Scholarlink

<p>Low-speed heavy load bearings are widely used in industrial equipment; however, their low rotational speed makes fault feature extraction extremely challenging, and reproducible fault samples are usually scarce. In addition, the inherent “black-box” nature of deep learning models makes it difficult to construct accurate and interpretable diagnostic models under limited data conditions. To address these issues, this paper proposes a lightweight and interpretable fault diagnosis method for low-speed heavy load bearings based on a few-shot learning strategy. First, an efficient few-shot learning model named SATNet is developed based on the N-shot K-way framework, in which a self-attention mechanism is incorporated to enhance the learning capability of complex fault features under low-speed heavy load conditions. Second, the Teager Energy Operator (TEO) is employed as a signal enhancement technique to preprocess both acoustic emission and vibration signals under normal and faulty states. This process highlights nonlinear and energy-related signal characteristics and enables the extraction of relevant time-domain features. The processed signals are then fed into the SATNet model for training and testing to achieve accurate fault classification under different operating conditions. Finally, the SHAP and LIME algorithms are adopted to perform interpretability analysis of the diagnostic results, revealing the key time-domain features influencing classification decisions and quantifying their contributions. Experimental results demonstrate that the proposed method significantly improves both diagnostic accuracy and robustness in the fault diagnosis of low-speed heavy load bearings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A low-speed heavy load bearing fault diagnosis method based on interpretable few-shot learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月27日

查看全文
  • Jie Bai,
  • Xianbin Sun,
  • Liming Zhang,
  • Xin Zheng,
  • Peihan Zhang

摘要

Low-speed heavy load bearings are widely used in industrial equipment; however, their low rotational speed makes fault feature extraction extremely challenging, and reproducible fault samples are usually scarce. In addition, the inherent “black-box” nature of deep learning models makes it difficult to construct accurate and interpretable diagnostic models under limited data conditions. To address these issues, this paper proposes a lightweight and interpretable fault diagnosis method for low-speed heavy load bearings based on a few-shot learning strategy. First, an efficient few-shot learning model named SATNet is developed based on the N-shot K-way framework, in which a self-attention mechanism is incorporated to enhance the learning capability of complex fault features under low-speed heavy load conditions. Second, the Teager Energy Operator (TEO) is employed as a signal enhancement technique to preprocess both acoustic emission and vibration signals under normal and faulty states. This process highlights nonlinear and energy-related signal characteristics and enables the extraction of relevant time-domain features. The processed signals are then fed into the SATNet model for training and testing to achieve accurate fault classification under different operating conditions. Finally, the SHAP and LIME algorithms are adopted to perform interpretability analysis of the diagnostic results, revealing the key time-domain features influencing classification decisions and quantifying their contributions. Experimental results demonstrate that the proposed method significantly improves both diagnostic accuracy and robustness in the fault diagnosis of low-speed heavy load bearings.