Scholarlink

Purpose <p>The purpose of this study is to address the limitations of existing intelligent fault diagnosis methods, including poor generalization ability, weak adaptability under varying operating conditions, and excessive model complexity. </p> Methods <p>A lightweight transfer learning network that combines adaptive channel enhancement with a sparse optimization model (ACESOM) is proposed. Specifically, a continuous wavelet transform (CWT) of the vibration signal is performed to generate a time–frequency map, enabling ACESOM to effectively extract key features. Then, the proposed domain adaptation joint loss function is combined with a differential metric mechanism to enhance the mapping of characteristics across different domains in an adversarial transfer strategy. In addition, a pruning strategy is introduced to reduce redundant features, which decreases the complexity of the model while maintaining the diagnostic accuracies. </p> Conclusion <p>The adaptability and robustness of the model are verified by two experimental datasets, and the results show that ACESOM has excellent fault diagnosis performance and effectiveness.</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

Intelligent Fault Diagnosis of Bearing in Lightweight Transfer Learning Based on Enhanced Pruning Under Variable Conditions

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Qianqian Zhang,
  • Haitao Yan,
  • Zhongwei Lv,
  • Chan Xu,
  • Weihuang Liu,
  • Lei Xu,
  • Zhuang Wen,
  • Qiuxia Fan

摘要

Purpose

The purpose of this study is to address the limitations of existing intelligent fault diagnosis methods, including poor generalization ability, weak adaptability under varying operating conditions, and excessive model complexity.

Methods

A lightweight transfer learning network that combines adaptive channel enhancement with a sparse optimization model (ACESOM) is proposed. Specifically, a continuous wavelet transform (CWT) of the vibration signal is performed to generate a time–frequency map, enabling ACESOM to effectively extract key features. Then, the proposed domain adaptation joint loss function is combined with a differential metric mechanism to enhance the mapping of characteristics across different domains in an adversarial transfer strategy. In addition, a pruning strategy is introduced to reduce redundant features, which decreases the complexity of the model while maintaining the diagnostic accuracies.

Conclusion

The adaptability and robustness of the model are verified by two experimental datasets, and the results show that ACESOM has excellent fault diagnosis performance and effectiveness.