基于1D-CNN的混凝土内部缺陷超声信号的智能识别  

Intelligent recognition of ultrasonic signals from defects inside concrete based on 1D-CNN

作　　者：汪隆臻 葛维 赵骞 李坚林 刘常鸿 黄伟民 缪春辉 张洁 宋雷 WANG Longzhen;GE Wei;ZHAO Qian;LI Jianlin;LIU Changhong;HUANG Weimin;MIAO Chunhui;ZHANG Jie;SONG Lei(State Grid Anhui Electric Power Co.,Ltd.,Ma′anshan Power Supply Company,Ma′anshan 243000,China;Yunlong Lake Laboratory of Deep Underground Science and Engineering,Xuzhou 221116,China;State Grid Anhui Electric power Co.,Ltd.,Electric Power Research Institute,Hefei 230061,China;State Grid Anhui Electric power Co.,Ltd.,Hefei 230061,China;Department of State Key Laboratory for Geomechanics and Deep Underground Engineering,School of Mechanics and Civil Engineering,China University of Mining and Technology,Xuzhou 221116,China)

机构地区：[1]国网安徽省电力有限公司马鞍山供电公司,安徽马鞍山243000 [2]深地科学与工程云龙湖实验室,江苏徐州221116 [3]国网安徽省电力有限公司电力科学研究院,安徽合肥230061 [4]国网安徽省电力有限责任公司,安徽合肥230061 [5]中国矿业大学力学与土木工程学院深部岩土力学与地下工程重点实验室,江苏徐州221116

出　　处：《混凝土》2024年第5期75-81,共7页Concrete

基　　金：江苏省重点研发计划项目(BE2022709);国家自然科学基金(41974164)。

摘　　要：混凝土内部是否存在缺陷对混凝土工程的安全起着决定性作用。超声无损检测技术是应用最广泛的一种混凝土内部缺陷无损检测技术,依靠工程检测和传统的缺陷识别方法难以满足对混凝土内部缺陷的判别。本研究提出一种一维卷积神经网络模型,基于不同混凝土缺陷的有限元仿真数据库,直接将原始超声信号输入模型中,并且通过添加高斯白噪声来验证一维卷积神经网络模型的准确性和鲁棒性。再通过物理试验验证该模型的识别效果,该模型的识别准确率达到83%以上,且优于其他机器学习模型,验证了该模型对混凝土内部缺陷智能识别的有效性。The integrity of the internal structures of concrete plays a critical role in the safety of concrete projects.Ultrasonic nondestructive testing has become a widely used technique for identifying defects in concrete structures.However,it is difficult to adequately detect such defects through technical inspections and traditional defect detection methods.Therefore,it proposes a one-dimensional convolutional neural network(1D-CNN)based on a finite element simulation database for various defects in concrete.The raw ultrasonic signal was imported into the model and the accuracy and robustness of 1D-CNN model were validated by introducing Gaussian white noise.The fault detection capabilities of CNN model was then verified through physical testing.Experimental results show that the accuracy of the 1D-CNN model exceeds 83%which outperforms other machine learning methods.Indicating the high efficacy of CNN model in distinguishing the defects of concrete.

关 键 词：混凝土内部缺陷 超声信号 卷积神经网络 智能识别 

分 类 号：TU528.07[建筑科学—建筑技术科学]