【摘要】 大跨钢结构目前被广泛应用于体育馆等大型公共建筑中。但其缺陷损伤位置的确定,至今没有得到很好的解决,这必将影响其使用过程中的安全性。利用神经网络技术,以某高校体育场馆的大跨钢结构为工程背景进行模拟损伤定位研究,通过ANSYS计算软件对该大跨钢结构建模分析,得出了该结构在损伤前后的模态参数,并将其结果作为网络的输入参数。为了提高神经网络模型对该结构缺陷损伤判定的收敛速度及诊断精度,在进行损伤识别时,将该大跨结构细分成许多子结构,缩小损伤的范围,同时将高阶频率引入到不同的神经网络训练样本中进行网络训练,检验其对该结构及构件损伤识别的影响。分析结果表明,采用神经网络技术对大型复杂结构进行损伤定位是可行的,并通过该方法的改进,将识别精度大大的提高,所得结论为今后进行网络改进,提高网络的准确性、抗干扰性和泛化能力提供了有意义的参考。更多还原

【Abstract】 Under the background of the large span steel structure of a college stadium,a simulating study on the locating of defect based on neural network was put forward.The modal parameters before and after structural damage was calculated by use of ANSYS model analysis.These parameters were then taken as input parameters of the neural network.In order to improve the convergency speed and the diagnostics accuracy of the network model,when locating the structural damage,the structure was subdivided into a large amount of sub-structures to narrow the damage searching scope,meanwhile,high order frequency components of the signal were introduced into different training samples of the neural network to inspect their influences on the damage location and damage degree.The results show that the technology of neural network for detecting the defects location of large complex structure is feasible,and the identification accuracy is greatly enhanced by improving the method.The conclusions will provide meaningful references to further network improvements such as improving the accuracy,interference immunity and generalization ability of neural network.更多还原