Domain Adaptive Rolling Bearing Fault Diagnosis based on Wasserstein Distance

The rolling bearing usually runs at different speeds and loads, which leads to a corresponding change in the distribution of data. The cross-domain problem caused by different data distributions can degrade the performance of deep learning-based fault diagnosis models. To address this problem, this paper proposes a multilayer domain adaptive method based on Wasserstein distance for fault diagnosis of rolling bearings operating under different operating conditions. First, the basic framework uses deep Convolutional Neural Networks (CNN) to extract domain invariant features and Then an adaptation learning procedure is used for optimizing the basic CNN to adapt cross different domains. According to the experimental results, the network model has excellent fault diagnosis capability and adaptive domain capability and is able to obtain a high fault diagnosis rate under different working conditions. Finally, for investigating the adaptability in this method, we use t-SNE to reduce the high dimension feature for better visualization.