【作者】 邹万杰；

【导师】 瞿伟廉；

【作者基本信息】 武汉理工大学 ， 结构工程， 2009， 博士

【摘要】 随着社会的进步和经济的快速发展,土木工程结构的形式和功能日趋复杂,其安全越来越受到人们的重视。结构健康监测系统越来越多的安装到大跨桥梁、大型空间结构等大型土木工程结构中,但作为健康监测系统核心技术的损伤识别仍未从根本上得到解决,严重制约了结构健康监测系统的应用和发展。基于振动测试的结构损伤识别得到了广泛的研究和发展,人们提出了很多方法,这些方法在数值仿真分析及实验室试验中均取得了良好的识别效果。但由于实际的土木工程结构体积巨大,形式复杂,观测手段的限制等原因,通常难以获得这些传统的损伤识别方法所要求的测试信息及测试精度,因此,在已提出的结构损伤识别方法中能够成功应用到土木工程结构的损伤识别和健康监测中的例子尚不多见。结构损伤识别要在土木工程中走向实用还有很多问题需要解决,其中最主要的问题是需要解决测试信息不完备和测量噪声干扰条件下的参数识别及损伤识别问题。基于这样的背景,本文分别以杆系钢结构的轴向刚度和侧向刚度损伤识别为研究对象,提出了两种不同的损伤识别方法,即基于遗传算法的智能算法和基于改进残余力向量的解析算法,并在一些测试条件相同的benchmark结构算例中比较了它们的识别效果。本文主要研究工作分为六章:第一章是绪论。阐述了本论文的选题背景和目前结构损伤识别研究现状,并对本文的主要研究内容做了介绍。第二章提出基于频响函数和遗传算法的结构损伤识别方法。遗传算法是计算智能的一个重要组成部分,与常规的数学方法相比,它具有高度的自适应性、鲁棒性和并行性,因而对于包含非确定性和噪声的信息也有一定的处理能力,是一种典型的全局优化方法。如果结构输入(激励)已知,频响函数的测试简单且精度好,所包含的结构信息丰富。用测试频响函数及相应的计算频响函数构造遗传算法目标函数进行结构损伤识别,能够在考虑较高噪声水平情况下将结构的损伤识别出来。第三章在前人研究基础上提出基于残余力向量的杆系结构损伤直接识别法,克服了由残余力向量法一般只能得到结构损伤位置而无法识别其程度的缺点。该方法用刚度联系矩阵将损伤后结构刚度矩阵的变化矩阵展开,得到一个新的残余力向量方程,求解即可得到各单元刚度参数改变量的解析解,仅需一阶模态参数经一步计算就能够同时识别损伤的位置和损伤的程度。第四章研究了输出测量信息不足时结构的损伤识别问题。在基于残余力向量的结构损伤直接识别法中,当输出信息测量不足时无法得到完备的模态振型,完备的模态振型只能通过模态扩阶得到,而由扩阶模态进行结构损伤识别的效果与损伤处的测试信息完备程度有关。在基于遗传算法的结构损伤识别法中,提出直接由不完备的测试频响函数构造遗传算法的目标函数进行结构的损伤识别,在benchmark结构的算例中,识别效果良好。第五章研究了输入未知时结构的损伤识别问题。在噪声干扰下结构的模态参数识别误差较大,由此导致了依赖模态参数进行识别的残余力向量法的识别误差也较大,只有在很低水平的噪声条件下识别出来的结果才是可靠的。在基于遗传算法的结构损伤识别法中,在输入未知时无法获得结构的实测频响函数,如果输入的频谱平坦,近似为有限带宽的白噪声(比如脉动风),其功率谱为一常数,则结构的原点频响函数和自功率谱具有相似的形状。提出直接由测试的结构自功率谱和计算的原点频响函数来构造遗传算法的目标函数,进行结构的损伤识别,在benchmark结构算例中,获得了很好的识别效果。最后还讨论了在输入未知且输出测量信息不足时,仅以部分实测自功率谱和相应的计算频响函数构造遗传算法的目标函数,进行结构的损失识别,在算例中也能够获得很好的识别效果。第六章总结了本文的主要研究内容,指出尚需要进一步研究解决的问题并对将来的研究工作做了展望。更多还原

