摘要
在使用冷焊接技术(Cold Metal Transfer,CMT)作为铝合金薄板焊接的自动化焊接工艺过程中,由于电压电流过大或者薄板对接有间隙,常出现烧穿等缺陷。为进一步提高铝合金薄板CMT焊接过程中缺陷识别精准度和检测速度,提出一种基于改进YOLOv5的焊接烧穿缺陷检测模型。通过在主干网络中构建RepLKDeXt模块增大卷积核,更好地捕捉目标的上下文信息,并降低计算复杂度;引入F-EIoU损失函数替换CIoU损失函数实现对经典YOLOV5模型进行改进。搭建焊接试验数据获取平台,进行熔池图像采集。经图像裁剪、灰度化和图像增强等操作后,建立用于焊接缺陷检测模型的训练集和测试集。实验结果表明,与YOLOv3、Fast-RCNN、Faster-RCNN和YOLOv5相比,改进的YOLOv5模型在参数量和计算量减小的同时,提高了CMT铝合金薄板焊接过程中焊接缺陷的检测精度,降低了检测所需时间。
Cold metal transfer(CMT)is used as an automated welding process for welding of thin aluminum alloys.Defects such as burn-through often occur because of excessive voltage and current or because of gaps in the sheet butt joints.Aiming at the problems of low accuracy of defect recognition and slow detection speed in the CMT welding process of aluminum alloy thin plate,a burn-through defect detection model in welding based on improved YOLOv5 is proposed in this work.The RepLKDeXt module is introduced to increase the convolution kernel in the backbone network,the F-EIoU loss function is used instead of the CIoU loss function.A welding test data acquisition platform is built for image acquisition of the welding molten pool.The images are subjected to image cropping,gray scaling and image enhancement operations to build the training and test sets of the welding defect detection model.Experiment results show that compared with YOLOv3,Fast-RCNN,Faster-RCNN and YOLOv5,the improved YOLOv5 model increases the detection accuracy of welding defects in the welding process of CMT aluminum alloy thin plates while reducing the number of parameters and the detection time required.
作者
闫修韬
邹丽
杨鑫华
杨光
杨佳润
YAN Xiutao;ZOU Li;YANG Xinhua;YANG Guang;YANG Jiarun(Software Technology Institute,Dalian Jiaotong University,Dalian 116028;Liaoning Key Laboratory of Welding and Reliability of Rail Transportation Equipment,Dalian 116028)
出处
《舰船电子工程》
2025年第8期156-162,共7页
Ship Electronic Engineering
基金
辽宁省教育厅基本科研项目(编号:LJKMZ20220844)
辽宁省应用基础研究项目(编号:2023JH2/101300236)资助。
