Scholarlink

<p>In recent years, deep learning has achieved great success in the field of wheel tread fault diagnosis due to its robust feature learning capabilities. However, these methods often suffer from limited universality across different data samples due to the non-stationary working conditions of wheelsets, resulting in low diagnostic accuracy. To address this issue, this paper proposes a novel fault diagnosis method based on a Multi-Scale Convolutional Neural Network with Transfer learning and Efficient Channel Attention (TFECA-MCNN). By maximizing Multi-Scale Maximum Mean Discrepancy (MMD) and reducing feature discrepancy between the source and target domains, the model achieves deep transferable feature learning and effectively addresses cross-speed conditions. Experimental results from 12 transfer tasks set using wheel test rig demonstrate that the TFECA-MCNN model achieves an average diagnostic accuracy of 99.28% in cross-speed conditions, significantly outperforming several commonly used methods. This study provides a new method for monitoring and diagnosing wheel tread faults in rail transit vehicles.</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

Fault Diagnosis of Wheel Tread Based on Deep Transfer Convolution Neural Network

  • 期刊论文

  • 发表日期：

    2025年2月17日

查看全文
  • Aihua Liao,
  • Dingyu Hu,
  • Rongming Liu,
  • Wei Shi,
  • Yajing Huang

摘要

In recent years, deep learning has achieved great success in the field of wheel tread fault diagnosis due to its robust feature learning capabilities. However, these methods often suffer from limited universality across different data samples due to the non-stationary working conditions of wheelsets, resulting in low diagnostic accuracy. To address this issue, this paper proposes a novel fault diagnosis method based on a Multi-Scale Convolutional Neural Network with Transfer learning and Efficient Channel Attention (TFECA-MCNN). By maximizing Multi-Scale Maximum Mean Discrepancy (MMD) and reducing feature discrepancy between the source and target domains, the model achieves deep transferable feature learning and effectively addresses cross-speed conditions. Experimental results from 12 transfer tasks set using wheel test rig demonstrate that the TFECA-MCNN model achieves an average diagnostic accuracy of 99.28% in cross-speed conditions, significantly outperforming several commonly used methods. This study provides a new method for monitoring and diagnosing wheel tread faults in rail transit vehicles.