nonseparable regions. In order to verify practicality of the algorithm, it is used in the fault diagnosis of rolling bearing, and experimental results are efficient for classification .