Scholarlink

<p>Bearing fault diagnosis often struggles to extract reliable and discriminative features from original multi-modal fault samples that contain noise and redundant information. Aiming at the issue, we propose a bearing fault diagnosis method called category-aware multi-modal sparse space learning (CMSSL). In the method, we design a novel category-aware geometric metric to better capture structural differences among samples and construct category-aware graphs that preserve intrinsic local geometric information across modalities. These graphs are jointly incorporated with an L2, 1-norm sparsity constraint into a unified correlation analysis framework, enabling the extraction of compact and highly discriminative features. By learning a category-aware sparse projection space, CMSSL transforms the original multi-modal samples into feature representations with substantially improved category separability. Experimental results on Paderborn and our experimental platform demonstrate that the robustness and effectiveness of CMSSL in multi-modal bearing fault diagnosis.</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

Bearing fault diagnosis via category-aware multi-modal sparse space learning

  • 期刊论文

  • 发表日期：

    2026年6月3日

查看全文
  • Shuzhi Su,
  • Zini Qin,
  • Yanmin Zhu,
  • Mengjie Xu,
  • Yanyan Chen,
  • Chuang Dong

摘要

Bearing fault diagnosis often struggles to extract reliable and discriminative features from original multi-modal fault samples that contain noise and redundant information. Aiming at the issue, we propose a bearing fault diagnosis method called category-aware multi-modal sparse space learning (CMSSL). In the method, we design a novel category-aware geometric metric to better capture structural differences among samples and construct category-aware graphs that preserve intrinsic local geometric information across modalities. These graphs are jointly incorporated with an L2, 1-norm sparsity constraint into a unified correlation analysis framework, enabling the extraction of compact and highly discriminative features. By learning a category-aware sparse projection space, CMSSL transforms the original multi-modal samples into feature representations with substantially improved category separability. Experimental results on Paderborn and our experimental platform demonstrate that the robustness and effectiveness of CMSSL in multi-modal bearing fault diagnosis.