Joneses: bearing fault diagnosis based on joint feature extraction and enhancement

Abstract Bearing fault diagnosis is very important for the safe and stable operation of mechanical systems. In recent years, deep learning methods have made good progress in this field. However, there are still some challenges. First, the same fault may show different signal patterns under different working conditions. Second, some fault types have few samples in real scenarios. Third, although bearing signals are inherently time-series data, many existing methods fail to effectively capture their temporal dependencies. To solve these problems, this paper proposes a new method called Joneses 4 4 ‘Joneses’ embodies a dual ‘joint’ design concept: jointly extracting and utilizing fault and working-condition features, while jointly employing the Transformer and ViT to capture signal characteristics from multiple perspectives. for bearing fault diagnosis. Joneses includes Transformer and Vision Transformer (ViT) joint encoding, joint feature extraction of fault types and working conditions and feature enhancement. It first uses a Transformer encoder to process the time–frequency data and a ViT encoder to process the spectrograms. This helps extract temporal and visual features. Then, it separates fault-related features and condition-related features. Finally, feature enhancement module is introduced to improve the discriminative ability of fault features. Experiments have shown that Joneses can effectively distinguish different fault features on the CWRU dataset and accurately cluster the same fault features under different working conditions, achieving high-precision fault diagnosis. It also performs well on the BJTU-RAO dataset, effectively handling complex working conditions and compound fault types.