Scholarlink

When subgrade settlement occurs, especially if it is significant, the structural integrity and functionality of the track interlayer may be significantly and negatively affected. There may be discontinuous contact (void) between track and subgrade, which is particularly difficult to monitor. In this research, a damage identification method based on one-dimensional convolutional neural (1D-CNN) network and vehicle-track coupled dynamics is proposed. Frist, a large-scale vehicle-track-subgrade coupled dynamic model is established to simulate the impact of settlement-induced track voids on the train’s vertical vibration acceleration Then a 1D-CNN network capable of automatically extracting features is built to identify the track voids from the vehicle acceleration signals. The efficiency and accuracy of the method is improved through training, optimization, and validation adjustments. The results show that the identification rate of this method reaches 98%.

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

Identification of Track Voids Caused by Differential Subgrade Settlement Through Integration of 1D-CNN and Vehicle-Track Coupled Dynamics

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Youwei Zhang,
  • Yu Guo

摘要

When subgrade settlement occurs, especially if it is significant, the structural integrity and functionality of the track interlayer may be significantly and negatively affected. There may be discontinuous contact (void) between track and subgrade, which is particularly difficult to monitor. In this research, a damage identification method based on one-dimensional convolutional neural (1D-CNN) network and vehicle-track coupled dynamics is proposed. Frist, a large-scale vehicle-track-subgrade coupled dynamic model is established to simulate the impact of settlement-induced track voids on the train’s vertical vibration acceleration Then a 1D-CNN network capable of automatically extracting features is built to identify the track voids from the vehicle acceleration signals. The efficiency and accuracy of the method is improved through training, optimization, and validation adjustments. The results show that the identification rate of this method reaches 98%.