Scholarlink

Background <p>In bearing failure diagnosis, accurate fault detection is essential for ensuring the safe operation of machinery. However, acquiring enough labeled fault data for training is expensive and challenging.</p> Objective <p>This paper aims to address key challenges in few-shot fault diagnosis by proposing a novel framework incorporating an Adaptive Detail Convolution and a dual-branch architecture.</p> Methods <p>The proposed framework consists of three main components: an Adaptive Detail Convolution module for enhanced feature extraction, a Global KAN-Transformer learning branch to model long-range dependencies between global features, and an adaptive-regularized Mahalanobis distance module to measure the similarity of local features between support and query samples.</p> Results <p>Experimental results on the CWRU dataset show that the proposed framework significantly improves classification performance in terms of accuracy, robustness, and efficiency.</p> Conclusion <p>The proposed solution offers an effective approach for few-shot bearing fault diagnosis and outperforms existing methods in both accuracy and computational efficiency.</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

Few-shot Bearing Fault Diagnosis using Adaptive Detail Convolution and Global KAN-Transformer with Mahalanobis Distance

  • 期刊论文

  • 发表日期：

    2025年5月14日

查看全文
  • Xinyu Zhao,
  • Zhanshan Zhao,
  • Xiubin Cui,
  • Jiao Yin,
  • Jinli Cao,
  • Hua Wang

摘要

Background

In bearing failure diagnosis, accurate fault detection is essential for ensuring the safe operation of machinery. However, acquiring enough labeled fault data for training is expensive and challenging.

Objective

This paper aims to address key challenges in few-shot fault diagnosis by proposing a novel framework incorporating an Adaptive Detail Convolution and a dual-branch architecture.

Methods

The proposed framework consists of three main components: an Adaptive Detail Convolution module for enhanced feature extraction, a Global KAN-Transformer learning branch to model long-range dependencies between global features, and an adaptive-regularized Mahalanobis distance module to measure the similarity of local features between support and query samples.

Results

Experimental results on the CWRU dataset show that the proposed framework significantly improves classification performance in terms of accuracy, robustness, and efficiency.

Conclusion

The proposed solution offers an effective approach for few-shot bearing fault diagnosis and outperforms existing methods in both accuracy and computational efficiency.