Research On Rolling Bearing Fault Diagnosis Method Based On Deep Learning

Rolling bearings are widely used in machinery and equipment in all walks of life.They are indispensable parts in the industry and play a vital role in mechanical mechanisms.The healthy state of rolling bearing is of great significance to the whole machine.So it is necessary to diagnose the bearing health.Due to the influence of large-scale and integration of modern machinery,a large number of real-time data reflecting the health status of bearings are collected for mechanical equipment.Therefore,bearing fault diagnosis based on big data has become a hot research topic in recent years.In this paper,aiming at the fault diagnosis of rolling bearing,combined with data acquisition technology and deep learning theory,three neural network fault diagnosis methods for different working conditions of bearing are proposed.The main research contents are as follows:Firstly,a bearing fault diagnosis method based on compressed data acquisition and CNN+ SVM is proposed to solve the problem of large dimension of bearing data and excessive redundant information.Compressed sensing technology is used to reduce the dimension of the sampled data,reduce the amount of data from the source,remove the irrelevant fault feature information in the data,so that the compressed collected data retains the fault data in the original data,and use the compressed data for fault diagnosis.This can reduce the storage pressure and calculation pressure of post-processing fault data.Convolution neural network has a good ability of feature extraction,therefore,one-dimensional convolutional neural network(1DCNN)is used for feature extraction.In order to improve the classification effect of the network,support vector machine(SVM)is used instead of softmax for fault classification.Secondly,one-dimensional convolutional neural network is an end-to-end diagnosis model,which only uses time-domain signal as the input of fault diagnosis and lacks the recognition of fault information in frequency domain,A bearing fault diagnosis model based on STFT and2 DCNN is proposed.The original data is sampled by sliding window sampling technology,and1024 data points are taken as a sample.Then the two-dimensional time-frequency data was obtained in this paper after the transformation of the acquisition samples using the STFT technique.The two-dimensional data is used as the input signal of the 2DCNN and the 2DCNN is trained.2DCNN has good feature extraction ability and can achieve 100% classification effect,In the actual working condition,the sample data set is unbalanced,and this model can still achieve good fault diagnosis accuracy in the case of unbalanced data set.Thirdly,in the process of bearing fault data acquisition,due to the influence of working environment,the data is mixed with noise data,which affects the feature extraction of data,For the problem that the influence of noise on data is different under different loads,the feature extraction of noisy data using 1DCNN and 2DCNN is not obvious,and the classification accuracy is low.Therefore,a fault diagnosis model based on stacked noise reduction selfencoder(SDAE)with a modified long and short term memory neural network(Attention＿LSTM)is proposed.The original data is obtained by overlapping sampling,and different kinds of noise with different intensities are added to the data.SDAE is used to denoise the noisy data,LSTM has the ability of nonlinear data mapping and feature extraction.In order to enhance the feature extraction ability of LSTM,attention mechanism is added to LSTM,Different probabilities are added to the features extracted from the network to make a trade-off,and the features of irrelevant bearing fault data are discarded.Therefore,the Attention＿LSTM was used to extract features from the noise reduction data,and finally the Softmax was used to classify the faults from the extracted features.This paper proposes CS+1DCNN＿SVM,STFT＿2DCNN and SDAE+ALSTM fault diagnosis models,which can effectively cope with fault diagnosis under different operating conditions of bearings and have high research significance and practical significance.