【Abstract】 With the quickly development of the societies and economy, the shape and function of the civil engineering structures become more and more complication, and the security of them is pained more attention to. More and more structural health monitoring(simplified in SHM) system are installed on the long-span bridge、large-scale space civil engineering, however, the key technique of the SHM system—damage identification has never been ultimately resolve, and it restricts the application and developing of SHM seriously.The structures damage identification based on vibration test was researched and developed widely, lots of methods were presented in past several decades, most of these methods can get a good results in the numerical analysis or in the experiment in laboratory. However, because the volume of the civil engineering is very huge, and the shapes are more complicated, observation means are restrict, and so on, the prerequisite test information of these methods are generally hard to get completely, the successful examples of damage identification and SHM of civil engineering structures are rare. The damage identification of civil engineering structures go to practical, there are a lot of problems need to be solved, most important of all, is to solve the parameter identification and damage identification in incomplete of the test information and the disturbing of test noise. In such background, two different methods of structures damage identification on steel member structures were presented in this dissertation, they are intelligent method based on genetic algorithm(simplified in GA) and analysis method based on residual force vector separately. In some examples of benchmark structure with the same test condition, the identification results were compared. This dissertation includes six chapters as follows:In the first chapter, the background of this dissertation are introduced firstly, Secondly, the structures damage identification based on vibration test is surveyed. Finally, the major research work of this dissertation is produced.In the second chapter, the structures damage identification methods based on FRF and GA is presented. GA is an important part of computation intelligence, compare with other mathematical methods, it has the characteristic of adaptive itself, robustness, and parallelism, so it can process the information with the uncertainty and noise, it is a typical global optimization method. if the input of structures is know, the testing of frequency response function(simplified in FRF) is simple and the precision is high, and it contains abundance information. The test FRF and the corresponding calculated FRF are used to construct the target function of GA, the damage of structures can be identified well in difference level noise in the examples.In the third chapter, based on the work of previers, a direct damage identification method of structures based on residual force vector is presented. Compare with the traditional method of residual force vector, the weakness that it can only identify the location of damage but not extent in generally directly is overcome. The stiffness matrix perturbation of the damaged structure is expanded by the stiffness connectivity matrix at first, then a new equation of residual force vector is got, it can be solve directly, so the analytical solutions of the stiffness parameter perturbation of elements are got. This method can identify the location and extent of structures damage at the same time only need one calculation step, and need only the first mode parameters of the damaged structures.In the fourth chapter, the damage identification of structures with insufficient output information is researched. In the direct damage identification method of structures based on residual force vector, when the output information is insufficient, the completed mode shapes can not be got, so the completed mode shapes can be got only by mode extend, the damage identification results based on extend mode shape depend on the complete degree of the test information of damage location. In the damage identification of structures base on GA, when the output information is insufficient, the insufficient test FRF is used to construct the target function of GA is presented, and GA is used to identify the structures damage. In the example of benchmark structure, the identify results is well.In the fifth chapter, the damage identification of structures with unknown input(excitation) information is researched. When the input is unknown, the errors of identification mode parameters are larger than known input in considering of test noise, and it lead to larger errors of damage identification results, in the example, the identification results is reliable only in considering a very low level noise. In the structures damage identification base on GA, when the input is unknown, the test FRF can not be got, if the spectrum of input is flat, and approximately to the white noise(such as fluctuating wind), it’s the power spectrum is a constant, so the origin FRF of structures are similar to the power spectrum in form. So the target function of GA constructed by testing power spectrum and corresponding calculating FRF is presented, and used to identify the structures damage. In the example of benchmark structure, a good identification results is got. At last, the case considering with the unknown input and output incomplete is researched, only parts of test power spectrum and corresponding calculating FRF are used to construct the target function of GA, in the example of benchmark structure, the good identification results are got too.The last chapter summarizes conclusions in this dissertation, pointes out the present problems in the damage identification of structures, and narrates the prospect for the research work of future in this field.更多还原