基于Vision Transformer网络的旋转设备迁移故障诊断OA

With the rapid development of intelligent manufacturing,a large number of monitoring data are generated during the operation of rotating machines,which promotes the vigorous development of intelligent diagnosis methods based on deep learning.However,the intelligent diagnosis method based on deep learning needs a large number of labeled sample data when updating its network parameters.Most of the monitoring data collected from rotating machines in the industrial field are sample data without labels,and labeling these data require high cost.Transfer learning aims to use existing knowledge to solve the problem that the label information of target domain sample data is scarce or even difficult to obtain.Most of the existing transfer fault diagnosis methods use convolutional neural network(CNN)as the basic network model.A fixed local receptive field adopted in the CNN cannot effectively represent the global information of high-dimensional features.As a comparison,a self-attention mechanism adopted in Vision Transformer(ViT)network can fully mine the correlation between different parts of features and effectively represent the global information of high-dimensional features.Thus,a transfer fault diagnosis method of rotating machines based on ViT is proposed,the feature transfer from source domain to target domain is realized through an adversarial training approach.The application of the proposed method on the rolling bearing fault dataset verifies that the proposed method owns a superior diagnosis effect in comparison with the other comparative methods.