联合卷积与长短记忆神经网络的桥梁结构损伤识别研究  

Research on bridge structural damage detection based on convolutional and long short-term memory neural networks

作　　者：杨建喜[1] 张利凯 李韧[1] 何盈盈 蒋仕新 邹俊志 YANG Jianxi;ZHANG Likai;LI Ren;HE Yingying;JIANG Shixin;ZOU Junzhi(College of Information Science and Engineering,Chongqing Jiaotong University,Chongqing 400074,China;College of Civil Engineering,Chongqing Jiaotong University,Chongqing 400074,China)

机构地区：[1]重庆交通大学信息科学与工程学院,重庆400074 [2]重庆交通大学土木工程学院,重庆400074

出　　处：《铁道科学与工程学报》2020年第8期1893-1902,共10页Journal of Railway Science and Engineering

基　　金：国家自然科学基金资助项目(51608070);重庆市教委科学技术研究资助项目(KJQN201800705,KJQN201900726)。

摘　　要：为改进传统方法在时空相关特征联合提取及结构损伤识别效果等方面存在的不足,结合结构健康监测加速度振动信号的数据特性,将结构损伤识别归约为多变量时间序列分类问题,提出一种联合卷积神经网络(Convolutional Neural Network,CNN)和长短记忆(Long Short-Term Memory,LSTM)循环神经网络模型的桥梁结构损伤识别方法。以结构健康监测获取的加速度振动响应为输入,通过CNN模型提取其多时间窗口内传感器拓扑相关性特征,然后将该特征矩阵输入以Softmax为输出层的LSTM模型,以进一步提取其时间维度特征,并进行结构损伤模式分类。以某连续刚构桥结构缩尺模型的一种无损伤及3种不同程度损伤工况为试验数据环境,验证了提出方法在准确率、精确率、召回率和F值等方面优势。In order to improve the bottlenecks in traditional methods about the combined extraction of spatio-temporal correlation features and structural damage detection,this paper combined the data characteristics of acceleration vibration signal of the structural health monitoring to reduce the structural damage detection to the problem of multivariate time series classification.This paper proposes a novel structural damage detection approach based on Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)neural network models.Taking the acceleration vibration response data obtained by structural health monitoring as input,the correlation features between sensors in multiple time frames is extracted by CNN model,and then the feature matrix is input into the LSTM model which uses Softmax as output layer to further extract time-related features and classify the structural damage patterns.An actual monitoring dataset obtained from a bridge scaled-model is employed as the experimental context.The experimental results verify the advantages of the proposed method in terms of accuracy,precision,recall and F-scores.

关 键 词：桥梁健康监测 结构损伤识别 卷积神经网络 长短记忆神经网络 

分 类 号：U446.2[建筑科学—桥梁与隧道工程]