Rolling Bearing Fault Monitoring And Diagnostic Method Based On KICA And LSSVM

As an important rotating mechanical equipment, rolling bearing’s service life and fault time is uncertain because of the influence of its work environment and intensities. Therefore, condition monitoring and timely fault diagnosis of rolling bearing has a great value. In this paper, the research object is the rolling bearing. Research is mainly focused on the local mean decomposition’s (LMD) application in vibration signal’s feature extraction field, the superiority of reflecting the fault feature information by mixed domain feature set, the application of Kernel Independent Component Analysis (KICA) for rolling bearing’s condition monitoring and feature extraction, Least Squares Support Vector Machine (LSSVM) classification method for fault classification.First of all, for the lack of traditional fault diagnosis method, using the signal characteristic parameter method to calculate the time domain and frequency domain features of the signal, and then combined with time-frequency domain features which is obtained by LMD decomposition method to construct the mixed domain feature set. The purpose of this is to fully reflect the fault feature information and prepare for further condition monitoring and fault classification.Secondly, extracting the time domain features to construct the feature set, and then put it into KICA for further analysis, the results show that KICA can effectively monitor the running state of rolling bearing and detect the fault timely. And then comparing with independent component analysis method (ICA), the results indicate that KICA has better performance than ICA in state monitoring field.Finally, for the information redundancy and high dimension defect of mixed domain feature set, using the KICA method to reduce the noise and dimension of fault feature set, and then input the feature set into LSSVM for classification, the results show that by constructing mixed domain feature set can improve classification accuracy, comparing with the classification result of the support vector machine method (SVM), verified the superiority of LSSVM.The simulation results show that the method: constructing the mixed domain feature set— KICA—LSSVM for rolling bearing fault monitoring and diagnosis which proposed in this paper is effective, it does a great value for further improving the accuracy of mechanical equipment fault monitoring and diagnosis.