基于YOLO v5的带涂层钢结构亚表面缺陷脉冲涡流热成像智能检测

张玉彬; 刘鹏谦; 陈丽娜; 韩雅鸽; 刘蕊; 谢静; 徐长航

基于YOLO v5的带涂层钢结构亚表面缺陷脉冲涡流热成像智能检测

中国石油大学(华东) 机电工程学院, 山东青岛 266580

详细信息

作者简介:
张玉彬（1998-），男，河南商丘人，博士研究生，主要从事红外热成像无损检测技术研究。E-mail: zzhyubin@163.com

通讯作者:
徐长航（1976-），男，山东巨野人，博士，教授，主要从事安全工程信息化与智能安全工程方面的研究。E-mail: chxu@upc.edu.cn

中图分类号: TP391
计量
- 文章访问数: 167
- HTML全文浏览量: 75
- PDF下载量: 44
出版历程
- 收稿日期: 2023-08-10
- 修回日期: 2023-09-23
- 刊出日期: 2023-10-19

YOLO v5-based Intelligent Detection for Eddy Current Pulse Thermography of Subsurface Defects in Coated Steel Structures

College of Mechanical and Electronic, China University of Petroleum, Qingdao 266580, China

摘要

摘要: 带涂层钢结构亚表面缺陷的存在例如腐蚀、钢基体裂纹及涂层脱粘等，会对整体结构的性能产生影响，并加速涂层系统退化过程。因此，提出一种基于YOLO v5的带涂层钢结构亚表面缺陷脉冲涡流热成像智能检测方法。这一方法可以在不移除涂层的情况下自动检测带涂层钢结构亚表面缺陷，具有重要的工程应用价值。通过所提方法，在保留涂层的情况下，对带涂层钢结构中的腐蚀、裂纹、脱粘等亚表面缺陷进行智能检测。检测结果表明，本文所提方法能够精确地识别和分类带涂层钢结构的4种亚表面缺陷类型：钢基体裂纹、脱粘、严重质量损失（如腐蚀凹坑、腐蚀磨损）以及轻微质量损失（如腐蚀薄层）。4种缺陷类型的检测精度分别高达96%、97%、95%和93%，同时满足实时性检测需求。
- 带涂层钢结构 /
- 缺陷智能检测 /
- 脉冲涡流热成像 /
- YOLO v5
Abstract: Subsurface defects in coated steel structures, such as corrosion, steel matrix cracks, and coating debonding, affect the overall structural performance and accelerate the degradation of coating systems. Therefore, this study proposes a YOLO v5-based intelligent detection method for pulsed eddy current thermography of subsurface defects in coated steel structures. This method can automatically detect subsurface defects in coated steel structures without removing the coating, which is of significant importance for engineering applications. The proposed method intelligently detects subsurface defects such as corrosion, cracks, and debonding in coated steel structures without removing the coating. The detection results show that the proposed method can accurately identify and classify four types of subsurface defects in coated steel structures: cracks in the steel matrix, debonding, severe quality loss (corrosion pits and corrosion abrasion), and slight quality loss (thin corrosion layers); the four defect types can be detected with accuracies of 96%, 97%, 95%, and 93%, respectively, while meeting real-time inspection requirements.
- coated steel structures /
- intelligent defect detection /
- eddy current pulse thermography /
- YOLO v5

HTML全文

图 1 带涂层钢结构缺陷ECPT检测原理

Figure 1. Schematic of ECPT detection for defects in coated steel structure

下载: 全尺寸图片幻灯片

图 2 YOLO v5算法性能测试

Figure 2. The performance test diagram of YOLO v5 algorithm

下载: 全尺寸图片幻灯片

图 3 YOLO v5网络结构

Figure 3. The network structure diagram of YOLO v5

下载: 全尺寸图片幻灯片

图 4 Mosaic数据增强

Figure 4. Mosaic data enhancement

下载: 全尺寸图片幻灯片

图 5 CSP结构

Figure 5. CSP structure

下载: 全尺寸图片幻灯片

图 6 FPN+PAN结构

Figure 6. FPN+PAN structure

下载: 全尺寸图片幻灯片

图 7 YOLO v5模型训练结果

Figure 7. The training results of YOLO v5 model

下载: 全尺寸图片幻灯片

图 8 钢基体裂纹缺陷识别分类效果

Figure 8. Identification and classification effect of steel matrix crack defects

下载: 全尺寸图片幻灯片

图 9 涂层脱粘缺陷识别分类效果

Figure 9. Identification and classification effect of coating debonding defects

下载: 全尺寸图片幻灯片

图 10 严重质量损失缺陷识别分类效果

Figure 10. Identification and classification effect of large material loss defects

下载: 全尺寸图片幻灯片

图 11 轻微质量损失缺陷识别分类效果

Figure 11. Identification and classification effect of minor material loss defects

