基于卷积神经网络的桥梁损伤识别方法应用研究  

Application of bridge damage identification based on convolutional neural network

作　　者：李雪松 林逸洲 马宏伟[1,3] 聂振华 LI Xuesong1, LIN Yizhou2, MA Hongwei1,3, NIE Zhenhua2(1. School of Civil Engineering, Qinghai University, Xining 810016, China;2. School of Mechanics and Constrution Engineering, Jinan University, Guangzhou 510000,China; 3. Dongguan University of Technology, Dongguan 523808,Chin)

机构地区：[1]青海大学土木工程学院,青海西宁810016 [2]暨南大学力学与土木工程学院,广东广州510000 [3]东莞理工学院,广东东莞523808

出　　处：《青海大学学报（自然科学版）》2018年第2期41-46,共6页Journal of Qinghai University(Natural Science)

基　　金：青海省科学技术厅项目(2016-ZJ-721);国家自然科学基金项目(11472146)

摘　　要：目前桥梁健康监测系统普遍存在数据量过大无法高效分析的缺点。为了改善健康监测系统数据灾难问题,本文提出基于卷积神经网络(CNN)的桥梁损伤识别方法。通过简支梁振动试验,取得9个测点加速度数据训练CNN,测试网络识别准确率,分析CNN在桥梁损伤识别应用中的有效性。在此基础上分析各种激励大小对CNN桥梁损伤识别影响,以及模拟真实环境在信号中添加噪声测试CNN性能。结果表明:CNN具有在噪声环境以及弱激励环境下良好的损伤识别性能。本文方法的阶段性试验成果能为桥梁监测系统数据灾难问题提供新的解决思路。The current bridge health monitoring systems have the disadvantage of too large data volume to high-efficiency analysis. To improve the data disaster of health monitoring system,this paper presents a bridge damage identification method based on Convolutional Neural Network（ CNN）.Nine points of acceleration data trained CNN were obtained through the vibration test of simple supported beams to test the accuracy of network identification,and the effectiveness of CNN in the application of bridge damage identification was analyzed. Based on this,the influences of various incentive sizes on the damage identification of CNN bridges were investigated,and the performance of CNN was tested by adding noise to the real environment. The experimental results show that CNN has good damage identification performance under noisy environment and weakly tense atmosphere.The preliminary results of this method can provide a new solution to the data disaster of bridge monitoring system.

关 键 词：卷积神经网络 损伤识别 简支梁试验 弱激励环境 

分 类 号：TU317[建筑科学—结构工程]