基于BP神经网络的网架结构损伤定位研究

Damage location of grid structure based on BP neural network

针对实际工程中某正放四角锥网架结构进行了1064种损伤工况的动力计算,定义了归一化固有频率,面向网架节点分别分析了损伤位置和损伤程度对结构各阶固有频率的影响,结果表明:损伤位置对各阶固有频率的影响模式有显著差异,而损伤程度对其影响模式相似,未在本质上影响损伤模式性态;各阶固有频率对损伤位置的响应分布较为开展,包含了损伤模式的个性化信息。据此建立了描述网架结构损伤模式的BP神经网络,对其进行了训练学习,并在实际应用中验证了其损伤定位的可靠性。

The dynamic calculation of 1064 damage conditions is carried out for a square pyramid grid structure in practical engineering,and the normalized natural frequency is defined.Facing the grid nodes,the effects of damage location and damage degree on each order of natural frequency are analyzed respectively,and the following conclusions are obtained:the influence modes of damage location on each order of natural frequency are significantly different,while the influence modes of damage degree are similar,it does not essentially affect the damage mode behavior.The response distribution of each natural frequency to the damage location is relatively developed,including the personalized information of the damage mode.Based on this,the BP neural network describing the damage mode of grid structure is established,and its reliability of damage location is verified in practical application after training and studying.