| Bearing is one of the key components of rotating machines,and its working environment is complicated,easy to cause a malfunction or even failure of the bearing,cause serious economic losses and casualties,so bearing fault diagnosis as well as bearing health monitoring is an important means of modern machinery and equipment maintenance The traditional method of rolling bearing fault signal diagnosis rely on experts experience Artificial feature extraction and the low accuracy problem,can’t meet in the big data under the background of modern intelligent fault diagnosis To solve above problems,this paper put forward on the basis of deep learning theory model of bearing fault diagnosis network,studied the different network algorithm,and can be achieved under the complex environment of noise and variable condition of rolling bearing fault of high recognition rate,The main research contents are as follows:(1)According to the characteristics of time-domain signal of bearing fault vibration,a Parallel Multichannel Deep Convolution Neural Network(PMDCNN)based on Parallel Multichannel is proposed in this paper.The experimental results show that the network model can extract different signal characteristics through different channels,which can realize the bearing fault diagnosis with high accuracy.On the basis of the model,the performance of the model is improved through the study of the model layer number.The ADAM optimizer and the appropriate learning rate are selected to accelerate the convergence speed of the model.The batch normalization algorithm is used to improve the stability of the network model.Dropout technology is used to improve the generalization ability of the network.Aiming at the problem of excessive number of parameters in the network model,the number of parameters in the PMDCNN network is optimized by local sparsity principle to make it 1/70 of the previous number.(2)To solve the problem that the accuracy of network identification is reduced when bearings are disturbed by noise and variable working condition environment,a Long ShortTerm Memory Network(LSMT)is adopted,The combination of Memory(LSTM)and the model(PMDCNN-LSTM)enables it to better extract bearing signal fault features.The ELU activation function is used to alleviate the problem of network gradient disappearance.Through the study of batch normalized number and training samples,the training time of the model is greatly reduced and the performance of the model is improved,The results show that the network has strong anti-noise and anti-variable performance.By comparing with other algorithms,the test results show that the optimized PMDCNN-LSTM has good applicability,strong stability and high accuracy of bearing fault identification.(3)Aiming at the problems of the small amount of measured bearing and gear data and the difficulty of approximate domain training model,a fault diagnosis network of measured bearing and gear based on migration model was designed.The experimental results show that the model trained on the CWRU data set can retain the fault identification ability of the original model to a great extent when it is transferred to the field of measured bearings and gears. |
