Research On Fault Diagnosis Technology Of Asynchronous Motor Based On Multi-source Heterogeneous Information Fusion

As the most widely used power equipment, asynchronous motor consists of manymechanical system and electrical system, which is a typical strong-coupled andnonlinear system. There is a strong nonlinear relation between fault features. Affectedby many uncertain factors, the fault diagnosis process has inherent uncertainty.Therefore, it’s difficult to characterize equipment running status fully to make anaccurate diagnosis by single signal feature. Information fusion provides a new way tosolve the problems in asynchronous motor fault diagnosis. Beginning with theretrospection of research status of asynchronous motor fault diagnosis technology, onthe basis of analyzing asynchronous motor structure performance and common faultmechanism, experimental scheme is devised, and the fault diagnosis method ofasynchronous motor based on heterogeneous information fusion technology isinvestigated from both feature level and decision level in this paper. The maincontents are as follows:1. Research heterogeneous information feature fusion fault diagnosis method ofasynchronous motor based on CMKPCA.(1) Aiming at the requirement of dimension reduction in asynchronous motorfeature fusion fault diagnosis and the insufficiency of KPCA (Kernel PrincipleComponent Analysis), an improved KPCA algorithm-CMKPCA (Class Mean KernelPrinciple Component Analysis) is proposed. Apply KPCA idea to class mean vector ofmapping data, the CMKPCA algorithm model is established by building class meankernel matrix. The eigenvector formed by class mean kernel principle componentabsorb all classified information of class mean vector and its dimension is lower thanthe number of fault category, which realize dimensionality reduction withoutinformation loss based on class mean vector. CMKPCA has the stronger ability ofintegrating original variable information than KPCA, which is an effective featurefusion method.(2) In view of the nonlinear features of asynchronous motor fault and highdimensionality of combined feature, a fault diagnosis model of multi-sourceheterogeneous information feature fusion based on CMKPCA and SVM (SupportVector Machines) is established. CMKPCA is used to carry out heterogeneousinformation feature fusion. SVM is used to identify fusion feature. The experimentwith vibration signal and current signal as heterogeneous source demonstrate the validity of this method.2. Research heterogeneous information decision fusion fault diagnosis method ofasynchronous motor based on weighted evidence theory.(1) Aiming at the problems that need to be solved in the process of using D-Sevidence theory to decision fusion, posterior probability modeling approach based onmulti-classification SVM is introduced, which combined the advantage of SVM indealing with non-linear problem with the advantage of posterior probabilisticmodeling. Meanwhile, a weighted combinational algorithm based on weightedevidence model and matrix analysis is proposed. It improves the reliability of theconflict evidence fusion by using weighted average evidence and reducescomputational complexity by using weighted average evidence fusion formula derivedfrom matrix analysis.(2) Aiming at the uncertainty of asynchronous motor fault diagnosis and highlyconflict evidence fusion problem, a fault diagnosis model of multi-sourceheterogeneous information decision fusion based on weighted evidence theory andSVM is established. The posterior probability modeling method based on SVM is usedto structure basic probability assignment function and the weighted combinationalalgorithm is used to decision fusion. The experiment with vibration signal and currentsignal as heterogeneous source demonstrate that this method can deal with conflictevidence reasonably and improve the accuracy and confidence of fault diagnosis,while it has have strong fault-tolerant ability.