Gear Fault Diagnosis based on Distribution Adaptation Layer and Soft Label Learning

Abstract

The intelligent gear recognition method based on convolutional neural network can effectively identify the gear fault， but the convolutional neural network needs a lot of labeled training data， which limits the application of convolutional neural network in gear fault diagnosis. To solve this problem， a gear fault diagnosis method based on distribution adaptation layer and soft label learning is proposed. The convolutional neural network is used to extract features and soft labels. The distribution discrepancy is extracted by the distribution adaptation layer， and the soft label loss is generated by the soft label learning. The joint loss of distribution discrepancy， soft label loss and classification loss are used as the objective function， and the model is trained to diagnose the faults of target domain. The proposed method is verified by gear vibration signals. The results show that the proposed method can classify gear fault data accurately and effectively.

Keywords

• Gear fault diagnosis;Convolutional neural network;Deep transfer learning;Distribution adaptation layer;Soft label learning