中国机械工程 ›› 2025, Vol. 36 ›› Issue (9): 2108-2116.DOI: 10.3969/j.issn.1004-132X.2025.09.023

• 服务型制造 • 上一篇    

基于改进YOLOv5s的风电叶片表面缺陷检测方法

王俊, 高贵兵   

  1. 湖南科技大学机电工程学院, 湘潭, 411100
  • 收稿日期:2024-09-27 出版日期:2025-09-25 发布日期:2025-10-15
  • 基金资助:
    湖南省自然科学基金(2023JJ60145);湖南省杰出青年基金(2024JJ2031);湖南省科技创新计划(2023RC3174)

A Method for Detecting Surface Defects on Wind Turbine Blades Based on Improved YOLOv5s

Jun WANG, Guibing GAO   

  1. School of Mechanical and Electrical Engineering,Hunan University of Science and Technology,Xiangtan,Hunan,411100
  • Received:2024-09-27 Online:2025-09-25 Published:2025-10-15

摘要:

为了提高风电机组叶片健康监测技术的智能化、高效化、便捷化发展,依据目标识别技术提出一种基于改进YOLOv5s算法的风电叶片表面缺陷检测方法。首先将YOLOv5s算法的原始骨干网络用渐进特征金字塔网络(AFPN)替换,增强了网络的学习能力;其次将卷积块注意力模块(CBAM)嵌入到主干提取网络中,提高了模型对叶片表面缺陷特征的提取能力;然后使用最小点距离交并比(MPDIoU)损失函数替换CIoU损失函数,提高了边界框定位精度;最后,采用改进的检测方法对某风电机组叶片进行缺陷检测。检测结果表明,改进后的算法在精确率、召回率和平均精度均值(mAP)等方面分别提高了4.1%、2.9%和4.8%,达到了91.9%、89.3%和93.5%,具有显著的精度优势和更好的模型稳定性。

关键词: 风电叶片, 缺陷检测, 渐进特征金字塔网络, 卷积块注意力模块

Abstract:

In order to improve the intelligent, efficient, and convenient development of wind turbine blade health monitoring technology, a wind turbine blade surface defect detection method was proposed based on improved YOLOv5s algorithm according to target recognition technology. Firstly, the original backbone network of YOLOv5s was replaced with an AFPN to enhance the network's learning ability. Secondly, the CBAM was embedded into the backbone extraction network, which enhanced the model's ability to extract surface defect features of leaves. Then, the minimum point distance intersection over union(MPDIoU) loss function was used to replace the CIoU loss function, improving the precision of bounding box localization. Finally, an improved detection method was used to detect defects in the blades of a certain wind turbine unit. The detection results show that the improved algorithm improves precision, recall and mean average precision(mAP) by 4.1%, 2.9% and 4.8%, respectively, reaching as 91.9%, 89.3% and 93.5%, which has significant precision advantages and better model stability.

Key words: wind turbine blade, defect detection, asymptotic feature pyramid network(AFPN), convolutional block attention module (CBAM)

中图分类号: