【作者】 吴金志；

【导师】 张毅刚；

【作者基本信息】 北京工业大学 ， 结构工程， 2005， 博士

【摘要】 正如人会得病一样,建筑结构在设计、施工尤其是使用过程中,必然要遭受人为因素和自然因素的影响而出现老化或破损,严重者会导致生命和财产的重大损失。对建筑结构的损伤识别已经成为目前国内外研究的热点,并取得了一些阶段性的成果,但远未达到实际应用的程度。本文以结构较复杂并具有众多自由度的网格结构为对象,对基于动力检测的结构整体损伤识别技术进行了深入系统的研究,并为该技术的实际应用提供了有价值的理论方法和试验依据。主要内容和成果如下: 1.在对已有的建筑结构损伤识别方法进行了系统研究的基础上,分析了目前广泛使用的基于传统数学方法和人工神经网络方法的优缺点。由于实测信息的不完备,使得在利用传统的基于模型修正的数学方法进行结构损伤识别时,必须对结构的模型进行缩聚或对实测振型进行扩展。这个过程中的误差再加上模型误差和测试误差,必然使得损伤识别结果的可靠性大大减小。神经网络具有很强的模式识别能力,而且对结构模型的依赖程度不高。可以用有限的测试信息形成训练样本,并建立起结构损伤与结构模态参数变化之间的稳定映射关系,从而可以完成对结构损伤的识别,具有很好的应用前景。2.通过对算例的损伤识别表明,为了获得具有较高模式识别能力的神经网络,对其输入输出参数进行归一化处理是必要的。同时,在训练中加入一定水平的噪音可以提高神经网络的泛化能力,增强其适应性,并对具有测试误差的数据具有更好的识别效果。3.在分析了损伤对结构模态参数的影响的基础上,提出了直接利用结构低阶频率变化率和少数测点振型分量构造结构损伤特征参数的方法。该方法可以充分利用振型和频率的优点,如结构振动频率的变化率可以很好地反映结构的整体特性,测试简单而且精度较高,而振型对结构的损伤更敏感等等。同时,这种输入参数又克服了只利用频率或振型以及目前广为使用的具有歧义的振型差的不足。通过对数值算例和试验模型的损伤识别,证明了所用参数的有效性。4.在对目前常用的测点布置法进行深入研究的基础上,提出了测点布置优先级综合排序法,对结构中的各个自由度进行测点布置优先级排序。该方法综合了单元模态应变能系数法和有效独立法的优点,既考虑了将测点布置在模态能量较大的节点,使其具有很好的可测性,同时又考虑了使实测模态之间保持较大的空间交角,从而使得有限的实测信息能够更充分地反映结构的性态。5.对网格结构的构成型式进行了分析,并针对网格结构中杆件和节点众多,更多还原

【Abstract】 Just like people may get sick, building structures will inevitably suffer from influence of artificial factors and natural factors and then arises aging and worn during the design, construction and especially during service, and the structural accidents will cause the loss on lives and property. So, structural damage identification has been the hotspot in the current domestic and oversea research. The damage identification methods for complex grid structures have been studied in this paper, based on their dynamic characters. Main contents and achievements are as follow: 1. Systematically research has been carried out on the exist methods of the structural damage identification. And the priorities and drawbacks of the traditional mathematical method and artificial neural network-based method for damage identification have been analyzed. For the measured information is far from complete, the model condensation or the mode shapes extension techniques have to be used in the traditional model updating method. And confidence level of the structural damage identification’s result would greatly reduce due to the errors in each procedure. The neural networks have great capability of the pattern recognition, and thus can be well used for structural damage identification with less dependent on the structural model. 2. From the damage identification results of several examples, it is found that, in order to obtain the neural networks with higher pattern recognition performance, it is necessarily for the input and output parameters of the neural networks be normalized. Meanwhile, training with certain level of noise injection will improve the generalization of the neural networks, and will lead to better identification results to the error polluted measured data. 3. Based on analyzing the influence of the damage on the structural modal parameters, grid structural damage signature (GSDS) parameters have been defined, with the first few lower mode frequencies and mode shapes from limited sensor placements, where the normalized mode shapes are used directly instead of the variation of the mode shapes, for the variation of the mode shapes is meaningless. This method can make good use of both advantages of the frequencies and mode shapes. For the variation of the frequencies can well reflect the structural integral performance under damages, and can be easily measured. While the mode shapes are more sensitive to the damages in the structures. At the same time, GSDS can also overcome the shortage with only frequencies or mode shapes used. 4. Based on the study on commonly used sensor placement optimization methods, a sensor placement priority compositor method has been given in this paper. This method takes the model strain energy of the elements and the effective independent of measured mode shapes into consideration. And so it can more sufficiently reflect the structural performance. 5. Based on the study on the formation of grid structures, in which there are plenty of elements and nodes, but the number of nodes is far less than that of the elements. A three-step damage identification method for grid structures is given in this paper. In the first step, the node oriented damage primary localization method is used to更多还原