Transfer Learning Based Rolling Bearing Fault Diagnosis

In recent years, transfer learning has been an important method to address the problem that labeled data are rarely in the real world. In many industry scenarios, collected labeled sample signals are usually not in the same data distribution. A major assumption for traditional learning and deep learning based bearing fault diagnosis methods is the training data and testing data must follow the same data distribution. However, this assumption may not hold in reality. To address the different distribution problem, this paper proposed an unsupervised approach for bearing fault diagnosis based on transfer learning. The correlation alignment (CORAL) algorithm is used to align data distribution of domains in the proposed approach, then the proposed approach applies statistical algorithms to extract shallow features and wavelet scattering network to extract deep features. The 1 nearest neighbors (1-NN) classifier is trained with the feature vector matrix of source domain, which is able to classify the unlabeled samples of target domain presenting the effectiveness of the proposed approach. Different experiments were carried out to demonstrate the performance of the proposed approach. The experiment results show that the proposed model is superior to other bearing fault diagnosis methods.