Research on Bearing Fault Diagnosis Based on Few-shot Adversarial Domain Adaptation

In cross-domain fault diagnosis tasks of bearings, the data distributions between the source domain (SD) and target domain (TD) often exhibit differences. Directly applying models trained on the SD to the TD typically fails to achieve satisfactory diagnostic results. Existing cross-domain diagnostic methods often require a large number of fault samples, which increases data collection costs and does not meet the practical needs of industry. Therefore, this paper proposes a few-shot adversarial domain adaptation method based on lightweight convolutional networks. By utilizing a lightweight feature extractor and adversarial mechanism to align features between the source and target domains, the classification accuracy of the TD is enhanced. Experimental results demonstrate that this method significantly improves the accuracy of the TD in multiple cross-domain classification tasks.