Bearing Fault Diagnosis Based on Deep Residual Network and Particle Swarm Optimization

Bearings as the key component of a rotating machinery, their health affects the operation of the entire rotating machinery. Effective and accurate bearing fault diagnosis can ensure the safety of machinery operation and reduce maintenance costs. Nowadays, more and more bearing fault diagnosis methods are developed based on deep learning (DL) network. Although great progresses have been made from the DL-based methods, the diagnosis results highly rely on expert experience and manual hyperparameter settings. To solve such an issue, this paper proposes a bearing fault diagnosis method based on deep residual network (ResNet) and particle swarm optimization (PSO), the proposed method is named as ResNet-PSO-based bearing fault diagnosis method. In the proposed method, ResNet is used to improve the diagnosis accuracy, and PSO is adopted to automatically optimize the hyperparameters in ResNet. The effectiveness of the proposed ResNet-PSO-based bearing fault diagnosis method is validated on the bearing dataset from Paderborn Bearing Data Center. Some state-of-the-art methods are compared, and the results show the outperformance of the proposed method.