加速度与计算机视觉多尺度滤波融合的桥梁位移响应监测方法  

Bridge displacement monitoring approach using multi-scale filtering fusion of acceleration and computer vision

作　　者：吴桐 周志祥 唐亮[1] Wu Tong;Zhou Zhixiang;Tang Liang(Chongqing Jiaotong University,Chongqing 400074,China;Shenzhen University,Shenzhen 518060,China)

机构地区：[1]重庆交通大学,重庆400074 [2]深圳大学,广东深圳518060

出　　处：《土木工程学报》2025年第4期58-74,共17页China Civil Engineering Journal

基　　金：国家自然科学基金(51708068、51778094);深圳市科技计划(JCYJ20220818095608018)。

摘　　要：为改善现有计算机视觉技术在桥梁结构动态位移监测精度、采样频率、便捷性等方面存在的不足,充分挖掘各类监测响应中的有效信息,提出一种基于视觉位移和加速度多尺度滤波重构位移的方法,可实现位移在时域和频域的精度提升。首先,推导位移重构过程在时域和频域的理论表达式,分析目标精度、时窗长度及采样频率的影响及取值依据,提出由视觉位移提供位移的准静力分量,由加速度提供位移的动力分量,二者通过互补滤波融合实现高精度位移重构;其次,由视觉位移和加速度中的同频带响应进行最小二乘拟合,可以在无需任何结构尺寸与测量参数条件下,快捷、自动计算视觉位移的比例因子。通过对实验室悬索桥缩尺模型进行两种工况的试验和户外两跨连续梁桥的动载试验,验证所提出方法对桥梁位移监测的可行性。模型桥试验中,以位移计实测位移为参考,融合位移的误差比视觉位移最大减小了约50%,比例因子计算误差小于2%;实桥试验中,比例因子计算与传统计算方法的误差小于8%。在两类试验中,相比视觉位移,融合位移均能够获取试验桥梁的高阶响应。研究结果表明:所提出的方法能得到高精度、宽频域和低噪声的位移,可推进计算机视觉在桥梁结构健康监测领域的进一步应用。To address the limitations of existing computer vision techniques in accuracy,sampling frequency,and convenience for bridge displacement monitoring,and to fully exploit the valuable information contained in the monitored physical quantities,a method is proposed based on multi-scale filtering reconstruction of displacement using visual displacement and acceleration.By applying this method,the enhanced accuracy of displacement reconstruction in both the time and frequency domains can be achieved.Firstly,theoretical expressions for the displacement reconstruction process in the time and frequency domains are derived,and the influences and value-taking criteria of target accuracy,time window length,and sampling frequency are analyzed.In this algorithm,visual displacement is utilized to provide quasi-static displacement components and acceleration is utilized to provide dynamic displacement components,which are then fused with complementary filtering to achieve high-precision displacement reconstruction.Additionally,a least squares fitting is performed on responses in the same frequency band from visual displacement and acceleration.This allows for the quick and automatic calculation of the scaling factor for visual displacement without requiring any structural dimensions or measurement parameters.Feasibility of the proposed method for bridge displacement monitoring is validated through experiments on a reduced-scale model of a suspension bridge under two scenarios and on a two-span continuous girder bridge under dynamic loads.In model tests,compared to visual displacements,the fused displacements reduced the error by approximately 50%when compared to displacement measurements obtained using displacement transducers.And the calculation error of scaling factor was less than 2%.In field tests on a real bridge,the error between scaling factor calculations and traditional methods was less than 8%.Additionally,in both of tests,the fused displacement could capture higher-order responses of the target bridges as compared to

关 键 词：桥梁工程 桥梁位移监测 计算机视觉 数据融合 互补滤波 比例因子 

分 类 号：U446.2[建筑科学—桥梁与隧道工程]