利用压电传感器基于GAF-ResNet的管道焊缝缺陷分类  

Pipeline Weld Defect Classification Based on GAF-ResNet Using Piezoelectric Sensors

作　　者：卫小龙 杜国锋[2] 余泽禹 袁洪强 马骐 Wei Xiaolong;Du Guofeng;Yu Zeyu;Yuan Hongqiang;Ma Qi(Jingzhou Vocational and Technical College,Jingzhou,Hubei 434023,China;School of Urban Construction,Yangtze University,Jingzhou,Hubei 434023,China;School of Electronic Information,Yangtze University,Jingzhou,Hubei 434023,China)

机构地区：[1]荆州职业技术学院,湖北荆州434023 [2]长江大学城市建设学院,湖北荆州434023 [3]长江大学电子信息学院,湖北荆州434023

出　　处：《化工设备与管道》2024年第1期87-93,共7页Process Equipment & Piping

基　　金：国家自然科学基金项目(51778064,52078052);湖北省技术创新专项重大项目(2019AAA011);荆州市2023年度科技计划项目(2023EC36);荆州职业技术学院重点科技创新成果培育工程项目(jzzp202302)。

摘　　要：针对管道焊缝缺陷分类难度大的问题,提出了利用压电传感器数据,结合格拉姆角场(Gramian Angular Field,GAF)和残差神经网络(ResNet)的焊缝缺陷分类方法。先采用GAF原理将一维时间序列数据转化为二维图像,将转化后的二维图像数据集输入,训练最优二维残差神经网络模型用于焊缝缺陷分类。实验中管道焊缝预制了10个缺陷(5种类型),使用导波和超声技术分别对焊缝中1-5号缺陷进行检测,分析Precision(精确率)、Recall(召回率)、F1-score(F1评分)三个指标,证实了基于GAF-ResNet方法的可行性,同时6-10号缺陷验证了该方法的可靠性和普适性。Aiming at the difficulty of classification of pipeline weld defects,a welding defect classification method using piezoelectric sensor data,combined with Gramian Angular Field(GAF)and Residual Neural Network(ResNet)was proposed.Firstly,the GAF principle is used to convert one-dimensional time series data into two-dimensional images,and the converted two-dimensional image data set is used as input to train the optimal two-dimensional residual neural network model for weld defect classification.In the experiment,10 defects(5 types)were prefabricated in the pipeline weld.Guided wave and ultrasonic technology were used to detect the 1-5 defects in the weld respectively,and the three indicators of Precision,Recall and F1-score were analyzed.The feasibility of the GAF-ResNet method,and defects 6-10 verify the reliability and universality of the method.

关 键 词：管道焊缝 缺陷分类 GAF 残差神经网络 导波 超声 

分 类 号：TB553[理学—物理] TP391.4[理学—声学]