基于小波散射卷积神经网络的结构损伤识别被引量：5

Structural damage identification based on the wavelet scattering convolution neural network

在线阅读下载PDF

导出

摘要损伤识别是结构状态评估领域的关键问题之一,对确保结构安全性有重要意义。深度学习算法在基于振动的结构损伤识别方面带来了许多突破,但从海量数据中挖掘结构损伤关键信息仍是亟待解决的技术难题。该研究提出了基于一维卷积神经网络(one-dimensional-convolutional neural network,1D-CNN)深度学习的结构多类型损伤识别模型,采用小波散射变换对1D-CNN架构第一层卷积滤波器进行替换,通过散射系数实现输入层原始数据降维与特征提取,结合CNN卷积层、激活层和池化层实现监测数据特征增强处理。在此基础上,结合1D-CNN全连接层与Softmax函数实现特征数据分类,从而实现结构多类型损伤定位与定量高效识别。通过钢桁架结构和斜拉桥两种数值模型对上述框架进行了验证。结果表明:与普通卷积神经网络模型相比,基于小波散射卷积神经网络的结构损伤识别精度显著提升,损伤分类准确率达95.0%以上。随着传感数据环境噪声比例的增加,小波散射卷积神经网络损伤分类准确率虽略有下降,但仍保持较高精准度,说明该方法具有较强的鲁棒性抗噪能力。 Damage identification is one of the key issues in the field of structural condition assessment,which is of great importance to ensure structural safety.The deep learning algorithm has led to many breakthroughs in vibration-based structural damage identification,but it is still an urgent technical challenge to obtain the key information of structural damage from massive amounts of data.A multi-type structural damage identification model was proposed based on the one-dimensional-convolutional neural network(1D-CNN)deep learning.The wavelet scattering transform was used to replace the convolutional filter in the first layer of the 1D-CNN architecture.The scattering coefficients were used to achieve dimensionality reduction and feature extraction of the original data in the input layer,and the CNN convolutional layer,activation layer and pooling layer were combined to achieve feature enhancement processing of monitoring data.The 1D-CNN fully-connected layer and Softmax function were combined to classify the feature data,thus realizing the location and quantitative identification of multi-type structural damages.The above frame was verified by two numerical models of a steel truss structure and a cable-stayed bridge.The results show that compared with the normal convolutional neural network model,the accuracy of structural damage identification based on the wavelet scattering based convolutional neural network is significantly improved,and the accuracy of damage classification is more than 95.0%.In addition,with the increase of the proportion of environmental noise in sensor data,the accuracy of the wavelet scattering convolutional neural network damage classification slightly decreases but still has high accuracy,indicating that the method has strong robustness and anti noise ability.

作者马亚飞李诚何羽王磊涂荣辉 MA Yafei;LI Cheng;HE Yu;WANG Lei;TU Ronghui(School of Civil Engineering,Changsha University of Science&Technology,Changsha 410114,China;Traffic Engineering Management Center of Zhejiang Province,Hangzhou 310014,China)

机构地区长沙理工大学土木工程学院浙江省交通工程管理中心

出处《振动与冲击》 EI CSCD 北大核心 2023年第14期138-146,共9页 Journal of Vibration and Shock

基金国家重点研发计划项目(2019YFC1511000) 国家自然科学基金(52078055) 湖南省自然科学基金创新研究群体(2020JJ10060)。

关键词结构状态评估深度学习小波散射变换卷积神经网络(CNN) 损伤识别 structural condition assessment deep learning wavelet scattering transform convolutional neural network(CNN) damage identification

分类号 TU312.3 [建筑科学—结构工程]

作者简介第一作者:马亚飞,男,博士,教授,1984年生;通信作者:王磊,男,博士,教授,1979年生。