【作者】 吴大宏；

【导师】 赵人达；

【作者基本信息】 西南交通大学 ， 桥梁与隧道工程， 2003， 博士

【摘要】 桥梁建成通车后，由于受气候、环境因素的影响，结构材料会被腐蚀和逐渐老化，长期的静、动力荷载作用，使其强度和刚度随着时间的增加而降低。这不仅会影响行车安全，更会使桥梁的使用寿命缩短。在结构布局和规模都十分复杂的大型桥梁上仍然沿用传统的桥梁外观检查、养护、维修程序以及常规的局部检测，显然已难以全面反映桥梁的健康状况，尤其是难以对桥梁的安全储备以及退化途径作出系统的评估。建立和发展某种能够提供整体和全面的全桥结构检测和评估信息的监测系统，随时了解大桥结构的承载能力和安全储备，对保证大桥运营的安全性和耐久性都是十分必要的。 目前，在许多大跨度桥梁上都已经安装了永久性的监测系统，但目前的桥梁监测系统中不含结构模型，因而无自动损伤识别的能力。本文在前人研究工作的基础上，根据遗传算法和神经网络在处理复杂非线性问题时的各自特点，分别将其用于桥梁结构健康监测系统的不同部分，提出了建立基于遗传算法与神经网络的桥梁结构健康监测系统的基本设想。为此，本文主要进行了以下几个方面的研究工作。 1．在既有桥梁的研究中，主要包括桥梁总体性损伤评估、桥梁承载能力鉴定、桥梁剩余寿命评估、桥梁处置对策确定及其经济分析等，其中在日常管理养护中比较常用、比较重要的问题之一就是桥梁的总体性损伤评估。根据《中国公路桥梁综合评定方法》，引入数理统计中的“正交试验法”和“均匀设计法”，应用神经网络对桥梁结构进行模糊综合评判。 2．应用神经网络较强的模式分类功能，选择合适的模态信息或动静力响应数据，对桥梁结构进行损伤识别。结合基于静态应变及位移测量的结构参数识别算法，借助遗传算法强大的优化搜索能力，探讨基于遗传算法的桥梁结构损伤识别技术。 3．根据安装在桥梁结构上的监测系统传递来的位移、应变信息，进行作用在桥梁结构上的荷载识别，可以：① 对监测系统传递来的荷载信息加以校核和补充；② 为下一步的损伤识别提供依据。而荷载识别的关键技术在于快速、准确地模拟桥梁的荷载-挠度曲线。以“在位移控制下部分预应力混凝土梁非线性行为试验研究”的试验数据为建模样本集，分别以作用于梁上的荷载和梁的跨中挠度为输入、输出，利用神经网络来对混凝土梁的荷载-挠度曲线进行模拟，并用其进行荷载识别的试验研究，取得了较好的效果。第11页西南交通大学博士研究生学位论文 4.裂缝开展宽度是衡量部分预应力混凝土桥梁使用性和耐久性的一个重要指标，精确地模拟部分预应力混凝土梁的荷载一裂缝关系有着十分重要的意义。然而，混凝土结构裂缝形成和发展的过程十分复杂，具有一定的随机性，难于用常规的方法进行建模。应用神经网络较强的函数映射能力和联想、记忆功能，对部分预应力混凝土梁的荷载一裂缝关系进行建模，通过试验数据进行验证，效果良好，证明应用神经网络来对部分预应力混凝土梁的荷载一裂缝关系进行建模是可行的。 5.混凝土结构中钢筋的锈蚀将直接影响其安全性、可靠性和耐久性，快速、准确地预测混凝土构件中的钢筋锈蚀量有着重要的意义。应用神经网络对锈蚀开裂后的混凝土构件中的钢筋锈蚀量进行预测建模，并通过工程检测结果验证了该建模方法的可行性。 6.应用遗传算法强大的非线性搜索能力，分别以年均投资差额和总投资差额最大为优化分析目标，提出了用遗传算法对常遇大气环境下的混凝土桥梁进行耐久性优化设计的方法，通过算例进行验证，效果良好。 7.将桥梁结构健康监测系统监测到的结构响应看作一个非线性时间序列，应用神经网络较强的数据压缩能力和联想、记忆功能，选取若干天的结构响应数据(应变一时程曲线等)进行学习，并对未来一段时间内的结构响应进行预测。以此为根据，对桥梁结构进行疲劳可靠性分析。 8.由于几何非线性和材料非线性的影响，构件某一失效模式的极限状态方程很可能是强非线性方程，甚至无法得到显式的极限状态方程。工程实际中常采用的传统方法难以精确或有效地对桥梁结构进行可靠性分析。根据可靠度指标刀的几何意义，将求解可靠度指标刀的问题转化为有约束最小化问题，应用遗传算法强大的非线性搜索能力进行求解。更多还原

