基于深度学习的钢桥面板U肋-顶板节点Lamb波损伤检测
DOI:
作者:
作者单位:

1.天津城建大学土木工程学院,天津 300384 ;2.中国铁建大桥工程局集团有限公司,天津 300300 ;3.天津市装配式桥梁智能建造技术与装备重点实验室,天津 300300 ;4.清华大学土木工程系,北京 100084

作者简介:

通讯作者:

田亮(1984—),男,副教授,博士,硕士生导师。研究方向为高性能钢桥结构、钢桥疲劳与断裂。E-mail:sjtu_tl@126.com。

中图分类号:

[U24];U441

基金项目:

天津市自然科学基金项目(24JCYBJC00850);中国铁建股份有限公司科研重大专项(2023-A01);中国铁建大桥局集团有限公司科技创新项目(DQJ-2024-B05)


Research on Lamb Wave Damage Detection in U-Rib-Deck Joints of Steel Bridge Decks Based on Deep Learning
Author:
Affiliation:

1.College of Civil Engineering, Tianjin Chengjian University, Tianjin 300384 , China ; 2.China Railway Construction BridgeEngineering Bureau Group Co., Ltd., Tianjin 300300 , China ; 3.Tianjin Key Laboratory of Prefabricated Bridge IntelligentConstruction Technology and Equipment, Tianjin 300300 , China ;4.Department of Civil Engineering, Tsinghua University, Beijing, 100084 , China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对钢桥面板等复杂结构中Lamb波多模态传播、频散效应及信号衰减导致的损伤特征识别困难问题,本研究提出一种基于深度学习的钢桥面板U肋-顶板节点损伤检测方法。通过将挤压和激励(squeeze-excitation,SE)注意力机制与长短时记忆网络(long short-term memory,LSTM)嵌入卷积神经网络(convolutional neural networks,CNN),并结合Hilbert变换提取包络曲线构建数据集,实现钢桥面板U肋-顶板节点典型疲劳损伤的有效识别。研究结果表明:① 损伤状态下直达波包相位呈现右移且幅值衰减,验证信号时域变化特征在损伤检测上应用的可行性;② SE-LSTM-CNN模型在验证集与测试集分别达到 93.67%与95.00%的准确率,且各类损伤识别精度均超过90%,验证该模型在钢桥面板U肋-顶板节点损伤检测任务上有良好适用性;③ SE-CNN与LSTM-CNN模型的分类准确率较基础CNN模型分别提升1.00%与3.33%;而SE-LSTM-CNN模型的分类准确率较单一改进模型再提升7.33%与5.00%,验证SE注意力机制与LSTM的协同增效作用。此外,使用包络曲线数据集使模型在验证集上的准确率较原始信号提升21.33%,说明该方法能有效增强SE-LSTM-CNN模型对Lamb波损伤特征的辨识能力;④ 基于MATLAB APP Designer构建的智能检测软件实现了损伤检测全流程优化,降低了人工干预误差。本研究有望为钢桥面板U肋-顶板节点的损伤检测提供新的技术方案。

    Abstract:

    To address the challenges in identifying damage characteristics caused by multimodal Lamb wave propagation, dispersion effects, and signal attenuation in complex structures like steel bridge decks, this study proposes a deep learning-based damage detection method for U-Rib-Deck joints in steel bridge decks. By embed-ding squeeze-excitation (SE) attention mechanisms and long short-term memory (LSTM) networks into convolutional neural networks (CNN), combined with constructing datasets using Hilbert transform envelope curves, effective identification of typical fatigue damages in U-Rib-Deck joints is achieved. The research results demonstrate: ① Under damage conditions, the direct wave packet exhibits a rightward phase shift and amplitude attenuation, confirming the feasibility of using time-domain signal changes for damage detection. ② The SE-LSTMCNN model achieved validation accuracy and test accuracy of 93.67% and 95.00%, respectively, with the recognition accuracy for all types of damage exceeding 90%, indicating the model’s excellent applicability for damage detection tasks in steel bridge deck U-Rib-Deck joints. ③ The classification accuracy of the SE-CNN and LSTM-CNN models improved by 1.00% and 3.33%, respectively, compared to the baseline CNN model, while the SE-LSTM-CNN model further improved accuracy by 7.33% and 5.00% compared to the single-improvement models, validating the synergistic effectiveness of SE attention mechanism and LSTM for damage detection in steel bridge deck U-Rib-Deck joints; furthermore, using the envelope curve dataset increased the model’s validation accuracy by 21.33% compared to raw signals, demonstrating this method’s effectiveness in enhancing the SE-LSTM-CNN model’s ability to identify Lamb wave damage features. ④ The intelligent detection software developed based on MATLAB APP Designer achieved full-process optimization for damage detection, reducing errors from human intervention. This research is expected to provide a new technical solution for damage detection in steel bridge deck U-Rib-Deck joints.

    参考文献
    相似文献
    引证文献
引用本文

田亮,宋鹏飞,张海顺,等. 基于深度学习的钢桥面板U肋-顶板节点Lamb波损伤检测[J]. 华东交通大学学报, 2025,42(6):17-30.

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2025-05-27
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-01-15
  • 出版日期:
关闭