【作者】 谭冬梅；

【导师】 瞿伟廉；

【作者基本信息】 武汉理工大学 ， 结构工程， 2003， 硕士

【摘要】 现代的许多土木结构正不断向大型化、复杂化方向发展，新型结构也不断出现，同时，我国在役结构物中有许多建筑物存在不同程度损伤或结构已经进入服役后期，急需损伤鉴定与维修加固，因此结构损伤诊断方法的研究成为新世纪土木工程领域的一个前沿研究热点。 对建筑结构进行动力响应预测、振动控制和结构状态评估及健康监测，首先必需详细了解结构的动力特性。结构的动力特性和结构参数直接相关，这些模态特性可以通过有限元分析得到其理论值，也可以由实验模态分析得到其实测值。由于结构的损伤将引起相应动力特性的改变，结构自振频率的实测值与有限元理论值之间常常存在较大的差异，则需要解决的问题是如何修正结构的有限元模型，使得结构模态特性的理论值趋近于实测值。神经网络方法因其具有非线性映射能力强、计算速度快、容错性好等优点，非常适合于结构有限元模型修正。但实验数据不完备、结构损伤标识量的选取、神经网络的选定等问题没有得到很好的解决。 本文的工程背景为深圳市民中心屋顶网架的健康监测。以工程中广泛应用的网架结构为对象，针对其节点固结系数，利用结构的频率与模态变化量为损伤标识量，研究了神经网络方法在结构有限元模型修正中的应用。 本文建立了节点连接刚度发生变化的网架结构的三维有限元模型，定义了固结系数来描述节点的单元刚度变化，研究了用于振动测试的加速度传感器的优化配置与网架结构模态参数的实测方法。 本文提出了基于神经网络对网架结构有限元模型进行修正的方法。首先确定网架结构以固结系数表示的三维空间有限元模型，根据单元刚度矩阵中的固结系数的变化，建立固结系数与结构动力特性的关系。其次，根据损伤后的结构频率与模态的变化，应用径向基神经网络，进行结构节点固结系数的识别，从而实现对网架结构有限元模型的修正。 根据本文的研究可以得知，径向基神经网络可以很好的用于结构有限元模型的修正，利用频率与模态作为损伤标识量，改善单纯频率作为损伤标识量对结构损伤不敏感的特点，完全可以实现基于神经网络的对网架结构的有限元模型修正。更多还原

【Abstract】 Since more and more structures are becoming larger and more complex than before, there are many new structures in modern time. At the same time, there are various damage in many existed structures, and they have been used for several decades and are to be surveyed and strengthened. Therefore, structural damage diagnosis has become one of the advancing fronts of civil engineering researches.To forecast structural response, vibration controk state evaluation and health monitor, the first thing is to know the structural dynamic characteristic. Structural dynamic characteristic is related to structural parameter, its theory value can be gained by finite element model (FEM) analysis, its practice value can be achieved by experiment modal analysis. Since the structural dynamic characteristic is changed for its damage, the large difference of structural frequency is existed between the theory value and the practice value. The problem that is to be resolved is how to correct structural FEM to make the theory value equal to the practice value, neural network technique is adapted to the FEM correction for its strong non-linear mapping ability, rapid computation and anti-interference capability. But there are still some problems to be solved such as selection of neural network, determination of structural damage indicator and incompletion of measurement.The project background of this paper is the health monitor of roof truss of Shenzhen citizen center. The object is truss structures that are extensively used in civil engineering, the fixity factor is considered and the damage indicator of frequency and modal are used, and neural network technique that is used for FEM correction is studied in this paper.The space FEM is established when joint link stiff is changed in this paper, the fixity factor is defined to show the element stiff change of joint, theplacement of sensor in an optimal fashion for vibration experiment and modal parameter identification of truss structure are studied.The FEM correction method of truss structure based neural network technique is developed in this paper. Firstly, the space FEM of truss structure that based fixity factor is determined, the relation between fixity factor and structural dynamic characteristic is established according to the change of fixity factor of element stiff matrix. Secondly, according to the change of damage structural frequency and modal, the fixity factor is identified by RBF neural network, then FEM correction of truss is finished.According to the research of this paper, RBF neural net can be well adapted to structure FEM correction, the damage indicator of frequency and modal is used to improve the characteristic that damage indicator of pure frequency is not sensitive to structural damage, the FEM correction of truss structure based neural network can be fully finished.更多还原