Unsupervised Domain Adaptive Method Based on Time Domain Mask Consistency in Bearing Fault Diagnosis

Unsupervised domain adaptation approaches have been introduced into the fault diagnosis domain due to the domain bias caused by the differences in real-life working conditions. However, the model still lacks effective supervised learning to guide, resulting in poor identification of fine-grained differences in the target data and difficulty in predicting data far from the sample centroid. This paper proposes a time domain mask consistency method based on unsupervised domain adaptation. Since there is a good contextual correlation ability in time series, we encourage the model to predict the masked vibration data as a way to enable the model to learn the contextual correlations of the time domain data. In this paper, the method is validated using a bearing fault dataset for different domain adaptation tasks, and the results show that the method can be flexibly applied to different unsupervised domain adaptation methods for bearing fault diagnosis and achieves a performance improvement of $12.8 \%$ on the open-set domain adaptation task, which is the most effective.