基于机器学习的黏钢构件黏接层缺陷识别方法研究  

A Defect Identification Method for Bonding Layers of Adhesive Steel Members Based on Machine Learning

作　　者：姚浩 夏桂然 刘泽佳[2] 周立成[2] YAO Hao;XIA Guiran;LIU Zejia;ZHOU Licheng(Guangzhou Communications Investment Group Co.,Ltd.,Guangzhou 510330,P.R.China;School of Civil Engineering&Transportation,South China University of Technology,Guangzhou 510641,P.R.China)

机构地区：[1]广州交通投资集团有限公司,广州510330 [2]华南理工大学土木与交通学院,广州510641

出　　处：《应用数学和力学》2024年第4期429-442,共14页Applied Mathematics and Mechanics

基　　金：广东省自然科学基金(2023A1515012942)。

摘　　要：对黏钢加固结构黏接层缺陷对超声检测信号的影响进行了深入研究,并提出了一种基于机器学习的黏接层缺陷识别的新型方法.首先,该文基于直接接触式的脉冲回波反射法对黏钢构件进行有限元模拟,并阐述了超声波在黏钢构件中的传播规律;其次,通过分析局部段超声回波信号及相关信号特征,讨论了不同缺陷变量对超声回波信号的影响规律;最后,建立了黏钢构件超声时程响应数据集,并对比了不同机器学习模型对缺陷大小、位置的分类识别性能,形成了黏钢构件黏接层缺陷识别方法.结果表明,局部段超声回波信号及其特征随着缺陷大小、位置的改变呈规律性变化,能够对缺陷信息进行初步区分.同时,该文提出的基于RF模型的黏钢构件黏接层缺陷识别方法能够有效识别黏钢构件黏接层缺陷,具有较广阔的工程应用前景.The effects of bonding layer defects on ultrasonic detection signals of bonded steel reinforced structures were deeply studied and a new method for the bonding layer defect identification based on machine learning was proposed.Firstly,based on the direct contact pulse-echo reflection method,the finite element simulation of the viscous steel member was carried out,and the propagation law of ultrasonic waves in the viscous steel member was expounded.Secondly,the characteristics of local ultrasonic echo signals and related signals were analyzed,and the effects of different defect variables on ultrasonic echo signals were discussed.Finally,the ultrasonic time-history response data set of the adhesive steel member was established,and the classification and recognition performances of different machine learning models for the size and location of defects were compared,and the defect identification method for the adhesive layer of the bonded steel member was built.The results show that,the local ultrasonic echo signal and its characteristics change regularly with the defect size and location,which can help preliminarily distinguish the defect information.Meanwhile,the proposed RF model-based defect identification method can effectively identify the defects of the adhesive layer in the bonded steel member,and has a broad engineering application prospect.

关 键 词：超声检测 机器学习 黏钢构件 黏接层缺陷 

分 类 号：O3[理学—力学]