Adversarial Domain Adaptation for Transfer Fault Diagnosis of Roller Bearings based on Multi-MMD Alignment

A new fault diagnosis algorithm is proposed called adversarial domain adaptation for transfer fault diagnosis of roller bearings based on multi-MMD alignment, which is abbreviated ADA-MMA, It is designed to solve the problem that the different sample distribution caused unsatisfactory diagnosis effect due to the variable rotation speed under real conditions. ADA-MMA makes the model continuously learn and optimize through the adversarial mechanism of feature extractor and domain classifier. In order to prevent the negative transfer caused by the sample differences in the process of confrontation, multi-layer MMD was added in the feature extraction stage to further align the target domain features with the source domain features. The experimental results show that the model can solve the difficulties in the current fault diagnosis and has good robustness and diagnostic accuracy.