下载: 全尺寸图片幻灯片

图 12 不同模型的损失函数比较(a) GIoU损失函数曲线；(b) 目标检测损失函数曲线；(c) 目标分类损失函数曲线

Figure 12. Comparison of loss functions between different models: (a) GIoU loss function curves; (b) Target detection loss function curves; (c) Target classification loss function curves

下载: 全尺寸图片幻灯片

图 13 不同模型的评价指标比较：(a) mAP@0.5曲线；(b) mAP@0.5:0.95曲线；(c) 精确率曲线；(d) 召回率曲线

Figure 13. Comparison of evaluation indexes of different models: (a) mAP@0.5 curves; (b) mAP@0.5:0.95 curves; (c) Accuracy curves; (d) Recall rate curves

下载: 全尺寸图片幻灯片

表 1 带涂层钢结构缺陷数据集信息

Table 1 Information on defect data sets for coated steel structure

Training classification	Defect type	Number of images
Training set 909 images	Crack in steel substrate	221
	Coating debonding	70
	Serious quality losses	595
	Minor quality losses	23
Validation set 10 images	Crack in steel substrate	2
	Coating debonding	2
	Serious quality losses	5
	Minor quality losses	1
Test set 90 images	Crack in steel substrate	23
	Coating debonding	15
	Serious quality losses	42
	Minor quality losses	10

下载: 导出CSV

表 2 四种网络结构训练检测性能对比

Table 2 Comparison of training and detect performance of four network architectures

Evaluation indicators	YOLO v5s	YOLO v5m	YOLO v5l	YOLO v5x
Number of network layers/layers	191	263	335	407
Model weights/MB	14.1	43.4	95.3	177.5
Model parameters	7.26318e+06	2.14979e+07	4.74095e+07	8.84538e+07
Anchor box/target	5.56	5.53	5.54	5.54
Training hours/h	0.913	1.096	1.313	1.783
Single-frame inference speed/s	0.009	0.010	0.012	0.017
mAP@0.5	99.5%	99.5%	99.5%	99.5%
mAP@0.5:0.95	88.2%	92.9%	87.0%	89.4%

下载: 导出CSV

表 3 基于测试集的网络性能比较

Table 3 Network performance comparison based on test set

Network models	YOLO v5s	YOLO v5m	YOLO v5l	YOLO v5x
Image/frame	90	90	90	90
Target/piece	90	90	90	90
Batch size	16	16	16	16
AP	0.786	0.851	0.835	0.792
Recall	1	1	1	1
mAP@0.5	0.995	0.995	0.995	0.995
mAP@0.5:0.95	0.812	0.764	0.787	0.762
Single-frame inference time consumption/ms	1.6	3.3	5.0	9.0
Single-frame NMS consumption time/ms	1.0	1.0	0.9	1.0
Total single-frame detection time/ms	2.6	4.3	5.9	9.9

下载: 导出CSV

参考文献(20)

