基于多注意力机制的双通道模型对风机叶片表面目标识别

doi:10.19936/j.cnki.2096-8000.20250228.016

复合材料科学与工程 ›› 2025, Vol. 0 ›› Issue (2): 129-136.DOI: 10.19936/j.cnki.2096-8000.20250228.016

基于多注意力机制的双通道模型对风机叶片表面目标识别

柯灿阳, 文传博^*

上海电机学院电气学院,上海 201306

收稿日期:2023-11-13 出版日期:2025-02-28 发布日期:2025-03-25
通讯作者: 文传博(1981—),男,博士,教授,硕士生导师,研究方向为复杂设备故障诊断与寿命预测、多源信息融合等,wencb@sdju.edu.cn。
作者简介:柯灿阳(1998—),男,硕士研究生,研究方向为风力发电机故障诊断、深度学习等。
基金资助:
国家自然科学基金(61973209)

Dual-channel model based on multi-attention mechanism for wind turbine blade surface target recognition

KE Canyang, WEN Chuanbo^*

School of Electrical Engineering, Shanghai Dianji University, Shanghai 201306, China

Received:2023-11-13 Online:2025-02-28 Published:2025-03-25

摘要/Abstract

摘要： 目前,风机叶片缺陷识别主要依赖于望远镜和停机吊篮这两种方法,但这些方法存在准确率和安全性方面的问题。随着无人机技术的广泛应用,利用无人机进行风机叶片表面缺陷的目标识别成为备受瞩目的选择。本文采用计算机视觉的方法,提出了一种结合Swin Transformer和轻量级神经网络两种不同架构的中量级网络对风机叶片表面目标进行识别。本文收集了华东某沿海风场的1 275张叶片图像,并通过图像增强技术将数据扩充为原来的5倍。模型中,Swin Transformer为主特征提取器,轻量级神经网络为辅特征提取器,引入CBAM注意力机制增强模型对局部关键信息的聚焦能力,并采用CosineAnnealingWarmRestarts的学习率调整策略优化模型性能。实验结果表明:本文所提出的模型在风机叶片目标识别中准确率、F1分数分别达到97.8%和96.35%,均领先于同等量级的主流模型,为风机叶片表面目标识别提供了一种新的思路。

关键词: 风机叶片, 复合材料, 图像识别, Swin Transformer, 轻量级神经网络, 注意力机制

Abstract: At present, the identification of defects in wind turbine blades mainly relies on two methods: telescopes and shutdown gondolas, but these methods have problems in accuracy and safety. With the widespread application of drone technology, the use of drones to identify surface defects on wind turbine blades has become a high-profile option. In this paper, a medium scale network combining Swin Transformer and lightweight neural network was proposed to identify the surface target of wind turbine blade using computer vision method. A dataset comprising 1 275 blade images from a coastal wind farm in East China was collected, and data augmentation techniques were employed to expand the dataset by a factor of five. In the model, the Swin Transformer served as the primary feature extractor, while the lightweight neural network functioned as the auxiliary feature extractor. The CBAM attention mechanism was introduced to enhance the model’s focus on crucial local information, and the learning rate was adjusted using the CosineAnnealingWarmRestarts strategy to optimize model performance. Experimental results show that the accuracy and F1-score of the model proposed in this paper reached 97.8% and 96.35% respectively, which are both ahead of the mainstream models of the same magnitude, providing a new method for wind turbine blade surface target recognition.

Key words: wind turbine blades, composites, image identification, Swin Transformer, CNN, attention mechanism

中图分类号:

TB332

柯灿阳, 文传博. 基于多注意力机制的双通道模型对风机叶片表面目标识别[J]. 复合材料科学与工程, 2025, 0(2): 129-136.

KE Canyang, WEN Chuanbo. Dual-channel model based on multi-attention mechanism for wind turbine blade surface target recognition[J]. COMPOSITES SCIENCE AND ENGINEERING, 2025, 0(2): 129-136.

