【作者】 祁泉泉；

【导师】 辛克贵；

【作者基本信息】 清华大学 ， 土木工程， 2011， 博士

【摘要】 本文采用理论分析、数值模拟和试验研究解决了结构健康监测领域中结构参数识别系统目前存在的一些关键问题,主要包括无线传感技术和识别算法。本文的主要研究工作和贡献如下：（1）设计了一种可以应用在结构参数识别系统振动信号采集的基于磁制爬行装置的移动无线传感器网络，主要包括构造设计，硬件和软件设计。进行了实验室门式刚架锤击试验，试验表明用移动无线传感节点采集的数据与静态传感器采集的数据基本一致。利用移动无线传感器网络采集的加速度数据识别门式刚架的模态参数，与有限元结果进行对比分析，表明本文设计的移动无线传感节点可靠，可以提供灵活的空间运行方式，为解决目前静态传感器布置成本高、耗能高等问题提供了可行性。（2）提出并推导了用于结构模态参数识别的EERA算法，该算法可以处理强迫振动响应，具有更高的精度和稳定性，尤其可提高基频的精度，可处理加速度、速度或位移时程数据，建议采用加速度时程识别结构的高阶模态。解决了HHT变换中存在的模态混叠和端点效应，推导了利用HHT变换计算比例阻尼结构体系的振型公式。分别用上述两种方法处理了数值模拟和实验室框架结构的振动台试验数据，模态参数识别精度好，鲁棒性强。（3）提出了IBB法，当精确质量阵为对角阵时，可以释放BB法的一个约束，解决了非质量归一化振型的问题，该方法获得的修正模型可以保持明确的物理意义和带状特性。利用GA算法最小化4种目标函数识别结构物理参数，结果表明基于GA算法的优化修正方法对目标函数有一定的敏感性，当损伤类型未知时，建议使用本文提出的组合目标函数以增强识别算法的精度和鲁棒性。通过DMU方法解决了不完备复模态的模型修正问题并研究了噪音的影响，该方法是一种理论上完备的算法，可保持模型的物理特性并降低计算量，对添加的质量块数量和重量要求不高，噪音影响可视为矩阵扰动问题。（4）提出了IUKF滤波，该方法利用SVD分解获取计算过程中的方差矩阵均方根，引入噪音形成增广状态变量，重新生成sigma样点。试验表明基于概率分布的IUKF可识别物理非线性较强的时变结构物理参数，可增强算法的鲁棒性和稳定性，提高识别结果的精度和抗噪能力。更多还原

【Abstract】 The key issues of structural parameter identification system in health monitoring,mainly including the fields of wireless sensing technology and identification algorithms,are studied via theoretical analysis, numerical simulation and experimental research inthis thesis. The main research work and contribution are as follows.(1) The mobile wireless sensor network (MWSN) aided by the magnet-wheelclimbing robot is developed to collect vibration signal for structural parameteridentification system. The design of MWSN consists of mechanical design, hardwareand software design. The hammer impact experiment of steel portal frame is carried outin the laboratory. The difference between acceleration data collected separately fromMWSN and static sensor is small. The comparison of frame’s modal parameteridentified separately from time history of acceleration data collected by MWSN andfinite element method shows that the MWSN could offer reliable, flexible and adaptivespatial resolutions for SHM that can overcome the limitation of high cost of sensors,limited power supply of current sensor networks.(2) Extended Eigensystem realization algorithm (EERA) is proposed and derivedfor structural modal parameter identification, which could be used for the measuredforced response. Any kind of time history of acceleration, velocity or displacement isavailable for EERA. The accuracy and stability of EERA are demonstrated, and it’ssuggested to use the time history of acceleration to identify higher modes of structures.The problems of mode mixing and endpoint effect in Hilbert-Huang Transform (HHT)for modal parameter identification are solved by adopting Chebyshev I type filter andremoving endpoints. And the format of mode shapes of proportional-damping-ratiostructure is achieved. The accuracy and robustness of the two proposed algorithms aredemonstrated by numerical simulation and shaking table test of frame.(3) The improved Berman-Baruch method (IBB) is proposed and derived toidentify structural physical parameters for finite element model updating. One constraintof Berman-Baruch method that the normalized modes with mass matrix are requiredcould be released with the diagonal mass matrix. The updated model via IBB is morephysical meaning and keeps the characters of band and sparse of system matrices. The structural physical parameters under3kinds of different damage scenarios are identifiedseparately by minimizing4kinds of different objective functions by genetic algorithm(GA), the comparison among those sets of identified parameters shows model updatingbased on GA is sensitive to the chosen objective functions, and it’s suggested to use thecombination objective function proposed in this thesis in order to obtain the accurateand robust detection when the kind of structural damage is unknown. The DifferenceModel Updating (DMU) is presented by adding the known blocks to the structure andits characteristics are concluded that it’s one kind of complete algorithm for modelupdating, the physical properties could be preserved, the computational cost could belower, the qualification of added mass block is easy to satisfy, and parameteridentification based on incomplete set of complex modes also can be achieved whileidentification based on noise-contaminated modes seems inefficiency.(4) The Improved Unscented Kalman Filter (IUKF) is proposed and derived by thefollowing two changes, using the Singular Value Decomposition (SVD) to format thesquare root of covariance matrix, and bringing the noise vector to state vector to formaugmented state vector and to regenerate the sigma sample points. Numerical simulationshows IUKF is capable of detecting the time varying history of nonlinear hystereticstructural physical parameters, and improving the robustness, stability, accuracy andnoise-immunity.更多还原