Fault diagnosis of imbalanced multi-channel data based on deep transfer learning

The fault diagnosis is an effective method to reduce downtime and maintenance costs. However, the fault samples obtained in actual production are far less than normal samples, which leads to the fault condition of small samples is difficult to be effectively identified. To solve the problem of imbalanced samples, this paper proposes a novel fault diagnosis model, which combines the Convolutional Neural Network and transfer learning. The proposed methodology can make the fault diagnosis model be more inclined to small sample fault data. The effectiveness of the proposed model is verified by a real-world case study. The results show that the deep transfer learning model has excellent fault diagnosis performance in the case of sample imbalance.