Most data-based bearing fault intelligent diagnosis methods have assumed that all data is under the same working conditions. However, the fault data under unknown working conditions cannot be fully obtained under industrial applications. When there is no prior data, the diagnostic accuracy rate of these methods will drop significantly. Therefore, we propose a zero-shot bearing fault diagnosis method based on extended semantic auxiliary information. The semantic autoencoder method is used to project the frequency-domain features of bearing signals into the semantic space and diagnoses bearing faults under unknown working conditions in the semantic space. To help the model achieve better classification, an attribute inference module based on the back propagation neural network and deep convolutional neural networks with wide first-layer kernels is proposed to extend abstract semantic information to supplement auxiliary semantic information. The bearing datasets from CWRU and MFPT are used to verify the effectiveness of the proposed method. Three comparative experiments prove the accuracy and robustness of the proposed method.
展开▼