【Abstract】 After bridges having been constructed and opened to traffic, their material will be deteriorated or aged gradually because the influence of the weather, environmental factors and their strength and stiffness will degrade with the time running for the action of the static and active loads applying on them. Not only will this endanger the safety of the traffic, but also it will shorten the life span of the bridge. Apparently, for bridges with a complex layout and large scales, it’ s difficult to hold all the information that can reflect the condition of the bridges globally and more difficult to evaluate their safety factors and tracks of deterioration systematically by the traditional ways of visual bridge inspection, maintenance or local inspection. Therefore, to assure the security and durability of the bridges, it is necessary to construct a health monitoring system that can provide all the information reflecting the condition of the bridges globally at any time needed.Nowadays, Structural Health Monitoring Systems have been installed on many long span bridges permanently. However, they do not have the ability of automatic damage detection because of no FEM model existing. Based on the achievements of the forerunners, a model of the Bridge Health Monitoring System is suggested by using Genetic Algorithms and Neural Network in different parts of the system on the basis of their characters while dealing with complex nonlinear problems. Therefore, the following research work has been carried out in this doctoral dissertation.1. In the research of the existing bridges, some problems need to be solved immediately such as how to evaluate the condition of a bridge globally, which is the best procedure to evaluate the strength of a bridge, how to evaluate the residual life of a bridge, how to maintain a deteriorated bridge and how to carry out an economy analysis. During the regular inspection and maintenance, the most important thing is to evaluate the global condition of a bridge correctly and timely. Taking<<Synthetic Method of Evaluating Highway Bridge in China> as a reference and making use of the Orthogonal Design Method and the Uniform Design Method in statistics, a forecast model is suggested to evaluate the global condition of a bridge based on Neural Network.2. Combined with some approximate modal information and response of the bridges in their static or active mode, a procedure for damage detection of bridges is suggested based on Neural Network because of its stronger mode clustering ability. On the other way, combined with some methods of parameter identification based on strain or displacement information coming from static measurement, another procedure for damage detection of bridges is suggested based on Genetic Algorithms because of its strong nonlinear searching ability.3. By identifying the load condition of a bridge with the information of strain and displacements from the Bridge Monitoring System stored on it, one can, first, replenish and check the information from the monitoring system, and second, provide information for the damage detection in the next step. The key point of load identification is to model the relation between the load and the deflection caused by load correctly and quickly. Utilizing the data from a research on the nonlinear behavior of a partially prestressed concrete beams controlled by displacement as samples and taking the loads applied on the beams and the deflection at the mid-span as input and output for the model respectively, a model of load and deflection is suggested and used to identify the load condition of concrete bridges based on Neural Network in the dissertation and the results are satisfactory.4. Crack width is one of the most important indexes to indicate the reliability and durability of a partially prestressed concrete bridge. It is very important to precisely model the relationship between the crack width and the applied load of a partially prestressed concrete beam. However, the process of the formation and development of the cracks on a concr更多还原