Characteristic Analysis Of Fault Features Of Rolling Bearings

Rolling bearings are widely used in large-scale machines,especially rotary machines,due to their high running accuracy,low price,etc.The running state of the rolling bearing is related to the safety and reliability of the entire mechanical system.Therefore,it is necessary to study the fault diagnosis and condition monitoring of rolling bearings.At present,many scholars have proposed a variety of feature extraction methods.However,not all features are valid,or the features that are effective for fault pattern recognition are not suitable for evaluating the degree of bearing wear.The effectiveness of selected features is directly related to the accuracy of fault identification and degree assessment.This dissertation takes the rolling bearing as the research object,applies modern signal processing technology,extracts the fault features of the vibration signal of the rolling bearing,and gives the evaluation index of the fault feature in the fault identification and the evaluation of the effectiveness.The research content of this dissertation is as follows:(1)Based on the characteristics of the rolling bearing vibration signal,the feature extraction method of the rolling bearing fault diagnosis field has been widely used to extract fault features from the vibration signal from multiple feature domains such as time domain,frequency domain,time-frequency domain,and entropy feature.Compared with the traditional fault diagnosis of rolling bearings from a single domain,fault signatures can be extracted from multiple domains,and the differences and complementarities between the features can be fully utilized to more accurately and comprehensively reflect bearing fault information.(2)Analysis of characteristics of fault identification features.The distance evaluation algorithm was used to evaluate the sensitivity of the feature to the fault.Based on the CWRU experimental data,the sensitivity of the extracted candidate features are analyzed,and the features that are more sensitive to the faults,such as the frequency-averaged statistical feature frequency,center-of-gravity frequency,and multi-scale sample entropy were selected,and the sensitive features were analyzed.(3)For the currently extracted fault features,it is difficult to accurately identify the fault through a single feature.This thesis proposes a method based on Euclidean distance-based Maximum Class Separability(MCS)to select the fault feature vector.This method can effectively select feature vectors with good intra-class aggregation degree and inter-class separation degree in feature space to achieve better recognition accuracy.Different eigenvectors are selected for different operating conditions.Selecting a feature through statistical analysis can not only effectively identify faults,but also have good generalization characteristics for different operating conditions.The validity of the MCS method and the generalization of the selected features were verified by MFPT experimental data.(4)Characterize of the state assessment feature.This thesis proposes to evaluate the effectiveness of fault features from two aspects of differentiation and trend.Taking the CWRU inner ring fault data as an example,the fault features under different conditions are evaluated and analyzed for the trend degree of the fault and the discriminability.The evaluation results show that the RMS and standard deviations,average frequency ? frequency domain characteristics,and wavelet packets The percentage of node energy has good differentiation and trend on the degree of failure.Taking into account the actual engineering,the bearing fault degree change process is more complicated,using the IMS whole-life bearing data,the whole-life data of the outer ring fault as an example,to evaluate the distinguishability and trend of the fault features,and to analysis of the trend of fault features.