Research On Rolling Bearing Fault Diagnosis Method Based On Feature Fusion And Transfer Learning

With the wide application of intelligent manufacturing equipment,the loss caused by mechanical failure is increasing.It is important to identify and troubleshoot problems at an early stage.The traditional fault diagnosis method is data-driven,which mainly includes data collection,feature extraction and fault classification.The result of feature extraction will affect the classification accuracy to some extent.Because Convolutional Neural Network(CNN)can adaptively extract features from original signals,it can eliminate the influence on traditional manual features.A bearing fault diagnosis method based on feature fusion and transfer learning is presented in this paper.In this paper,a convolutional neural network model based on feature fusion is proposed.After preprocessing,the data set is sent to the one-dimensional(1D-CNN)and twodimensional(2D-CNN)feature extractor for training.In the aggregation layer,the output of the two pooling layers is connected into a vector and sent to the full connection layer,and the output of the full connection layer is extracted as the feature data.The network can learn the local correlation between adjacent and non-adjacent intervals in the periodic signal of vibration data.Therefore,it is proposed to adopt different feature extractors to carry out feature fusion training.In this way,the fusion process of the two features is the training learning process,and the label of the output result can be backpropagated to turn the past feature combination into a real fusion.At the same time,compared with classifier softmax,Support Vector Machine(SVM)classifier is more local target,only need to meet the boundary value,and will not restrict the specific score of the operation too much.Therefore,the classifier decides to adopt SVM.Experimental results show that the proposed method can effectively improve the classification accuracy.In addition,a combination of transfer learning and deep learning is used to conduct adaptive feature extraction and multi-state fault identification of rolling bearing vibration signals.In the process of constructing the network model training,the statistical distribution and spatial joint tuning algorithm are introduced to reduce the differences between different domains,and a migration learning failure model based on intra-class convolution is constructed,which extracts samples from the target domain By adjusting the high-level features of the same type of samples in the source domain and the target domain as the input of the classifier,the experiment shows that compared with the traditional deep learning model,this method can accurately judge the health status of the train bearing under different working conditions,Has a good application prospect.