【作者】 冉志红；

【导师】 李乔；

【作者基本信息】 西南交通大学 ， 桥梁与隧道工程， 2007， 博士

【摘要】 近年来，结构损伤识别在土木和航空等领域迅速发展起来，但大量的损伤识别方法面对大型结构时却遇到了许多实践上的困难，尚有许多问题值得深入研究。在诸多结构损伤识别方法中，基于动力测试数据的识别方法有远程在线的潜在优势，因此动力损伤识别方法受到了极大的关注。本文首先对结构损伤识别的研究现状、研究的主要难点进行了分析。然后针对结构动力损伤识别中动力指纹的基本问题进行了系统研究，主要完成了以下六个方面的工作。1．对于目前损伤识别领域所用到的动力指纹进行了概括和分类。将动力指纹分为四个大类：传递特性类动力指纹、复杂函数类动力指纹、传递曲率类动力指纹和特征参数类动力指纹。逐一对这些动力指纹进行分析研究，对每种动力指纹的计算方法、适用范围进行了系统的总结。2．研究了几种常用动力指纹抵抗干扰的能力。首先对模态参数的噪声来源进行剖析，并据此给出其理想化的噪声分布公式。按概率统计相关理论推导了动力指纹的噪声分布，并得出了一些具有指导意义的结论。本文还基于系统灵敏度分析理论，推导了结构的损伤灵敏度理论公式。在动力指纹噪声灵敏度和损伤灵敏度分析的基础上，提出了衡量动力指纹抵抗干扰能力的综合性能评价指标，为特征比选提供了量化的参考依据。3．利用模式识别中特征提取和特征选择的基本理论，将动力指纹库进行适当的变换，得到了具有良好可分性的优化特征库。主要对基于相关分析、遗传算法的特征选择和基于主成分分析、Fisher判别分析、三谱相辅分析的特征提取等方法开展研究。并提出了针对结构损伤识别问题的特征向量优化策略，为工程的实际应用提供了有效的实施方案。4．从模式分类的角度对结构的损伤位置和损伤程度进行识别，其核心问题就是如何合理设计分类器。本文研究了近邻法、判别分析、人工神经网络(BP神经网络、RBF神经网络)三种分类器的具体算法，并分析了它们在结构损伤识别中各自的优势和局限性。5．多重子步法是解决复杂结构损伤识别问题的有效途径，对结构进行区域划分是多重子步法的基础工作。本文将模糊关系聚类和模糊C均值聚类用于损伤区域的划分中，提出了基于优化动力指纹库的软分区方法。这相当于对优化特征库进行重组，使动力指纹从特征和模式两个不同角度均进行了优化，降低了损伤定位的难度，提高了损伤识别的精度。6．研究了连续梁桥损伤识别过程中基本特征库的构建、灵敏度分析、特征库的优化、分类器的设计等相关问题的程序实现方法，并以MATLAB为平台编制相应的计算程序。以某三跨混凝土连续箱梁桥为例，验证了本文理论的正确性和可行性。更多还原

【Abstract】 In recent decades, the development of structural damage identification approaches has been accelerating in the fields of civil and aerospace structures. But most of the approaches faced with a number of practical challenges when applied to large structures. There are many problems required to be solved. Because the damage identification based on dynamic test data has latent virtues of long-range and on-line, the dynamical damage identification attracts great attention. In the first content of this thesis, the state-of-art of structural damage identification is depicted. And then the foundational issues of dynamic fingerprints are researched systemically. The main research work is summarized as follows:1. Dynamic fingerprints in the field of damage identification are generalized and classified. They are grouped into four categories: transfer property dynamic fingerprints, complex function dynamic fingerprints, transfer curvature dynamic fingerprints, and character parameter dynamic fingerprints. The computation methods and applicability of dynamic fingerprints are summarized systemically.2. The anti-jamming capability of dynamic fingerprints is researched. At first the noise source of mode parameters is analyzed, and the idealization noise expressions are brought forward. On the basis of probability theory, the noise expressions of dynamic fingerprints are deduced. The damage sensitiveness of dynamic fingerprints is researched based on system sensitiveness theory. In order to provide the definite quantity index for feature selection, the comprehensive performance evaluation criteria for anti-jamming capability of dynamic fingerprints is defined.3. According to feature selection and feature extraction technology in the filed of pattern recognition, the foundation feature storage is transformed with the proper means into the optimization feature storage. And the severability of the optimization feature storage will be improved. The pertinence analysis, genetic algorithm, principal component analysis, fisher discriminant analysis and three spectra analysis technology are researched for feature selection and feature extraction. And the optimization strategies of the foundation feature storage are advanced and provide the effective implementation for the practical application. 4. The structural damage location and damage extent are identified based on pattern classification. Near neighbor method, discriminant analysis method, artificial neural networks (Back-Propogation neural networks, Radial Basis Function neural networks) are researched, and their advantages and disadvantages are analyzed.5. The multistage damage identification method is an effective ways for complex structures, and sub-area’s partition is a foundational work. According to fuzzy relation clustering and fuzzy C means clustering, the soft sub-area’s partition method is advanced based on the optimization feature storage. It is corresponded with recombining of the optimization feature storage. This method optimizes dynamic fingerprints from feature and pattern, and reduces the difficulty and heightens the precision of structural damage identification.6. The plan of program design for structural damage identification of continuous girder bridge is researched. The main content includes building of the foundation feature storage, sensitiveness analysis, optimization of the feature storage, design of classification methods and so on. The corresponding software system is developed based on MATLAB. Finally, the correctness and feasibility of the presented methodology are demonstrated by a three spans continuous box-girder bridge.更多还原