东北大学学报(自然科学版) ›› 2024, Vol. 45 ›› Issue (11): 1595-1603.DOI: 10.12068/j.issn.1005-3026.2024.11.010

• 机械工程 • 上一篇    下一篇

基于改进YOLOv8的铆接孔表面缺陷检测

郝博1,2, 徐新岩1(), 赵玉欣1, 闫俊伟1   

  1. 1.东北大学 航空动力装备振动及控制教育部重点实验室,辽宁 沈阳 110819
    2.东北大学秦皇岛分校 控制工程学院,河北 秦皇岛 066004
  • 收稿日期:2023-06-21 出版日期:2024-11-15 发布日期:2025-02-24
  • 通讯作者: 徐新岩
  • 作者简介:郝 博(1963-),男,辽宁沈阳人,东北大学教授,博士生导师.
  • 基金资助:
    装备预先研究领域基金资助项目(61409230125)

Surface Defect Detection of Riveting Holes Based on Improved YOLOv8

Bo HAO1,2, Xin-yan XU1(), Yu-xin ZHAO1, Jun-wei YAN1   

  1. 1.Key Laboratory of Vibration and Control of Aero-Propulsion System,Ministry of Education,Northeastern University,Shenyang 110819,China
    2.School of Control Engineering,Northeastern University at Qinhuangdao,Qinhuangdao 066004,China.
  • Received:2023-06-21 Online:2024-11-15 Published:2025-02-24
  • Contact: Xin-yan XU
  • About author:XU Xin-yan, E-mail: 1035011108@qq.com

摘要:

飞机蒙皮、尾翼等零件上的铆接孔表面质量关乎飞机整体装配性能,目前铆接孔缺陷检测大多为传统人工检测,易出现漏检现象.因此,提出一种具有创新性的改进YOLOv8的铆接孔表面缺陷检测方法.采用可变形卷积替换常规卷积,解决特征提取中感受野形状固定的问题.嵌入SimAM注意力机制,增强网络在背景和目标对比度较低状况下的辨识能力.使用WIoU边界框回归损失函数代替CIoU损失函数,降低低质量图像对模型训练的影响,提高模型的鲁棒性和泛化能力.为验证本文模型的性能,以带铆接孔的6061铝合金板代替飞机蒙皮进行检测.实验结果表明,本文模型在铆接孔测试集上mAP_0.5和准确率分别达到了0.918和0.920,较原始YOLOv8模型分别提高了24.1%和25.3%.

关键词: YOLOv8, 铆接孔, 缺陷检测, 可变形卷积, 注意力机制

Abstract:

The surface quality of riveting holes on aircraft skin, tail and other components is crucial to the overall assembly performance of an aircraft. Currently, most riveting hole defect detection relies on the traditional manual methods, which are prone to missing detection. An improved detection method based on YOLOv8 for surface defect detection of riveting holes was proposed. The conventional convolution was replaced by deformable convolution to solve the problem of the fixed receptive field shape in feature extraction. The SimAM attention mechanism was embedded in order to enhance the recognition ability of the network under low contrast between the background and targets. The CIoU loss function was replaced by the WIoU bounding box regression loss function to reduce the impact of low?quality images during model training and improve the robustness and generalization ability of the model. To verify the performance of the improved model, 6061 aluminum alloy plates with riveting holes were used as a substitute for aircraft skin in the detection process. Experimental results demonstrated that the improved model achieved mAP_0.5 and accuracy of 0.918 and 0.920 on the riveting hole test set, which represents an improvement of 24.1% and 25.3% compared to the original model.

Key words: YOLOv8, riveting hole, defect detection, deformable convolution, attention mechanism

中图分类号: