Research On Bearing Fault Diagnosis Method Based On Deep Learning

Rolling bearings are not only an important part of industrial production equipment,but also one of the largest vulnerable parts in rotating machinery equipment,once there is a problem,it will cause the entire equipment to shut down,thus affecting the operation of the entire production line.Traditional bearing fault diagnosis methods rely on expert experience with certain limitations.With the rapid development of convolutional neural networks,deep learning-based methods are gradually applied to mechanical fault diagnosis,and based on the rolling bearing vibration database created by Case Western Reserve University,experts and scholars have carried out extensive research on the analysis methods of bearing vibration signals.In this paper,the fault diagnosis method of rolling bearings based on deep learning is studied,and the research content is as follows:This paper first aims at-the traditional mechanical equipment bearing diagnosis field,relying on manual diagnosis time-consuming and inefficient,combined with the current research results and advanced of deep learning in the field of computer vision,this paper proposes an automatic bearing fault diagnosis method,using GAF(Gramian Angular Field)-TSNE to convert one-dimensional vibration signal into two-dimensional image and then input it into Convolutional neural network,different sizes of images will also have a great impact on the diagnosis results,too small size will lead to the lack of image feature information,which will reduce accuracy.If the size is too large,the diagnostic accuracy will not be improved When the efficiency of diagnosis is seriously affected.The current method of converting one-dimensional vibration signal into two-dimensional image lacks research in this regard,and this paper takes advantages of T-SNE excellent ability to visualize the classification effect IV to compare the feature classification ability converted into images of different sizes,and select the best conversion size.In view of the weak data augmentation effect under the condition that the length of the time series data is insufficient,the data augmentation scheme adopted in this paper uses the GAF algorithm to convert each column of data into the corresponding GAF image for each column of data in the data matrix corresponding to the transformed image,which can effectively expand the data volume under the condition that the length of the time series data is insufficient.Then,the neural network of the first layer of 8*8 convolution kernel is designed for the features of the two-dimensional image of vibration data,which can effectively extract the features useful for diagnosis.Then,the dropout module is added at each fully connected level,which reduces the amount of neuron parameters,effectively alleviates the problems of diagnostic accuracy and loss curve oscillation,and has good robustness.Finally,in view of the lack of research on the diagnostic performance of the model under variable working conditions(noise and variable load)in the field of bearing faults,this paper adds the MMD(Maximum Mean Discrepancy)algorithm to the model.specifically the bearing vibration data under variable working conditions is added to the training data in a certain proportion to the training data to train the parameters of the model,and the MMD(Maximum Mean Discrepancy)algorithm is used to calculate the difference between the feature distribution of the source domain and the target domain data after model training and feature extraction,and take it into account in the loss function.The recalculated loss function is substituted into the original model and then the target domain data is classified and diagnosed,and the model diagnosis accuracy is greatly improved compared with the model diagnosis rate without MMD(Maximum Mean Discrepancy)algorithm.and it proposed that the model diagnosis scheme based on most unified rules can be used under complex working conditions such as high noise and large load,which can greatly improve the diagnostic accuracy,and compare it with some other model diagnostic performance popular in the field of fault diagnosis at this stage,and this model has unique advantages for bearing fault diagnosis.