中国安全科学学报 ›› 2022, Vol. 32 ›› Issue (5): 35-40.doi: 10.16265/j.cnki.issn1003-3033.2022.05.2166

• 安全工程技术 • 上一篇    下一篇

基于金字塔拆分注意力的列车轮对踏面损伤诊断

何静1(), 侯娜1, 张昌凡1, 胡新亮2, 刘建华2,**()   

  1. 1 湖南工业大学 电气与信息工程学院,湖南 株洲 412007
    2 湖南工业大学 轨道交通学院,湖南 株洲 412007
  • 收稿日期:2021-12-21 修回日期:2022-03-17 出版日期:2022-05-28
  • 通讯作者:
    **刘建华(1981—),男,河南周口人,博士,副教授,主要从事电力牵引、传动与控制理论及应用等方面研究。E-mail:
  • 作者简介:

    何 静 (1971—),女,广东开平人,博士,教授,主要从事故障诊断方法及应用等方面的研究。E-mail:

    张昌凡, 教授

  • 基金资助:
    国家自然科学基金资助(52172403); 国家自然科学基金资助(62173137); 湖南省自然科学基金资助(2021JJ30217); 湖南省教育厅资助项目(19A137)

Diagnosis of train wheelset tread damage based on EPSA-ResNet

HE Jing1(), HOU Na1, ZHANG Changfan1, HU Xinliang2, LIU Jianhua2,**()   

  1. 1 College of Electrical and Information Engineering, Hunan University of Technology, Zhuzhou Hunan 412007, China
    2 College of Railway Transportation, Hunan University of Technology, Zhuzhou Hunan 412007, China
  • Received:2021-12-21 Revised:2022-03-17 Published:2022-05-28

摘要:

为解决列车轮对踏面损伤诊断准确率低、速度慢和损伤样本少的问题,提出一种基于金字塔拆分注意力网络(EPSA-ResNet)的车轮对踏面损伤诊断模型。首先,采用迁移学习方法预训练ImageNet数据集,得到模型参数,然后将其迁移到轮对踏面损伤特征数据中,并进行微调,从而获得共享模型结构和参数;其次将ResNet-50残差块中的3×3卷积替换为一种金字塔拆分注意力(PSA)模块,得到新的EPSA-ResNet,融合空间和通道注意力多级别特征,自适应地进行特征重标定;最后通过Softmax分类器得到轮对踏面损伤情况的诊断结果。结果表明:该方法能够有效识别列车轮对踏面损伤状态以及周围环境间存在的局部细微差异,诊断精度可达99.79%,优于其他深度神经网络模型。

关键词: 金字塔拆分注意力网络(EPSA-ResNet), 轮对踏面, 损伤诊断, 迁移学习, 残差网络

Abstract:

In order to address problems of low accuracy, slow speed and few damaged samples in diagnosis of train wheelset tread damage, a diagnosis model based on EPSA-ResNet was proposed. Firstly, model parameters were obtained by pre-training ImageNet data set through transfer learning, and transferred to wheelset tread damage characteristics for fine-tuning, so that shared model structure and parameters were obtained. Secondly, 3×3 convolution in ResNet-50 residual blocks was replaced by a Pyramid Split Attention (PSA) Block before a new EPSA-ResNet was gained, which integrated multi-level features of spatial and channel attention, and feature re-calibration was carried out adaptively. Finally, damage diagnosis results were obtained by Softmax classifier. The results show that this method can identify damage state of train wheelset tread and local subtle differences between surrounding environment, with an diagnosis accuracy up to 99.79%, which is better than other deep neural network models.

Key words: efficient pyramid split attention-residual network (EPSA-ResNet), wheelset tread, damage diagnosis, transfer learning, residual network