RMHBG-SN: A Siamese Network Framework to Tackle Data Sparsity and Imbalance in Bearing Fault Diagnosis

Bearing fault detection is crucial for the safe operation of industrial systems. With the advancement of machine learning, numerous advanced machine learning methods have been applied to bearing fault detection and achieved favorable results. However, the complex real-world industrial environment often prohibits bearings from operating under faulty conditions, leading to issues such as scarce fault data and imbalanced fault datasets. To address these challenges, this paper proposes a fault diagnosis framework based on siamese network, which consists of three components: a data augmentation module, an RMHBG-SN feature extraction model, and a diagnosis module. The framework integrates generative and discriminative models via three collaborative components to boost fault diagnosis accuracy under small sample conditions. To validate the proposed method's effectiveness, comparative experiments were conducted on the CWRU dataset against baseline models and advanced small sample learning approaches. The experimental results demonstrate that the proposed method exhibits superior performance when confronting problems such as limited sample sizes and dataset imbalance.