参考文献

[1] DAVIDSON M, KARPLUS V J, ZHANG D, et al. Policies and institutions to support carbon neutrality in China by 2060[J]. Economics of Energy & Environmental Policy, 2021, 10(2): 7-25.
[2] VARGAS S A, ESTEVES G R T, MAÇAIRA P M, et al. Wind power generation: a review and a research agenda[J]. Journal of Cleaner Production, 2019, 218: 850-870.
[3] COUNCIL G W E. Global offshore wind report 2020[J]. Global Wind Energy Coucil: Brussels, 2020, 19: 10-12.
[4] CIVERA M, SURACE C. Non-destructive techniques for the condition and structural health monitoring of wind turbines: a literature review of the last 20 years[J]. Sensors, 2022, 22(4): 1627-1679.
[5] YANG R, HE Y, ZHANG H. Progress and trends in nondestructive testing and evaluation for wind turbine composite blade[J]. Renewable and Sustainable Energy Reviews, 2016, 60: 1225-1250.
[6] 李春雷, 王洪江, 尹常永, 等. 风机叶片故障诊断技术的研究进展[J]. 沈阳工程学院学报(自然科学版), 2022, 18(3): 1-5, 19.
[7] 周季峰, 石腾, 许波峰. 风电机组叶片损伤故障检测技术研究进展[J]. 新能源进展, 2023, 11(6): 556-563.
[8] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
[9] SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Boston: IEEE, 2015: 1-9.
[10] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778.
[11] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 7132-7141.
[12] LIU Z, LIN Y, CAO Y, et al. Hierarchical vision transformer using shifted windows[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2021: 10012-10022.
[13] DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16×16 words: transformers for image recognition at scale[EB/OL]. (2020-10-22)[2023-11-13]. https://arxiv.org/abs/2010.11929.
[14] HOWARD A G, ZHU M, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL]. (2017-04-17)[2023-11-13]. https//arxiv.org/abs/1704.04861.
[15] ZHANG X, ZHOU X, LIN M, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 6848-6856.
[16] TAN M, LE Q. EfficientNet: rethinking model scaling for convolutional neural networks[EB/OL]. (2019-05-28)[2023-11-13]. https://arxiv.org/abs/1905.11946.
[17] WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the European conference on computer vision. Munich: Springer, 2018: 3-19.

基于多注意力机制的双通道模型对风机叶片表面目标识别

Dual-channel model based on multi-attention mechanism for wind turbine blade surface target recognition

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	关纪文, 王玉梅, 韦丽兰, 孔德辅, 陈华, 熊超华. CFRP筋增强珊瑚骨料混凝土偏压短柱力学行为及配筋率研究[J]. 复合材料科学与工程, 2025, 0(2): 1-11.
[2]	陆春华, 李朝晖, 漆仲浩, 朱学武, 徐奕帆. 高温环境对混凝土内GFRP筋拉伸性能退化影响[J]. 复合材料科学与工程, 2025, 0(2): 12-19.
[3]	程艳雯, 赵菲, 黄振辉, 张青松, 李良勇, 习鑫强. 碱处理椰壳纤维对海水海砂混凝土力学性能影响研究[J]. 复合材料科学与工程, 2025, 0(2): 20-27.
[4]	王朔, 贾晨辉, 杨智明, 姚亚琳. 含水量对环氧/酸酐树脂性能的影响及解决措施研究[J]. 复合材料科学与工程, 2025, 0(2): 28-33.
[5]	牛芳旭, 孙超明, 贺靖, 尹航. 共固化蜂窝夹层结构长梁弯曲性能预测[J]. 复合材料科学与工程, 2025, 0(2): 34-39.
[6]	徐小魁, 潘江, 郝尚, 张娟娟, 吴春燕, 刘千立. 国产高导热中间相沥青基碳纤维复合材料性能研究[J]. 复合材料科学与工程, 2025, 0(2): 40-45.
[7]	伊文昭, 刘璐璐, 徐凯龙, 陈伟. 考虑湿热老化的单向纤维增强复合材料纵向脱黏行为的数值模拟[J]. 复合材料科学与工程, 2025, 0(2): 46-53.
[8]	闫珊, 付天宇, 许家忠, 史新民. 基于改进遗传算法的纤维张力模糊控制研究[J]. 复合材料科学与工程, 2025, 0(2): 54-61.
[9]	高文明, 刘琛, 聂海平, 阳铭广, 王显峰. 复合材料飞机副油箱结构铺层设计及性能分析[J]. 复合材料科学与工程, 2025, 0(2): 70-81.
[10]	杨瑞瑞, 王雅婷, 冉晓璐, 刘万双. 硅烷改性氧化石墨烯增韧乙烯基酯树脂[J]. 复合材料科学与工程, 2025, 0(2): 82-90.
[11]	张月义, 唐小惠, 高尚兵, 李刚, 杨小平. 氰酸酯树脂与高强高模型碳纤维的界面相容和性能匹配研究[J]. 复合材料科学与工程, 2025, 0(2): 91-98.
[12]	吕林泽, 张芝芳, 郑铭峰, 江剑. 温湿环境下CFRP层合板的弯曲疲劳寿命与结构频率的关联性研究[J]. 复合材料科学与工程, 2025, 0(2): 99-107.
[13]	房恩惠, 杨树桐, 庞瑞阳, 孙忠科, 兰天. 树脂预浸技术提升CFRP-混凝土界面力学性能研究[J]. 复合材料科学与工程, 2025, 0(2): 117-128.
[14]	颜丙越, 张卓, 尹立, 何剑飞. 基于数字单元法的平纹织物压实形态变化研究[J]. 复合材料科学与工程, 2025, 0(2): 137-144.
[15]	徐东亮, 薛紫阳, 赖九衡. 基于机器视觉玻璃纤维束缺陷检测技术的研究[J]. 复合材料科学与工程, 2025, 0(2): 145-150.