【作者】 安永辉；

【导师】 欧进萍； Billie F.Spencer,Jr.；

【作者基本信息】 大连理工大学 ， 土木工程、防灾减灾工程及防护工程， 2013， 博士

【摘要】 基于振动信息的结构损伤识别通过对安装在结构上的传感器测得的振动响应信号采用某种算法进行分析来发现、定位、甚至定量损伤,具有可在线实时监测、快速定期检测、节省人力等优点；自20世纪70年代末期以来受到了广泛关注并取得了很大进展,已经提出了很多种结构损伤识别方法,但还面临着受噪声干扰大、对结构数值模型精确度有较大依赖、对损伤的敏感性有待提高等挑战。本文主要研究基于振动信息的结构损伤识别(损伤定位和损伤定量),重点发展不需要结构有限元模型(或对模型精确度依赖性低)、对环境和测试噪声不敏感且对结构损伤敏感的损伤识别方法,主要研究内容如下：(1)建立了适用于结构精细化和稀疏测点下损伤定位的SDLV法观测矩阵C的选择方法,从而提高了SDLV方法损伤定位的准确度。从SDLV法的核心(未知输入下Q矩阵的构造)入手,通过比较不同测点数比例下观测矩阵C对Q矩阵的影响,发现不同测点数比例时基于不同C矩阵构建的Q矩阵与损伤识别结果相关；为了减少传感器的使用数量,用力平衡的概念研究给出了基于SDLV法的两种常见桁架结构稀疏测点布置方案,实现了测点数量经济性和损伤识别精细化程度之间的平衡；为了实现对桁架结构关键杆件上局部损伤的定位,提出了基于SDLV法的结构关键构件精细化损伤定位方法和步骤；通过两种常见的桁架结构对以上内容进行了模拟和试验验证。(2)提出了基于比例柔度矩阵的LU(QR)分解的结构损伤定位方法。柔度矩阵可以仅通过前几阶低阶模态参数构建,但需要己知结构输入(或至少在一个测点上已知输入)来获得质量归一化振型,这在实际中往往难以做到,本文采用结构比例柔度矩阵及其LU(QR)分解的结果,针对不同的结构对象构建不同的损伤指标进行损伤定位,并通过剪切型集中质量框架模型和桁架结构模型进行了模拟和试验验证。(3)提出了四种目标函数和相应的结构损伤程度识别方法。运用以上两种方法可以发现并定位损伤,但未涉及损伤定量。针对损伤定量问题,选择损伤单元及其所在小范围的子结构脉冲响应和低阶振型为目标向量,借助有限元模型修正的方法,分别采用相关系数和模态保证准则分析模型修正结果中对应向量与目标向量之间的相似度,进而提出了四种结构损伤程度识别的目标函数和识别方法,并在简支梁模型和桁架结构模型上进行了模拟和试验验证。(4)研究／提出了五种基于振动信息的损伤特征函数：分形维数、近似熵、对数加速度能量、离散度、急动度能量等；并且采用曲率差和概率平均的思想,分别提出了相应于每种损伤特征函数的平均归一化曲率差和归一化曲率差概率的结构损伤定位方法；在几个不同类型的结构仿真和试验模型上,分别进行了不同噪声水平下的模拟验证和试验验证。此外,基于以上几种损伤特征函数和进一步提出的“修正的对数加速度能量”损伤特征函数,研究了大跨度桥梁结构吊杆和斜拉索的损伤定位。结果表明：这些方法对损伤敏感、抗噪声能力强、不需要结构数值模型,避免了参数识别和建模过程中的误差。本文方法适用于基于实时监测(基于环境激励)或定期检测获得振动信息的结构损伤识别,验证模拟和试验的结构类型有：剪切型集中质量框架模型、梁结构模型、桁架(网架)结构模型、悬索桥结构算例、拱桥结构算例、斜拉桥结构算例等。更多还原

【Abstract】 The vibration-based damage identification technique can identify, localize and quantify the damage of structures through analysing vibration responses with intelligent algorithms; it can be used in real time monitoring and fast periodical detection whenever necessary, also it can save manpower. Since the late1970s, vibration-based damage identification methods have received considerable attention and many methods have been developed. However, it still suffers from some limitations, such as high sensitivity to noise, high-dependence on accuracy of structural numerical models and low sensitivity to damage; challenges still remain.The present work is about the study of vibration-based structural damage identification methods (damage localization and damage severity identification), and the main purpose is to develop some new damage identification methods which are model-free, robust enough against the noise and also sensitive to the small damage. The research contents are listed in detail as follows:(1) A guidance has been provided regarding the conditions under which the respective formulations of the observation matrix C should be used, which results in the higher accuracy of the damage identification. The present work has studied the SDLV method from its key (i.e. the extraction of Q matrix without the excitation information) to the influence of different C matrices on Q matrix. Additionally, several strategies of sensor layout to achieve the effective performance with the SDLV method with limited measured nodes of two common truss structures have been explored based on the force balance method. The present work has also proposed the precise damage localization based on the SDLV method for local damage on some key truss members. Experimental and numerical validation of these points have been achieved based on two common truss structures.(2) The present work has proposed a damage localization method based on LU(QR) decomposition of the proportional flexibility matrix. The modal flexibility matrix can be extracted only with the first several modal parameters. However, the mass-normalized mode shapes can be achieved when input excitation (or at least at one measured point) is known, which is usually not available in practice. The proportional flexibility matrix is used and different damage indices based on its LU(QR) decomposition are used for various types of structures, and it has been validated based on a shear building model and a truss model. (3) The above two methods have been only designed as damage localization methods, damage quantification has not been considered. Here, four cost functions and the corresponding damage severity identification methods have been proposed. Pulse responses/the first several mode shapes at the damaged elements/and the substructure including the damaged elements have been selected as objective vectors; correlation coefficient and modal assurance criterion have been selected as mathematical tools to describe the similarity between the objective vectors and the corresponding vectors in the updated finite element model. Finally, finite element model updating-based four cost functions have been validated using a beam model and a truss model through experiments and simulations.(4) The present work has studied/proposed five structural damage features:fractal dimension, approximate entropy, logarithmic acceleration energy, degree of dispersion and jerk energy. Moreover, the curvature method, mean and probability method are combined to deal with the proposed damage features. As a result, two methods, i.e. the mean normalized curvature difference method and the curvature difference probability method for every damage feature are developed. Their feasibility has been validated in several types of laboratory models with numerical simulation and experiments with different noise levels. Moreover, damage localization of suspenders (hangers) and cables in the long-span bridges have been studied based on the above five and another new (i.e. the improved logarithmic acceleration energy) damage features. The results show these methods are sensitive to small damage, robust enough against the noise, and these methods do not require the finite element model of the measured structures, which avoid the error in the process of parameter identification and finite element modeling.The proposed methods in this work can be used in the vibration-based damage identification for real-time health monitoring or periodical detection. The research objects refer to the following numerical examples and laboratory models:a shear building model, a beam model, a truss model, a suspension bridge numerical model, an arch bridge numerical model and a cable-stayed bridge numerical model.更多还原