Intelligent Diagnosis Of Rolling Bearings With Class Imbalance Based On Data Augmentation And Feature Enhancement

Rolling bearing is one of the crucial components in the machinery.Due to long-term work under high load and high speed,the rolling beairng will be damaged due to wear and fatigue,and its damage may lead to the inability of the transmission system to operate,so that the whole machinery and equipment cannot work.Therefore,timely fault diagnosis of rolling bearing can ensure the safe operation of machinery and equipment.In engineering practice,rolling bearing is in normal operation for a long time and should be stopped immediately for replacement when a fault occurs,so the normal samples that can be collected will be much more than the fault samples,leading to an imbalance between the various types of bearing health status data,i.e.,the class imbalance problem.The diagnostic model trained by the class imbalance data set is prone to overfitting of the fault samples,which seriously affects the performance of the traditional intelligent diagnostic model.Therefore,it is important to study the accurate identification of failure modes in the case of class imbalance of rolling bearings.This paper takes rolling bearing as the research object,proposes an intelligent diagnosis method of rolling bearing with class imbalance based on data augmentation and feature enhancement,and conducts theoretical research and application verification for related problems.The main research contents are as follows:(1)An intelligent diagnosis method of rolling bearing with class imbalance based on data augmentation is proposed.The data augmentation method based on varying-parameter time-frequency analysis is proposed.Using the feature that different parameters in timefrequency analysis produce both consistent and diverse time-frequency state features,the number of minority fault samples is expanded by two time-frequency analysis methods,STFT and CWT,respectively,and the diversity of bearing time-frequency state features is improved at the same time,so as to build a high-quality balance data set.The quality of the expanded samples is evaluated from both qualitative and quantitative perspectives..Using the constructed balanced dataset to train the intelligent diagnosis model can effectively improve the generalization ability of the diagnosis model.The proposed method is experimentally validated using rolling bearing datasets measured in a rail vehicle wheelset bearing fault simulation bench.The experimental results prove that the proposed data augmentation method can effectively improve the accuracy of bearing fault diagnosis,and the effect is better than other data augmentation methods,and it shows good generalization for different imbalance ratios and different intelligent diagnosis models.(2)An intelligent diagnosis method of rolling bearing with class imbalance based on feature enhancement is proposed.The feature enhancement method based on the timefrequency attention mechanism is proposed to build a time-frequency attention network by using the attention mechanism in the field of deep learning,which enhances the timefrequency state features in both frequency and time dimensions respectively and suppresses the noise in the time-frequency domain,so as to improve the feature learning effect under class imbalance.The physical interpretability of the proposed method is demonstrated by combining the analysis of bearing fault feature frequencies and signal envelope spectra..The time-frequency attention network module is embedded into the front-end of CNN to build a complete bearing intelligent diagnosis model,which can effectively improve the accuracy of intelligent diagnosis of bearing class imbalance.Experimental validation is performed using the bearing data set collected from the visualized radially loaded rolling bearing fault simulation experiment bench,and the diagnostic performance of the proposed method is significantly improved compared with the intelligent diagnostic model without embedding the time-frequency attention network module.(3)An intelligent diagnosis model of rolling bearing with class imbalance with integrated data augmentation and feature enhancement is constructed.A combination of data augmentation method based on varying-parameter time-frequency analysis and feature enhancement method based on time-frequency attention mechanism is proposed to further enhance the diagnostic performance of the class imbalance diagnostic model by simultaneously expanding samples and suppressing noise.Experimental validation is performed with the above two data sets,and the results demonstrate the effectiveness,superiority and generalization of the proposed method..In this paper,we propose novel data augmentation and feature enhancement methods from the perspectives of sample expansion and noise suppression,respectively,to improve the performance of intelligent diagnosis models for bearings under class imbalance.The proposed method combines signal processing and machine learning methods in the field of fault diagnosis with distinctive features,which provides new ideas for mechanical fault diagnosis and has important practical significance for the engineering application of fault diagnosis of rolling bearing with class imbalance.