【作者】 张文斌；

【导师】 王祁；

【作者基本信息】 哈尔滨工业大学 ， 仪器科学与技术， 2009， 博士

【摘要】 车辆动态荷载和车型分类一直是高速公路研究工作者研究的热点和难点问题。随着国民经济的快速发展,国家对高速公路建设费用的投入逐年加大,使得我国高速公路里程数已经跃居世界前列,同时国家因每年道路破坏造成的损失也高达几百亿元。传统的车辆动态荷载测量系统（即动态称重系统）由于系统造价高、获取信息少以及称量速度慢很难实现不间断测量等缺点,难以满足公路对车辆荷载监测和车型分类的要求。研究一种能够快速、准确地测量车辆轴重和总重并能对车辆进行分类的动态称重系统具有重要的意义。本课题得到了黑龙江省交通厅基金（No:HJZ₂004₁2）的资助,旨在研究一种廉价的能够满足高速公路对车辆荷载监测和车型分类要求的动态称重系统,为重载车辆对刚性路面破坏作用研究提供客观、准确的车辆荷载及车型分类数据。基于此目的本文在分析大量的实验数据的基础上,提出一种新颖的基于路面结构应变响应的车辆荷载测量和车型分类技术,并就若干关键技术进行深入研究,论文具体的研究工作如下:1、论文在全面了解国内外动态称重和车型分类技术研究现状的基础上,分析了现有的动态称重方式和车辆分类方法存在的问题。2、针对路面结构的力学特性,研究了各种移动荷载下刚性路面的响应模型。用三维傅立叶变换方法分析了移动两轴常荷载和谐波荷载下刚性路面的应变响应,建立了荷载、应变响应与移动车速之间的明确数学关系。为基于刚性路面应变响应的动态荷载测量的研究奠定理论基础。3、在研究影响动态称重的精度和称量速度主要因素的基础上,指出车辆振动是影响动态称重的效率和精度的主要因素,提出利用等间距分布的多传感器进行多点测量的方式提高动态称重精度和效率。从理论上给出了等间距分布多传感器的设计方法和设计公式,利用此方法和公式设计和建立了基于多个埋入式应变传感器的动态称重系统原型。4、在获取大量的实验车辆样本数据的基础上,研究了基于多分类支持向量机（MSVM）的多传感器融合分类方法,提出了两种融合分类策略,并用实际数据对两种不同MSVM算法（OAA,OAO）进行了融合分类实验,分别比较了不同算法和不同融合策略对分类准确率的影响。实验数据表明车辆分类准确率超过94.6%。5、对多传感器动态称重系统进行了实际标定实验,并研究了基于支持向量回归和熵权多传感器融合建模方法。并用此方法解决了三个应变传感器动态称重系统建模标定的问题。实验数据测试表明模型可以获得6%以内的动态称重预测误差。更多还原

【Abstract】 Vehicle dynamic loads and vehicle classification has been a research hotspot and difficult problems for highway researchers. With the rapid development of the national economy, the state highway construction cost inputs to increase year by year, which leads our country highway mileage in front of the word. At the same time the costs for the damage of the roads is also as high as several hundreds billions of RMB each year. Due to the disadvantage such as high costs, less information gaining and slow weighing measurement, the traditional vehicle dynamic loads measurement system （ie, weigh in motion, WIM） is hardly to meet the requirements for highway vehicle load monitoring and vehicle classification. Thus it is very important to study a dynamic weighing system for increasing the accuracy and feasibility in weighing the vehicle axle loads and classifying the vehicles. This research has been aided by Heilongjiang Communications Fund （No: HJZ₂004₁2）. The purpose for the research work is to find a cheap way to meet the requirements for vehicle load monitoring and vehicle classification in the dynamic weighing system. It can provide objective and accurate vehicle load and vehicle classification data for heavy-duty vehicles damaging effects of rigid pavement research. For this purpose, according to the analysis of abundant experimental data, a novel dynamic weighing classification models based on strain response of pavement structure has been provided. According to these technical characteristics, a number of key technical issues were carried out, which was listed as follows:R&D of the WIM and vehicle classification in the world was reviewed in detail. Consequently, the existing problems in WIM and vehicle classification were pointed out.According to the mechanical properties of rigid pavement structure, the rigid pavement response model under different moving load has been studied. The response of the rigid pavement under the two-axle moving loads of constant amplitude and harmonic loads using a triple Fourier transform were simulated and the relationship between loads, strain response and speed were derived. These results may lead a theoretical foundation for studying on rigid pavement strain response dynamic loads measurement technology.This dissertation studied the main factors affecting the accuracy and speed of dynamic loads measurement; pointed out that the efficiency and accuracy of dynamic loads measurement was mainly affected by vehicle dynamic vibration; presented that multi-sensor of multi-point measurement methods can improve dynamic weighing accuracy and efficiency. Furthermore, the design methods and formulas of multi-sensor spacing system are introduced. Dynamic loads measurement system prototype based on the multiple embedded strain sensors was designed and set up using this method and formula.On the basis of many experimental vehicle sample data, Multi-sensor vehicle classification methods based on two multi-class Support Vector Machines （OAA, OAO） and two data fusion are researched in this dissertation. Different algorithms and different fusion strategies on classification accuracy using the experimental data show that vehicle classification accuracy can be reached more than 94.6 percent.Multi-sensor dynamic loads measurement system calibration using the actual experiment and multi-sensor fusion modeling method based on support vector regression and entropy weight were researched in this dissertation. Experimental data show that the model can be tested within 6% dynamic loads measurement prediction error.更多还原