Study On Rolling Bearing Fault Diagnosis Based On Deep Learning

As a vulnerable part of mechanical equipment,the fault diagnosis and research of rolling bearing has important practical value.Considering the different characteristics of bearing data in different operation stages and different application scenarios,first of all,the original data collected by sensors in the early stage are often insufficient with fault label information,and it is impossible to analyze the fault types of bearings through modeling,secondly,in the process of bearing operation,changes in working equipment and working conditions will lead to differences in the distribution of vibration data.Finally,with the increase of running time,strong background noise and external disturbance will also pollute the quality of bearing data.Therefore,in order to improve the practicability and generalization of bearing fault diagnosis technology,this paper takes three data problems in three scenarios of lack of label information,cross-equipment and high noise as the research background.The main contents are as follows:Aiming at the difficulty of labeling on-site data,which makes it difficult to establish the traditional neural network model,a method of abnormal signal detection of rolling bearing based on STL-DCAE is proposed.This method can use the reconstruction error of healthy samples to establish a threshold standard for abnormality detection and realize the state judgment of vibration data.At the same time,it can adaptively determine the starting point of the failure through STL decomposition.Verified by a number of bearing full-cycle experimental data,the method in this chapter is feasible for the realization of bearing abnormal data labeling.There are differences in the distribution of vibration data collected under different equipment and different working conditions.Aiming at the problem that the traditional deep learning model is difficult to adapt to the inconsistent distribution of data sets,a bearing fault location model based on SE-SCNN is proposed.By mining the distribution of bearing data in the feature space,the model realizes fault classification across equipment and under multiple working conditions.Through the test of bearing hybrid data set,the experimental results show that the model can adapt to the difference of data distribution effectively and has good generalization performance.The bearing vibration data is easily disturbed by noise in the process of acquisition,which leads to the weak local fault pulse cannot be prominent and affects the accuracy of bearing fault diagnosis.In order to solve this problem,a rolling bearing fault diagnosis method based on OVMD-MPE group sparse total variational denoising is proposed.This method has the ability to automatically judge and extract noise components from original vibration signals.Meanwhile,the performance of signal decomposition and reconstruction is improved through the optimization algorithm,and the loss of characteristic information is avoided.The effectiveness of this method is proved by studying the denoising effect and diagnosis performance in different noise environments.