基于融合特征矩阵和CNN模型的管道埋藏微裂纹超声导波定位  

Localization of buried micro⁃cracks in pipelines using ultrasonic guided waves based on fusion feature matrix and CNN model

作　　者：颜劲夫 艾星 何其骏 李佳惠 胡碗铷 李义丰[1] Yan Jingfu;Ai Xing;He Qijun;Li Jiahui;Hu Wanru;Li Yifeng(College of Computer and Information Engineering(College of Artificial Intelligene),Nanjing Tech University,Nanjing,211800,China;Key Laboratory of Modern Acoustics,Ministry of Education,Institute of Acoustics,Nanjing University,Nanjing,210093,China)

机构地区：[1]南京工业大学计算机与信息工程学院(人工智能学院),南京211800 [2]近代声学教育部重点实验室,南京大学声学研究所,南京210093

出　　处：《南京大学学报(自然科学版)》2025年第1期28-37,共10页Journal of Nanjing University(Natural Science)

基　　金：国家自然科学基金(61571222);江苏省研究生科研与实践创新计划(SJCX24_0563)

摘　　要：针对石化行业管道中埋藏微裂纹的检测和定位问题,提出了一种基于融合特征矩阵和卷积神经网络(Convolutional Neural Network,CNN)模型的新方法.首先,使用ABAQUS软件建立包含埋藏微裂纹的三维管道仿真模型,并部署传感器阵列采集超声导波信号.然后,提取并融合两类特征因子来构建融合特征矩阵:反映微裂纹非线性特征的零频分量和表征微裂纹位置信息的损伤指数.最后,将该矩阵输入到构建的CNN模型中进行训练和测试,实现了对管道中埋藏微裂纹的精确定位.仿真分析结果表明,与传统的高次谐波分量相比,零频分量对微裂纹的非线性效应更敏感;损伤指数可以放大原始信号中包含的损伤信息,对埋藏微裂纹的定位具有重要价值;CNN模型可以有效地从融合特征矩阵中提取微裂纹的位置信息,确定其空间坐标,为无损检测领域管道微裂纹的定位提供了新的思路.This paper presents an innovative approach for accurately locating buried micro⁃cracks in petrochemical pipelines by integrating a fusion feature matrix with a convolutional neural network(CNN)model.The study addresses the challenge of detecting and localizing micro⁃cracks within pipeline structures,which are more difficult to identify and pose greater risks than surface⁃breaking cracks do.The methodology begins by establishing a three⁃dimensional finite element model of a pipeline containing buried micro⁃cracks using ABAQUS software,which is then simplified into a two⁃dimensional plane to facilitate crack localization.A total of 1353 different buried micro⁃crack scenarios are simulated by varying the crack positions.Two key feature factors are extracted from the collected ultrasonic guided wave signals:the zero⁃frequency component,which exhibits high sensitivity to the nonlinear effects from the interaction between guided waves and micro⁃cracks,and the damage index,which quantifies the extent of pipeline damage based on the time⁃of⁃flight difference between damaged and intact signals.These factors are fused to create a comprehensive feature matrix,which is then fed into a CNN model for training and testing.The results demonstrate that the zero⁃frequency component is significantly more sensitive to micro⁃cracks than conventional second harmonic components,capturing weak nonlinear effects that are often missed by traditional linear inspection methods.The damage index effectively amplifies the micro⁃crack⁃induced damage information in the ultrasonic signals,enabling precise quantification of pipeline damage.The fusion of these complementary features enhances the CNN model′s accuracy in localizing buried micro⁃cracks.In conclusion,this research offers a novel solution for micro⁃crack localization in petrochemical pipelines by combining nonlinear ultrasonic guided wave techniques with advanced machine learning algorithms.The proposed method,validated through extensive simulations,s

关 键 词：零频分量 损伤指数 融合特征矩阵 CNN 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TG115.285[自动化与计算机技术—控制科学与工程]