高效神经网络训练及其在桁架损伤识别中的应用

将应变模态相对变化率作为损伤指标,探索了与人工神经网络结合对桁架结构进行损伤识别的方法。利用ANSYS的命令流语言APDL编写二次开发程序,实现结构不同损伤工况下的模态自动求解过程,并定义APDL中多维数组参数,将不同损伤工况下计算得到的模态参数存储于外部设备。在人工神经网络训练阶段,再借助于Matlab软件的外部文件调用功能,将模态信息加以批量提取。数值仿真表明,本文的方法是一种高效的人工神经网络训练模式,且仅用一阶应变模态改变率就可实现对桁架结构的损伤识别,便于在实际工程中应用。

High Efficiency Training Method for Artificial Neural Network and Its Application to Damage Identification of Truss

This paper presents an investigation into the damage detection of truss structures by taking the relative variation ratio of strain mode as damage index and combining with artificial neural network(ANN) method.The automatic calculation of the mode parameters for different damage situations is achieved using APDL commander flow language in ANSYS,and the results are stored in external devices using the multi-dimension array defined by APDL.In the ANN training stage,the data are automatically loaded using the external file loading function in Matlab,and the mode information is extracted in a batch-bulk manner.Numerical simulation shows that the method proposed in this paper is a high efficiency method for training ANN,and only the variation ratio of the first order strain mode is required,which is convenient for practical application.