基于改进神经网络的建筑结构损伤识别方法研究

Structure Damage Identification with Improved Neural Networks

采用ANSYS有限元分析软件建立神经网络的训练样本,将结构的固有频率作为网络输入,结构损伤的位置和损伤程度作为网络输出,提出了基于神经网络的建筑结构损伤识别方法。讨论了神经网络训练方法和隐含层节点数目对目标函数的影响,分析了网络训练的训练不足和训练过度等问题。以简单的建筑结构为例,基于MATLAB的GUI工具进行了BP神经网络的设计和分析。数值反演结果表明,所改进的建筑结构损伤识别方法具有良好的反演精度和较快的收敛速度。

The training samples of neural network were performed by using finite element method. Modal frequencies of damaged structure were used as input of neural networks. The location and damage index were used as output Of neural networks. The influence of training procedures and the number of hidden layer to objective function was discussed. Taking a simple structure as an example and basing on MATLAB tools, the neural network for identifying structure damage is proposed and investigated. The investigation shows that to identify the location and magnitude of the damaged structure by using an artificial neural network is feasible and a well trained artificial neural network by Levenberg-Marquardt algorithm reveals an extremely fast convergence and a high degree of accuracy.