Bearing Fault Diagnosis Based On Hybrid Domain Attention Mechanism Convolutional Network And Residual Contraction Network

Rotating machinery plays a very important role in modern industry,and the failure of its components may cause machine failure,production delays and catastrophic accidents,causing huge economic losses.Therefore,in recent decades,the condition monitoring of rotating machinery has been paid more and more attention.Rolling bearings are the most widely used and most prone to failure in rotating machinery.It plays a vital role in the normal operation of rotating machinery.Therefore,the fault diagnosis of rolling bearings has always been the focus of research in the field of fault diagnosis.In recent years,with the widespread application of artificial intelligence and deep learning technologies in industrial production tasks,the application of deep learning in rolling bearing fault diagnosis has become a major research hotspot.In this thesis,for the problem of difficult fault diagnosis of rolling bearings under variable load conditions and high noise environments,respectively,a convolutional neural network method based on mixed domain attention and a residual shrinkage network model combining mixed domain attention are proposed.In order to solve these two problems,new ideas and methods were proposed.(1)Rolling bearing fault diagnosis under variable working conditionsIn the past,researches on rolling bearing fault diagnosis mostly focused on the diagnosis under the same working conditions,and all achieved high diagnostic accuracy.However,the research results on fault diagnosis under variable working conditions are relatively small,and the accuracy of diagnosis is mostly low.In this thesis,a fault diagnosis method based on a hybrid domain attention mechanism combined with a one-dimensional convolution neural network is used for bearing fault diagnosis under variable operating conditions.In order to verify the superiority of the mixed domain attention method in dealing with this problem,this thesis implements three attention models for fault diagnosis problems based on different attention angles,namely Time Attention Model(TA)The Channel Attention Model(CA for short)is also a mixed domain attention model(Mix Attention for MA)that works together in the time channel domain.By inserting three different modules between each convolution layer of the one-dimensional convolution neural network(CNN1D),three diagnostic models CNN1D-TA,CNN1D-CA,and CNN1D-MA are obtained.The diagnostic accuracy of the three diagnostic models under variable working conditions was analyzed and compared through experiments.The results show that the CNN1D-MA model of the one-dimensional convolution neural network model based on mixed domain attention has an average accuracy rate of 97.47%under variable working conditions,which is much higher than the traditional diagnostic models such as FFT-SVM,FFT-MLP,The average accuracy rate of FFT-DNN is 10%～20%higher than the average accuracy rate of CNN1D model without added attention module by 3.78%;2.09%higher than CNN1D-TA model;and 1.77%higher than CNN1D-CA model.(2)Fault diagnosis of rolling bearings in noisy environmentIn the actual industrial production,the working environment of rolling bearings is complex.Once a large noise interference occurs,it will bring problems to the maintenance and diagnosis of bearings.If diagnostic errors or inaccurate equipment maintenance are affected due to noise interference,once the equipment failure is caused,it will not only affect production and cause economic losses,but it may also cause safety accidents.Therefore,it is very important to ensure the accuracy of diagnosis in high noise environments.Aiming at the problem of rolling bearing fault diagnosis in a high noise environment,this thesis proposes an improved hybrid based on the mixed-domain attention model based on the channel-by-channel deep residual shrinkage network(DRSN-CW)that performs well under the existing noise environment.Domain Attention Deep Residual Shrinkage Network(DRSN-MA).By adding-5dB～5dB signal-to-noise ratio Gaussian white noise to the original bearing vibration signal to simulate the actual production noise environment,11 experiments with different signal-to-noise ratios were designed.A 10-fold cross-validation was used to compare the accuracy of the model before and after the improvement.The experimental results show that the improved DRSN-MA model improves the average accuracy by 0.4%compared with the DRSN-CW before the improvement,but it improves by 3%in a high-noise environment of-5dB,so the improved DRSN-MA model performance has been improved and is more stable in high noise environments.