Research On Fault Diagnosis For Induction Motor Based On EMD-ICA And SVM

During the electrification age, for its good performance, the induction motor hasbecome the main device that provides motive force and drive for automatic productionand daily life. In practice, however, induction motor failure occurs frequently becauseof many factors.Serious faults may cause the damage of equipment and even thesecurity problem. Therefore, it’s very necessary to find the fault in time for theeconomy and society.Based on the virtual instrument, this paper has completed the signal acquisition andprocessing. Then this paper studys the EMD-ICA, multi-source information fusionand KPCA, and they are combined and applied to construct diagnosis‘s feature vector.Finally feature vectors are put into the SVM diagnosis model, the induction motor’sfaults that include bearing fault diagnosis, end cover loose, under voltage operation arerecognized.Firstly, studied the structure, the working principle and characteristics, commonfaults and their characteristic mechanism analysis of the induction motor are studied, toinduction motor fault diagnosis, a lot of monitoring and intelligent diagnosis methodsare contrasted.Secondly, the virtual instrument technology is studied. In view of the motorvibration and noise, temperature and electric current signal, signal acquisition systemis designed. Acquisition system hardware design is mainly based on NI-PXI dataacquisition platform, the appropriate data acquisition cards are selected, and thecorresponding parameters are set. Acquisition system software design is realized byusing the LabVIEW software programming, make the signal acquisition has a goodinterface.Then, signal and data processing technology was studied. For acquisition betweenvibration signal and noise signal aliasing problems, ICA method is applied to deal with;Based on time domain, frequency domain, and the EMD, the fault feature parametersare extracted. According to the principle of multi-source information fusion, faultfeature vector are formed, and KPCA method is applied to filter characteristics. Finally, according to the characteristic vector, fault types are classified. The SVMfault diagnosis models are constructed to classify the feature vector. The resultindicates diagnosis results are preferable. In this paper, the fault diagnosis method isfeasible.