Research On Analysis Method Of Rolling Bearing Vibration Signal Based On Deep Learning

Rolling bearings are one of the core components of rotating machineries.Once a failure occurs,the equipment can not operate normally,or even lead to serious safety accidents.Therefore,it is of great practical significance to analyze the vibration signal of rolling bearing and predict the fault of rolling bearing.With the rapid development of artificial intelligence technology based on deep learning,the application research of deep learning has become a hot spot in the engineering field.In this paper,based on the one-dimensional time-domain vibration signals of rolling bearings,the deep neural network under different environments is studied to solve the problems of less data samples and weak fault signals with noise in the actual production environment.First,a deep learning analysis method for one-dimensional time-domain vibration signals is studied,and a convolutional neural network(CNN)model and a long short-term memory network(LSTM)model are established respectively.The average accuracy of the models on the test set can reach 99.8% and 98.1%,respectively.Aiming at the classification of bearing conditions under multiple working conditions,a model of convolutional circular neural network(CNN＿LSTM)was proposed by combining the advantages of high accuracy of convolutional neural network and good stability of long and short memory network.The accuracy of the CNN＿LSTM model on the test set can reach 98.6%.Secondly,aiming at the problem of less data samples in the actual production environment,a transfer learning method was proposed to analyze bearing vibration signals and realize fault diagnosis.One-dimensional original vibration signals are first transformed into two-dimensional matrix(GAF)and time-frequency image(STFT),and then the Inception V3 model is fine-tuned using fine-tuning techniques.Using GAF and Inception v3 models,the accuracy rate on the test set is 95.4%,and the accuracy rate on the test set is 97.6%combining the STFT and Inception v3 models.Finally,in view of the problem that rolling bearings are susceptible to various noise interferences in a complex environment,the wavelet threshold denoising(DWT)method is adopted in this paper to denoise vibration signals.In order to identify weak bearing fault signals from noise interference,a model of twin convolutional network(Siamese CNN)is proposed in combination with the strong feature extraction ability of convolutional neural network and the ability of twin network(Siamese)to amplify the weak gap between two similar samples.By comparing the Siamese convolutional network model with other network models,experiments prove that the Siamese convolutional network model has a good classification effect for samples with high similarity.The accuracy of twin convolutional network on training set,verification set and test set is 99.88%,99.5% and 99.3%,respectively.