神经网络法剔除结构损伤检测中的温变影响  

Eliminating temperature influences in structural damage detection by using neural network

作　　者：顾箭峰[1] 邬晓光[1] 姚玉玲[1] 

机构地区：[1]长安大学公路学院,陕西西安710064

出　　处：《长安大学学报（自然科学版）》2016年第3期41-48,共8页Journal of Chang’an University(Natural Science Edition)

基　　金：中央高校基本科研业务费专项资金项目(CHD2011ZD006)

摘　　要：为了研究一种基于神经网络与奇异分析技术的结构损伤检测方法来剔除不利温变影响,以1座能代表中小跨桥梁性能的基准结构的有限元模型为例,分析温变和多级损伤对结构频率的影响,研究该检测方法的有效性和可靠性。以温变条件下无损结构的温度数据和前10阶竖向模态频率训练BP神经网络(BPNN),来构建温度和频率间的量化模型;然后用该网络来预测不同温度条件下结构的频率,计算频率预测误差来消除温变影响;以该预测误差的欧式范数为奇异指标,用结构待检状态奇异指标序列均值与健康状态的相对变化率来指示损伤的存在。研究结果表明:该方法不仅能可靠地检测温变条件下结构的损伤,而且能定性地区分结构整体损伤程度的大小,具有很强的噪声鲁棒性。In order to eliminate adverse influences of temperature fluctuation on damage detection,a method based on neural network and a novel detection technique was proposed.Taking finite element models of a benchmark grid structure as example,which was representative of short to and medium-span bridges,this paper analyzed the influence of temperature variations and several damage scenarios on structural frequency to testify the validity and reliability of the proposed method.A backward propagation neural network（BPNN）was established to formulate the quantitative model of temperature and frequencies of the first ten vertical modes for the intact structure under varying temperatures.Then,structural frequencies were predicted by the BPNN under the conditions of different temperatures,and then prediction errors between the network outputs and the target frequencies were calculated to eliminate temperature effects.Subsequently,Euclidean norm of prediction errors was utilized as a novelty index,and the relative difference between average values of novelty index sequences of candidate structure and intact structure were adopted as an indicator to detect damage.The results show that the proposed method is capable of ascertaining whether damage has occurred reliably and discriminatethe general damage severity of the structure qualitatively regardless of varying temperatures.Additionally,this method has remarkable noise robustness.6tabs,9figs,20 refs.

关 键 词：桥梁工程 基于频率的损伤检测 BP神经网络 温变 

分 类 号：U447[建筑科学—桥梁与隧道工程]