基于多头卷积自编码器的桥梁结构信号重构与损伤识别方法研究  

Signal reconstruction and damage identification methods for bridge structures based on multi-head convolutional autoencoders

作　　者：陈鑫婷 张军 鲁东明 应柳祺 李强 CHEN Xinting;ZHANG Jun;LU Dongming;YING Liuqi;LI Qiang(School of Civil Engineering and Architecture,Zhejiang Sci-Tech University,Hangzhou 310018,China;School of Civil Engineering,Ningbo Tech University,Ningbo 315100,China;College of Civil Engineering and Architecture,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]浙江理工大学建筑工程学院,杭州310018 [2]浙大宁波理工学院土木建筑工程学院,浙江宁波315100 [3]浙江大学建筑工程学院,杭州310058

出　　处：《振动与冲击》2025年第6期298-305,共8页Journal of Vibration and Shock

基　　金：浙江省尖兵领雁研发攻关计划项目(2023C03183);浙江省自然科学基金(LY23E080005);宁波市自然科学基金资助项目(2023J041)。

摘　　要：针对传统监督学习需要大量的标记损伤数据问题,基于多头卷积自编码器建立了桥梁结构振动信号的重构方法,使用基于均方误差的损伤评估指标分析拱桥结构和梁桥结构振动信号重构的有效性以及不同损伤状态下的变化规律。结果表明:多头卷积自编码器在重构振动信号及其后续的损伤识别方面精度优良,多头一维卷积结构在损伤检测精度和灵敏度上优于传统的一维卷积结构;通过拱桥有限元仿真数据与连续梁桥损伤实测数据进行了方法验证,发现该方法能够准确地识别出桥梁结构的损伤发展趋势,在噪声环境下也具有较好的鲁棒性,可为桥梁结构健康监测数据分析提供参考。To address the challenge that traditional supervised learning methods require a large amount of labeled damage data,a reconstruction method was proposed for bridge structure vibration signals based on a multi-head convolutional autoencoder.The effectiveness of the vibration signal reconstruction for arch bridges and beam bridges,as well as the variation patterns under different damage states,were analyzed using a damage assessment metric based on the mean square error.The results indicate that the multi-head convolutional autoencoder achieves high accuracy in both signal reconstruction and subsequent damage identification.The multi-head one-dimensional convolutional structure outperforms traditional one-dimensional convolutional structures in terms of damage detection accuracy and sensitivity.The proposed method was validated through finite element simulation of arch bridges and damage measurement data of continuous beam bridges,demonstrating its ability to accurately identify the damage development trends of bridge structures.Furthermore,the method exhibits robust performance in noisy environments,providing a valuable reference for the analysis of structural health monitoring data in bridge engineering.

关 键 词：桥梁工程 多头卷积自编码器 振动响应 信号重构 损伤识别 

分 类 号：TU317.9[建筑科学—结构工程]