摘要
交叉焊缝结构复杂,连接方式不同于对接焊缝,易出现裂纹缺陷。焊缝外形轮廓漏磁信号与缺陷信号叠加,使基于电磁信号的检测方法难以提取裂纹特征,增加检测难度。为此提出基于DenseNet网络的红外成像检测方法,采用红外成像装置扫描交叉焊缝生成红外图像,可避免信号叠加问题。将图像输入由密集块、平均池化层等构成的DenseNet模型,能有效提取裂纹缺陷特征。针对100个不同材质的交叉焊缝试件展开测试与评估,实验结果显示,该方法能够更清晰地呈现大部分裂纹缺陷,且仅有一次误将61号试件交叉焊缝表面上的气孔检测成了裂纹,且最大计算量不超过25FLOPs,具有较高的准确性与轻量级。
The structure of cross welds is complex,with connection method differing from those of butt welds,making them prone to crack defects.The superposition of magnetic flux leakage signals and defect signals in the contour of the weld seam makes it difficult for electromagnetic signal-based detection methods to extract crack features,increasing the difficulty of inspection.To this end,an infrared imaging detection method based on DenseNet network is proposed,which uses an infrared imaging device to scan cross welds and generate infrared images,and can avoid the problem of signal superposition.Inputting images into a DenseNet model composed of dense blocks,average pooling layers,etc.can effectively extract crack defect features.Testing and evaluation are conducted on 100 cross weld specimens of different materials.The experimental results show that this method can more clearly present most of the crack defects,and only once mistakenly detects the pores on the surface of the cross weld of specimen 61 as cracks.The maximum calculation amount does not exceed 25 FLOPs,and it has high accuracy and lightweight.
作者
王红梅
强彦
WANG Hong-mei;QIANG Yan(Shanxi Institute of Technology,Yangquan 045000,China;North University of China,Taiyuan 030000,China)
出处
《激光与红外》
北大核心
2025年第8期1253-1258,共6页
Laser & Infrared
基金
山西省教育厅“人工智能程序设计”课程教学设计的探索与研究项目(No.J20221252)资助。
作者简介
王红梅(1980-),女,硕士,高级实验师,研究方向为人工智能。E-mail:whm362144@163.com。