[1]	葛玉龙, 王宁, 潘岩, 等. 环氧低表面带锈底漆在沿海炼化装置涂层修复中的应用[J]. 腐蚀科学与防护技术, 2016, 28(6): 558-602. https://www.cnki.com.cn/Article/CJFDTOTAL-FSFJ201606017.htm GE Y, WANG N, PAN Y, et al. Application of epoxy low surface rust primer in coating repair of coastal refining equipment[J]. Corrosion Science and Protection Technology, 2016, 28(6): 558-602. https://www.cnki.com.cn/Article/CJFDTOTAL-FSFJ201606017.htm
[2]	高燕, 任思明, 刘成宝, 等. 涂层/钢结构腐蚀与耐久性评价中的微区电化学实验研究[J]. 宁波工程学院学报, 2021, 33(1): 14-20. https://www.cnki.com.cn/Article/CJFDTOTAL-LBGS202101003.htm GAO Y, REN S, LIU C, et al. Experimental study of microzone electrochemistry in corrosion and durability evaluation of coatings/steel structures[J]. Journal of Ningbo University of Technology, 2021, 33(1): 14-20. https://www.cnki.com.cn/Article/CJFDTOTAL-LBGS202101003.htm
[3]	JIANG X, XU Y, HU H, et al. Nondestructive testing of corrosion thickness in coated steel structures with THz-TDS[J]. Measurement, 2023, 217: 113088. DOI: 10.1016/j.measurement.2023.113088
[4]	Permeh S, Lau K, Echeverria Boan M, et al. Electrochemical characteristics of antifouling coated steel structure submerged in Florida natural waters to mitigate micro- and macrofouling[J]. Construction and Building Materials, 2021, 274: 122087. DOI: 10.1016/j.conbuildmat.2020.122087
[5]	LIU S, GU Y, WANG S, et al. Degradation of organic pollutants by a Co₃O₄-graphite composite electrode in an electro-Fenton-like system[J]. Chinese Science Bulletin, 2013, 58(19): 2340-2346. DOI: 10.1007/s11434-013-5784-4
[6]	LIU S, ZHAO X R, SUN H Y, et al. The degradation of tetracycline in a photo-electro-Fenton system[J]. Chemical Engineering Journal, 2013, 231: 441-448. DOI: 10.1016/j.cej.2013.07.057
[7]	Hinder S J, Lowe C, Maxted J T, et al. Intercoat adhesion failure in a multilayer organic coating system: an X-ray photoelectron spectroscopy study[J]. Progress in Organic Coatings, 2005, 54(1): 20-27. DOI: 10.1016/j.porgcoat.2005.03.012
[8]	Nishida Y, Takahashi H, Iso M, et al. Surface modification of silica particles with polyimide by ultrasonic wave irradiation[J]. Advanced Powder Technology, 2005, 16(6): 639-648. DOI: 10.1163/156855205774483316
[9]	WANG H, QIAN M, LIU W. Laser ultrasonic characterization of adhesive bonds between epoxy coating and aluminum substrate[J]. Ultrasonics, 2006, 44: e1349-e1353. DOI: 10.1016/j.ultras.2006.05.050
[10]	刘栓, 赵海超, 顾林, 等. 有机涂层/金属腐蚀无损检测技术研究进展[J]. 电镀与涂饰, 2014, 33(22): 993-997. DOI: 10.3969/j.issn.1004-227X.2014.22.010 LIU S, ZHAO H, GU L, et al. Research advances of nondestructive measurement technology for organic coating/metal corrosion system [J]. Electroplating & Finishing, 2014, 33(22): 993-997. DOI: 10.3969/j.issn.1004-227X.2014.22.010
[11]	HE Y, TIAN G Y, PAN M, et al. An investigation into eddy current pulsed thermography for detection of corrosion blister[J]. Corrosion Science, 2014, 78: 1-6. DOI: 10.1016/j.corsci.2013.09.001
[12]	Hernandez J, Fouliard Q, Vo K, et al. Detection of corrosion under insulation on aerospace structures via pulsed eddy current thermography[J]. Aerospace Science and Technology, 2022, 121: 107317. DOI: 10.1016/j.ast.2021.107317
[13]	SHI Z, XU X, MA J, et al. Quantitative detection of cracks in steel using eddy current pulsed thermography[J]. Sensors, 2018, 18(4): 1070. DOI: 10.3390/s18041070
[14]	牟蕾. 基于多维EMD的红外弱小目标检测方法研究[D]. 西安: 西安电子科技大学, 2019. MU L. Research on Infrared Dim Small Target Detection Method Based on Multidimensional EMD[D]. Xi'an: Xidian University, 2019.
[15]	窦田玫. 基于低秩稀疏恢复理论的红外小目标检测算法研究[D]. 西安: 陕西师范大学, 2020. DOU T. Research on Infrared Small Target Detection Algorithm Based on Low Rank Sparse Recovery Theory[D]. Xi'an: Shaanxi Normal University, 2020.
[16]	李向荣, 孙立辉. 融合注意力机制的多尺度红外目标检测[J]. 红外技术, 2023, 45(7): 746-754. http://hwjs.nvir.cn/article/id/2e1d129d-a77a-4dba-8de5-135fb8b75ee7 LI X, SUN L. Multiscale infrared target detection based on attention mechanism[J]. Infrared Technology, 2023, 45(7): 746-754. http://hwjs.nvir.cn/article/id/2e1d129d-a77a-4dba-8de5-135fb8b75ee7
[17]	何赟泽, 李响, 王洪金, 等. 基于可见光和热成像的风机叶片全周期无损检测综述[J]. 机械工程学报, 2023, 59(6): 32-45. HE Y, LI X, WANG H, et al. A review: full-cycle nondestructive testing based on visible light and thermography of wind turbine blade[J]. Journal of Mechanical Engineering, 2023, 59(6): 32-45.
[18]	Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 779-788.
[19]	代牮, 赵旭, 李连鹏, 等. 基于改进YOLO v5的复杂背景红外弱小目标检测算法[J]. 红外技术, 2022, 44(5): 504-512. http://hwjs.nvir.cn/article/id/f71aa5f4-92b0-4570-9056-c2abd5506021 DAI J, ZHAO X, LI L, et al. Improved YOLO v5-based infrared dim-small target detection under complex background[J]. Infrared Technology, 2022, 44(5): 504-512. http://hwjs.nvir.cn/article/id/f71aa5f4-92b0-4570-9056-c2abd5506021
[20]	王洪金, 杜旭, 赵丽劼, 等. 人在回路的风机叶片红外缺陷检测与测量[J/OL][2023-02-08]. 中国测试, https://kns.cnki.net/kcms/detail//51.1714.TB.20230207.1448.012.html. WANG H, DU X, ZHAO L, et al. Human-in-the-loop infrared defect detection and measurement of wind turbine blades[J/OL] [2023-02-08]. China Measurement & Test, https://kns.cnki.net/kcms/detail//51.1714.TB.20230207.1448.012.html.