Scholarlink

<p>Roller bearings play a crucial role in mechanical systems, where their operational condition directly impacts system performance and lifespan. However, detecting early-stage bearing faults during routine maintenance remains a challenge due to the cost and technical limitations of current fault diagnosis methods, often resulting in reduced accuracy. To address this issue, this paper proposes a bearing fault diagnosis method based on the MBConv Vision Transformer (MB-ViT) with frequency feature fusion. Specifically, we introduce a novel approach that transforms bearing fault signals into RGB images by combining Continuous Wavelet Transform (CWT) and Gramian Angular Summation Field (GASF) images, thereby enhancing feature representation and improving fault recognition. Additionally, recognizing the limitations of the traditional Vision Transformer (ViT) in capturing local features, we introduce the MB-Multi-Head Self-Attention (MB-MSA) module to overcome this challenge. Experimental results using data from Case Western Reserve University and Xi’an Jiaotong University show that feature fusion significantly improves fault diagnosis accuracy, while the MB-MSA module enhances both diagnostic precision and robustness. MB-ViT achieves an accuracy of 99.90<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="521_2025_11509_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>%</mo> </math></EquationSource> </InlineEquation> in bearing fault diagnosis tasks, demonstrating its superior performance. In summary, the proposed model outperforms existing ViT-based methods, providing a promising solution for bearing fault diagnosis and supporting the advancement of next-generation industrial technologies. The code and data are available at <a href="https://github.com/viivan/MB-vit">https://github.com/viivan/MB-vit</a>.</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

MB-ViT: MBConv vision transformer with time–frequency feature fusion for bearing fault diagnosis

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Gang Xiao,
  • Junbo Yao,
  • Liubing Zhong,
  • Zhongcheng Xiao,
  • JiaWei Lu

摘要

Roller bearings play a crucial role in mechanical systems, where their operational condition directly impacts system performance and lifespan. However, detecting early-stage bearing faults during routine maintenance remains a challenge due to the cost and technical limitations of current fault diagnosis methods, often resulting in reduced accuracy. To address this issue, this paper proposes a bearing fault diagnosis method based on the MBConv Vision Transformer (MB-ViT) with frequency feature fusion. Specifically, we introduce a novel approach that transforms bearing fault signals into RGB images by combining Continuous Wavelet Transform (CWT) and Gramian Angular Summation Field (GASF) images, thereby enhancing feature representation and improving fault recognition. Additionally, recognizing the limitations of the traditional Vision Transformer (ViT) in capturing local features, we introduce the MB-Multi-Head Self-Attention (MB-MSA) module to overcome this challenge. Experimental results using data from Case Western Reserve University and Xi’an Jiaotong University show that feature fusion significantly improves fault diagnosis accuracy, while the MB-MSA module enhances both diagnostic precision and robustness. MB-ViT achieves an accuracy of 99.90 \(\%\) % in bearing fault diagnosis tasks, demonstrating its superior performance. In summary, the proposed model outperforms existing ViT-based methods, providing a promising solution for bearing fault diagnosis and supporting the advancement of next-generation industrial technologies. The code and data are available at https://github.com/viivan/MB-vit.