【作者】 沈东强；

【导师】 徐礼华；

【作者基本信息】 武汉大学 ， 结构工程， 2004， 硕士

【摘要】 工程结构随着使用时间的延长，不可避免的发生老化；自然灾害的频繁发生也导致工程结构产生损伤，如1995年的日本神户地震及1999年的台湾大地震使无数的房屋损坏和倒塌，造成人民生命财产的巨大损失。结构损伤的及时检测及修复对于减少生命财产损失具有重要的作用。同时，尽早的发现结构损伤，可大大降低维护、维修的费用。因此，对工程结构的实时健康诊断和安全评估是及其重要的。 结构发生损伤通常会导致结构动力特性的改变。近年来，基于结构动力特性参数的诊断和监测技术，由于其多种优点及便于实时诊断，已成为当前学术界和工程界的研究热点。同时，结构损伤识别中的神经网络方法受到了广泛的关注和研究。本文在武汉大学科技创新基金项目“建筑物安全监测与安全状态评估方法研究”的支持下，系统研究了基于动力学参数和神经网络的结构损伤识别方法，本文的工作归纳起来主要有以下几个方面： 1．从损伤力学的基本概念出发，对钢筋混凝土结构在材料、构件、结构三个层次上的损伤模型进行综述，介绍具有代表性的结构损伤模型，探讨其损伤发生、发展的机理。 2．对基于动力参数的损伤识别方法进行了系统的分析和总结。用动力学参数构造的损伤标识量往往需要求解复杂的数学反演问题，使得较难应用于实际工程中。鉴于神经网络以其优异的非线性映射能力可以使逆问题正问题化，将振动模态分析方法与神经网络技术结合起来，以振动模态参数构造损伤标识量，作为神经网络输入的特征参数，从而进行结构损伤检测。 3．结构损伤识别包含有无损伤判别、损伤定位及损伤程度的判断三个不同层次的问题。本文以两种形式的结构为例，尝试分阶段逐层深入地解决损伤识别问题。通过数值仿真分析，结果令人满意。 4．结构的动力特性与结构物理参数直接相关，结构的损伤将引起相应动力特性的改变。本文讨论了频率、振型、振型差等模态参数变化与损伤位置、损伤程度之间的内在联系。 5．实际识别过程中，不可避免的会存在误差影响。本文分析讨论了测量误差和模型误差对识别结果可信度的影响，数值模拟和理论分析表明:低水平的测量误差对识别结果影响有限;对于模型误差影响，文章提供了解决意见。 6.对国际上标准的结构健康监测Benchmark问题进行了研究，结果表明分阶段的识别有助于问题的简化和解决。 最后，总结了本课题的研究成果，并对后续的研究工作做了展望。更多还原

【Abstract】 Engineering structures are inevitably ageing with increased serviceability time while frequent occurrences of natural disasters are also causing damage in structures, such as Kobe earthquake in Japan (in 1995) and Taiwan earthquake (in 1999), which cause great damage in numerous houses and even total collapse, taking a heavy toll on people’s lives and their properties. The detection and repair of structural damage play an important role in greatly reducing the loss of lives and properties. Meanwhile, maintenance and repair costs can be lowered substantially if early recognition of structural damage is done. Therefore, real time health diagnosis and safety appraisal are most crucial for engineering structures.Dynamic characters usually may change due to structural damage. Recently, diagnosis and monitoring techniques based on characteristic parameters of structural dynamics have stood in the spotlight of research fields by academic and engineering society owing to their various advantages and convenience for real time diagnosis. In the meantime, artificial neural network in structural damage recognition has been widely heeded and researched. With the support of the fund for science and technology innovation of Wuhan university-Research on Safety Monitoring and Safety Condition Evaluation of the Building. We systemically study the method in damage identification based on modal parameters and artificial neutral network .The main contents are as follows.1. Based on the basic conception of damage mechanics, reinforced concrete -three level damage models are summarized. Representational damage models are also introduced, the mechanism of damage occurring and developing is discussed.2. Prevalent testing method based on dynamic parameters are analyzed and summarized systematically, whereas forming parameters needed often meet complicated reverse calculation problem .So the method is hard been used in practical application. With the strong non-linear mapping ability, artificial neural networks can make inverse problem into obverse one, so we combine modal analysis method with neural networks technique, taking modal parameters to construct identification parameters diagnostic vector as input vector for nondestructive testing.3. Generally, Structural Damage Identification include existence , location anddegree of damage-three level problem. Through two factual examples, we try to solve damage identification problem step by step. The method is effective with satisfactory result4. Dynamic characteristics of the structure are directly related to physical parameters, the relationship between frequency -. mode -. mode variance and damage location together with damage degree is discussed.5. During course of practical identification, actual errors inevitably impact the result. Study of the relationship between errors and reliability has shown that the influence of low-level measured errors is limited. As for model errors, a useful proposal is given.6. Nowadays, large numbers of research results are focusing on simple component such as beam ^ board and pillar. Benchmark problem in structural health monitoring was studied in this paper, and the results shows that identification by stages can be useful for simplification and solution to a question.Finally, some results are summarized in this project with concerning prospect for the further research.更多还原