Research On Fault Diagnosis Method Of Rolling Bearing Based On EEMD

Rolling bearing is one of the most vulnerable components in the gearbox of train vehicles.The effective extraction of bearing fault feature and the pattern recognition of different states are the keys to bearing fault diagnosis.The fault signals of rolling bearing are non-linear and non-stationary.It is difficult to separate the fault feature because of the noise interference.Based on this,in this thesis,the rolling bearing is taken as the research object,the signal processing method of ensemble empirical modal decomposition and singular value decomposition are combined to extract fault feature.The support vector machine is used in fault state recognition of rolling bearing.The main contents are as follows:(1)Briefly discuss the rolling bearing fault mechanism and fault signal analysis method.In view of the fault feature extraction and state recognition methods used in the rolling bearing fault diagnosis,and the related signal noise reduction methods,the current research status at home and abroad is discussed in detail.Based on the analysis of the failure mechanism of the rolling bearing,the failure form and the calculation of the fault characteristic frequency are analyzed.Singular value decomposition(SVD)can effectively reduce the influence of background noise on signal feature extraction.Reasonable selection of SVD noise reduction rank order is the key.The influence of selecting singular value rank order on noise reduction effect under different criteria is discussed.(2)The combination of EEMD and the improved SVD noise reduction method are used in rolling bearing fault feature extraction.After the EEMD is decomposed,IMF components with large amount of fault information are selected with different evaluation criteria,and the comprehensive evaluation indexes constructed by EEMD entropy,kurtosis,and mean square error are used to select IMF components,which are verified by actual bearing fault signals.Firstly,the improved SVD is used to reduce the noise of the original signal.Secondly,the EEMD is used to obtain several IMF components,and the IMF component with a large amount of fault information is selected according to the comprehensive index.Finally,the enveloped demodulation analysis is performed on the selected IMF.The characteristic frequency of the bearing fault is extracted.The method is verified using simulation signals and test signals.(3)Support vector machine(SVM)is introduced to recognize the different fault states of rolling bearings.The basic principle of SVM is introduced,and the SVM model optimized by genetic algorithm(GA)is established.Considering that EEMD can extract the characteristic frequency of bearing fault,it is combined with GA-SVM model to construct the method of rolling bearing fault diagnosis based on Hilbert-Huang transformation and GA-SVM.Firstly,the correlation coefficient is used to screen the decomposed IMF component after EEMD,and the Hilbert marginal spectral energy and Lempel-Ziv complexity of IMF component are calculated to form the bearing high-dimensional feature vector.Secondly,Laplace score(LS)is used to reduce the dimensionality of high-dimensional eigenvectors.Finally,GA-SVM is used to identify different fault states of bearings.The method is verified by the test data of bearing in different states from Case Western Reserve University.