In practical engineering,it may be difficult and expensive to acquire numerous fault samples,that is,the fault samples used for training are relatively few,which makes it difficult for the fault diagnosis mod-els trained by the traditional deep learning methods to quickly adapt to the new working conditions.There-fore,an intelligent fault diagnosis method based on meta-transfer learning is proposed.Firstly,a well-de-signed self-attention network(SAN)that effectively extracts fault features is pre-trained.Secondly,a base-leaner and a meta-learner are trained on each"N-way M-shot"meta-task to adaptively adjust the parame-ters of the pre-trained SAN model.Finally,the effectiveness of the proposed method is verified through dif-ferent few-shot fault diagnosis(FSFD)tasks.The results prove that the method achieves high diagnosis ac-curacies under cross-working condition FSFD tasks.
维普
万方数据