Fault Diagnosis Of Rolling Bearing Based On Wavelet Analysis And Variational Mode Decomposition

As an important component of rotating machinery,the performance of rolling bearing greatly affects the stability and safety of rotating system.Therefore,the fault diagnosis of rolling bearing has always been the focus of traction drive system.In this thesis,through the analysis of vibration signals generated in the running process of rolling bearing,the current health status of bearing is identified and diagnosed.The main research contents are as follows:This thesis briefly expounds the research significance and purpose of rolling bearing fault diagnosis,and introduces the origin,the development process and research vein of rolling bearing fault diagnosis.This thesis introduces the de-noising principle and development process based on wavelet analysis.The traditional wavelet threshold de-noising,improved wavelet threshold de-noising,wavelet spatial correlation de-noising,wavelet modulus maximum de-noising and wavelet segmented threshold de-noising proposed in this thesis are used to de-noising the simulation signal and bearing real vibration signal respectively.The results show that the wavelet segmentation threshold de-noising method proposed in this thesis has the best de-noising effect.After de-noising by the wavelet segmented threshold de-noising method,the signal features are extracted.The basic time-domain and frequency-domain features are introduced.The time-frequency domain analysis methods,such as wavelet packet transform(WPT),empirical mode decomposition(EMD),ensemble empirical mode decomposition(EEMD),local mean decomposition(LMD)and variational mode decomposition(VMD).The principle,advantages and disadvantages of each method are described,and the simulation experiments are carried out for comparative analysis to conclude that VMD has the best effect.Besides,the high dimensional features need dimension reduction.In this thesis,principal components analysis(PCA)and local target space alignment(LTSA)are used to reduce the dimension of features.In view of the problem that VMD method needs to set decomposition parameters in advance,a scheme of VMD decomposition parameters based on Genetic Algorithm-Fruit Fly Optimization Algorithm(GA-FOA)is proposed,which realizes the adaptive decomposition of vibration signals.The VMD method of optimization parameters is compared with the VMD method of empirical parameters.The results of envelope spectrum analysis are used to compare the performance of fault characterization,so as to verify the superiorty of VMD method.Based on the wavelet segmented threshold de-noising method and GA-FOA optimized parameter VMD method proposed in this thesis,the feature matrix obtained.PCA and LTSA are used to identify the fault state and fault depth respectively by combining SVM model and BP neural network model in machine learning.The results show that LTSA+SVM method has the best diagnosis accuracy.Then,the ratio of the experimental training set and the test set is adjusted to study the generalization ability of the SVM classification model.