基于GoogLeNet的混凝土活动裂纹电磁辐射信号识别方法  

Method for identifying electromagnetic radiation signals of active cracks in concrete based on GoogLeNet

作　　者：侯春尧 谭大文 黄松岭[2] 张敬华 张洪毅 毛延翩[3,4] 周益 HOU Chunyao;TAN Dawen;HUANG Songling;ZHANG Jinghua;ZHANG Hongyi;MAO Yanpian;ZHOU Yi(Yongshan Xiluodu Power Plant,Three Gorges Jinsha River Chuanyun Hydropower Development Co.,Ltd.,Zhaotong 657300,China;Department of Electrical Engineering,Tsinghua University,Beijing 100084,China;China Yangtze Power Co.,Ltd.,Wuhan 430000,China;State Key Laboratory of Hydroscience and Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]三峡金沙江川云水电开发有限公司永善溪洛渡电厂,云南昭通657300 [2]清华大学电机系,北京100084 [3]中国长江电力股份有限公司,湖北武汉430000 [4]清华大学水沙科学与水利水电工程国家重点实验室,北京100084

出　　处：《中国测试》2025年第2期30-38,共9页China Measurement & Test

基　　金：国家重点研发计划项目(2022YFF0605600);全国重点实验室自主课题(SKLD22M02);三峡金沙江川云水电开发有限公司永善溪洛渡电厂科研项目资助(合同编号:Z412302001)。

摘　　要：混凝土随结构构件的受力、变形和温度等变化容易形成无法维持稳定的活动裂纹,并在裂纹产生瞬间向空间中发射电磁辐射信号。针对混凝土开裂产生的电磁辐射信号的幅值低、频带宽,容易与电磁环境噪声混淆的问题,该文提出一种基于GoogLeNet的混凝土活动裂纹电磁辐射信号识别方法。该方法搭建电磁信号采集实验平台,通过实际混凝土开裂实验构建原始信号数据集,利用连续小波变换(CWT)将混凝土活动裂纹电磁辐射原始时域信号转化为二维时频域图以强化信号在开裂早期的局部时频域变化特征,并在GoogLeNet基础上构建并迭代训练模型。结果表明,该文提出的方法平均识别准确率为99.63%,泛化识别成功率高于97%,与支持向量机和残差网络等信号识别方法相比,更适用于判定混凝土结构是否存在活动裂纹。Concrete tends to develop unstable active cracks in response to structural elements experiencing changes in stress,deformation,temperature,and other factors.Upon formation,these cracks emit electromagnetic radiation signals.Addressing the challenges of low signal amplitude,wide frequency bandwidth,and susceptibility to interference from electromagnetic environmental noise,this paper presents a concrete active crack electromagnetic radiation signal recognition method based on GoogLeNet.This method establishes an electromagnetic signal acquisition experimental platform,constructs a raw signal dataset through actual concrete cracking experiments,and utilizes Continuous Wavelet Transform(CWT)to transform the original time-domain electromagnetic radiation signals of active concrete cracks into two-dimensional timefrequency domain images.This transformation enhances the presentation of detailed signal information,highlighting the local time-frequency domain variations in the early stages of cracking.A model is constructed and iteratively trained on the foundation of GoogLeNet.The results demonstrate that the proposed method achieves an average accuracy rate of 99.63% in signal recognition,with a generalization recognition success rate exceeding 97%.In comparison with commonly used signal recognition methods such as Support Vector Machines and Residual Networks,it is better suited for determining the presence of active cracks in concrete structures.

关 键 词：混凝土 活动裂纹 电磁辐射 GoogLeNet 信号识别 

分 类 号：TB9[一般工业技术—计量学]