Research On Rolling Bearing Fault Intelligent Diagnosis Algorithm Based On Improved Convolutional Neural Network

Bearing is an important part of rotating machinery,which plays the role of supporting and and rotating mechanical parts,and the failure of bearing may lead to serious economic loss and human casualties.With the development of information technology in machinery and equipment monitoring,deep learning is applied to bearing fault diagnosis,therefore,bearing fault diagnosis using information technology becomes an important issue in industrial internet.Based on this,for the key problems in rolling bearing fault diagnosis,this paper takes rolling bearing vibration signal as the research object and convolutional neural network as the benchmark framework,proposes three kinds of bearing intelligent diagnosis models,and verifies the effectiveness of the algorithms through experiments,as follows:In order to improve the performance of bearing fault diagnosis under strong noise,CNN-GRU-A-SVM based rolling bearing fault diagnosis method under strong noise is proposed,convolutional neural network is used to extract the feature extraction in the time domain signal,gated cyclic unit screens the important information in the fault features,attention mechanism improves the self-adaptability of the network and reduces the difficulty of hyperparameter selection,support vector machine is used for the feature extraction The support vector machine is used to classify the faulty signals after feature extraction.The experimental dataset uses the Western Reserve University bearing dataset and compares with other intelligent diagnosis algorithms.The results show that the proposed method outperforms several other algorithms,has good noise immunity under high noise,and improves the accuracy of the convolutional neural network in diagnosing bearing faults under high noise.In order to solve the problems of huge number of deep convolutional neural network parameters,low computational efficiency and slow convergence speed,therefore,a hollow convolutional structure is proposed to optimize the network bearing fault diagnosis algorithm,the introduction of hollow convolution in the convolutional layer improves the perceptual field without increasing the size of the feature map,and then improves the feature extraction ability,replaces the last layer of convolution in the original WDCNN model with the Inception structure layer and pooling layer in the original WDCNN model,and replacing the fully connected layer with the pooling layer.It is demonstrated that the proposed method can reduce the number of parameters in the neural network through the derivation of the number of parameters.It is demonstrated experimentally that the proposed method does not degrade the diagnostic accuracy of bearing faults while improving the convergence speed and consumption time,thus enhancing the efficiency of convolutional neural networks in bearing fault diagnosis.In order to solve the problem of degradation of fault diagnosis performance in cross-domain scenarios,a convolutional adaptive network in cross-domain scenarios is proposed,and the input is based on a one-dimensional convolutional neural network with joint embedding of multicore MMD metric and Wasserstein metric for aligning edge distribution and and conditional distribution,which is trained with unlabeled data in the source and target domains,and the target domain data are input into the trained network model.It is demonstrated that the proposed network model has good fault diagnosis performance under different working conditions.