In order to realize accurate bearing fault identification under variable equipment and working conditions,a diagnosis method based on domain adaptive migration depth convolution neural network is proposed.For the training set and test set with different distribution features(i.e.different domains),the fault feature extraction module,domain recognition module and label classification module are constructed in the deep convolution neural network.The adaptive transfer ability of domain is realized by the way of feature extraction module and domain recognition module against training,so that the deep convolution neural net-work can effectively extract the common feature parameters of different domains.Four groups of migration experiments are de-signed based on the data of Case Western Reserve University and intelligent maintenance system center.The average recognition accuracy of traditional deep convolution neural network is 64.5%,and the average recognition accuracy of domain adaptive mi-gration convolution neural network is 94.9%,which fully shows that the domain adaptive migration deep convolution neural net-work can effectively identify bearing faults under variable equipment and working conditions.
关键词
轴承故障诊断/域自适应迁移/深度卷积神经网络/对抗训练
Key words
Bearing Fault Diagnosis/Domain Adaptive Migration/Deep Convolutional Neural Network/Adver-sarial Training
维普
万方数据