基于机器视觉和最优组合应变影响线的车辆荷载识别方法研究

Vehicle Load Identification Based on Machine Vision and Influential Line of Optimal Strain Combinations

为方便快捷地识别正交异性钢桥面板箱梁桥的车辆时空信息,实现不依赖车辆横向位置精确识别车辆荷载,提出一种基于机器视觉和遗传算法优化组合应变影响线的车辆荷载识别方法。该方法首先通过桥梁侧面安装的摄像头采集车辆视频流数据,基于机器视觉技术识别车速、车轴数和轴距等信息;然后采集U肋下缘应变响应,利用已知轴重的标定车辆识别测点的应变影响线;再基于变异系数指标采用遗传算法优化确定对荷载横向位置不敏感的最优组合应变影响线;最后进行车辆荷载识别。以某正交异性钢桥面板箱梁桥为背景,开展数值模拟和缩尺模型试验,研究不同车型、车重、横向位置和噪声水平等多种工况下所提方法的识别效果,验证方法的有效性和抗噪声性能。结果表明:该方法对车速、车轴数和轴距识别具有较好的精度和稳定性;基于变异系数优化后的组合应变影响线对车辆横向位置不敏感,可在不预估车辆横向位置的情况下有效地识别车辆荷载;数值模拟的轴重和总重最大识别误差分别为5.57%和4.03%(考虑20%噪声),模型试验测试的轴重和总重最大识别误差分别为7.16%和4.90%,该车辆荷载识别方法具有较好的适用性和准确性,便于工程应用。

This paper presents a vehicle load identification method based on machine vision and influential lines of strain combinations optimized by generic algorithms,which can rapidly identify spatio-temporal information of vehicles on orthotropic steel box girder bridge with ease and achieve accurate vehicle load identification without referencing the exact transverse locations of vehicles.First,traffic flow data are collected by the cameras installed on the lateral surfaces of the bridge,based on the machine vision technology,vehicle speed,number and spreads of axles are identified.Second,the strain responses at lower edges of U ribs are collected,and the already-known axle weights of trains are used to mark the strain influential line of vehicle identification points.Third,considering the indexes of variation coefficient,the generic algorithms are used to determine the influential line of optimal strain combinations that are not sensitive to transverse loading locations.At last,the vehicle loads are identified.Numerical modelling and scale-down model test were carried out,with an existing orthotropic steel box girder bridge as a case,to study the identification effects of the proposed method under multiple loading issues,including different types of vehicles,vehicle weights,transverse location and noise levels,and verify the efficiency and noise resistance of the method.It is shown that the method can identify vehicle speed,number and spreads of axles with great accuracy and stability.Following the principle that the influential line of strain combinations optimized by variation coefficient is not sensitive to vehicle transverse locations,the vehicle loads can be effectively identified without pre-estimating the transverse locations of vehicles.The identification bias of axle weight and total weight in numerical simulation are 5.57% and 4.03%(considering 20% noise),respectively,while 7.16% and 4.90%,respectively in model test,proving that the presented vehicle load identification method has high applicability and accuracy,and is easy for engineering application.