Kernel Local Fuzzy Clustering Margin Fisher Discriminant Method Faced on Fault Diagnosis

摘要

In order to better identify the fault of rotor system,one new method based on local fuzzy clustering margin fisher discriminant (LFCMFD) was proposed. For each point on manifold, the farthest point in local neighborhood and the nearest point outside local neighborhood usually constituted the local margin. LFCMFD introduced fuzzy cluster analysis algorithm, eliminated the influence of pseudo-margin points, obtained real local margin, compute with-class scatter and between-class scatte, established local magin fisher discriminant function, found optimal fault diagnosis vector,and then identified the fault class of new testing data by this vector. In order to improve the nonlinear analysis ability of LFCMFD, considering kernel mapping idea, training data with supervision information were mapped to kernel space, constructed kernel fisher discriminant function, LFCMFD algorithm based on kernel method (KLFCMFD)was proposed. The experiment showed, KLFCMFD algorithm had best effect in comparison to other manifold learning algorithm to the rotor fault diagnosis,and fully identify fault class when selecting the appropriate parameters.