Fault Diagnosis And Condition Monitoring Of Rolling Bearing Based On Kullback-Leibler Divergence

As one of the key components of rotating machinery,rolling bearing directly affects the reliability and stability of mechanical operation.How to use a large number of monitoring data to achieve in-depth mining of fault information in order to realize fault diagnosis and condition monitoring is worth in-depth study.Most of the existing methods of rolling bearing fault diagnosis and condition monitoring independently consider the characteristic information of a single sample,and pay less attention to the correlation between the samples in same state.This correlation may contain important fault and status information.If we can make full use of the similarity of statistical characteristics between samples in the same state and the different characteristics of statistical characteristics between samples in different states,it is helpful to improve the accuracy of fault identification.Based on this idea,this thesis takes rolling bearing as the research object to carry out the research on the method of fault diagnosis and condition monitoring of rolling bearing based on Kullback-Leibler divergence.The main work and innovation of the thesis are as follows:(1)Due to Ensemble Empirical Mode Decomposition(EEMD)has the problem of false Intrinsic Mode Function(IMF),Kullback-Leibler divergence and kurtosis coefficient are combined to remove the false components in this thesis so as to extract more effective features for subsequent fault diagnosis and condition monitoring.(2)In order to fully describe the characteristic information of each state of the rolling bearing vibration signal,a method is adopted to extract and fuse the multi-domain features of the signal such as time domain,frequency domain and time domain.Time-frequency feature extraction refers to the feature extraction of time-domain parameters and frequency-domain parameters of the improved EEMD method.Based on this,alternative feature sets of each state of rolling bearing are constructed.In order to select more sensitive features from the alternative feature set,K-L divergence can be used to characterize the similarity between the two distributions,and to modify traditional feature selection method which cannot effectively extract sensitive features with similar feature means but different distributions.(3)A fault diagnosis method of rolling bearing based on K-L divergence is studied in this thesis.Since this method makes full use of the correlation between samples,to a certain extent it is better than the traditional fault diagnosis methods based on Support Vector Machine(SVM)and BP(Back Propagation)neural network in terms of fault identification accuracy.The method is applied to vibration signal of rolling bearing and it is found that the method has better stability of fault diagnosis.(4)The evaluation model of rolling bearing state is established by K-L divergence.The sliding window is used to select the data set to be evaluated and calculate the K-L divergence between it and the normal state data to obtain the state evaluation value and finally form the state evaluation curve by analogy.The effectiveness of the method is verified by the application of vibration signal of the rolling bearing.Figure 36,table 21,73 references.