基于数据增强与多尺度融合的航空发动机叶片损伤检测

贾路; 曾昊; 彭靖波; 候欣耀

doi:10.13224/j.cnki.jasp.20250155

基于数据增强与多尺度融合的航空发动机叶片损伤检测

doi: 10.13224/j.cnki.jasp.20250155

空军工程大学航空工程学院，西安 710038

详细信息

作者简介:
贾路（1993-），男，硕士生，研究领域为航空器能源动力工程。E-mail：jialu20120812@163.com

通讯作者:
候欣耀（2000-），男，硕士生，研究领域为航空器能源动力工程。E-mail：19192411230@163.com

中图分类号: V263.6；TP391.41
计量
- 文章访问数: 623
- HTML浏览量: 544
- PDF量: 74
- 被引次数: 0
出版历程
- 收稿日期: 2025-03-29
- 网络出版日期: 2025-07-30

Aero-engine blade damage detection based on data augmentation and multi-scale fusion

Aviation Engineering School，Air Force Engineering University，Xi’an 710038，China

摘要

摘要:
为解决航空发动机叶片服役损伤检测中小尺寸损伤漏检率高、检测条件差异较大等问题，以YOLOv5框架为基础，提出了一种航空发动机叶片服役损伤检测方法。提出了一种数据增强方法，通过随机仿射变换、HSV空间增强、随机横向纵向翻转3种方式，解决了叶片损伤类别样本不平衡，检测条件差异大的问题。针对模型小尺度损伤检测能力弱的问题，提出了多核自适应卷积模块和多尺度特征聚合模块，提升了网络提取不同尺度损伤信息和损伤特征判别的能力，解决了叶片损伤样本尺寸较小，损伤外观特征不明显、检测精度低的问题。实验结果表明：改进模型相比基准模型，检测精度mAP@0.5提升了16.8%。对比实验中，改进模型检测精度高于Faster-RCNN、SSD、Deformable DETR、YOLOv8、YOLOv11和YOLOv12等主流的目标检测框架。另外与基于YOLO模型现有航空发动机缺陷模型SW-YOLO、YOLOv8-EMA对比，检测精度mAP@0.5分别提升了13.9%和15.3%。
- 航空发动机叶片服役损伤 /
- 损伤检测 /
- 数据增强 /
- 多尺度融合 /
- YOLOv5网络模型
Abstract:
To address high missed detection rates for small-sized damages and significant environmental variations in aero-engine blade inspection during service-induced damage, an enhanced YOLOv5-based detection method was developed. A data augmentation strategy integrating random affine transformation, HSV space enhancement, and bidirectional flipping was proposed, which effectively mitigated sample imbalance across damage categories. Second, a multi-kernel adaptive convolution module and multi-scale feature aggregation module were introduced to improve network capability in extracting fine-grained damage features and discriminating subtle characteristics, effectively addressing the technical bottlenecks of insufficient detection performance for small-sized damages with indistinct morphological features. The experimental results showed that compared with the baseline model, the detection accuracy (mAP@0.5) of the improved model increased by 16.8%. In the comparison, the detection accuracy of the improved model was higher than that of mainstream object detection frameworks such as Faster-RCNN, SSD, Deformable DETR, YOLOv8, YOLOv11, and YOLOv12. Furthermore, the proposed method exhibited 13.9% and 15.3% improvements in mAP@0.5 over existing defect detection models SW-YOLO and YOLOv8-EMA, respectively.
- aero-engine blade service-induced damage /
- damage detection /
- data augmentation /
- multiscale fusion /
- YOLOv5 network model

HTML

图 1 YOLOv5-6.0的结构

Figure 1. Structure of YOLOv5-6.0

下载: 全尺寸图片幻灯片

图 2 YOLOv5中的FPN和PAN结构

Figure 2. FPN and PAN structures in YOLOv5

下载: 全尺寸图片幻灯片

图 3 本文网络结构图

Figure 3. Proposed network architecture diagram

下载: 全尺寸图片幻灯片

图 4 多核自适应卷积模块结构图

Figure 4. Architecture diagram of the multi-core adaptive convolution module

下载: 全尺寸图片幻灯片

图 5 多尺度特征金字塔融合模块结构图

Figure 5. Architecture diagram of the multi-scale feature pyramid fusion module

下载: 全尺寸图片幻灯片

图 6 数据集叶片损伤类别图片

Figure 6. Blade damage category images in the dataset

下载: 全尺寸图片幻灯片

图 7 不同检测模型可视化结果对比

Figure 7. Comparison of visualization from different detection models

下载: 全尺寸图片幻灯片

图 8 本文检测算法的PR曲线

Figure 8. PR curve of the proposed detection algorithm

下载: 全尺寸图片幻灯片

图 9 本文检测算法的混淆矩阵

Figure 9. Confusion matrix of the detection algorithm

下载: 全尺寸图片幻灯片

图 10 本文算法在测试集上的可视化结果图

Figure 10. Visualization results of the proposed algorithm on the test set

下载: 全尺寸图片幻灯片

图 11 本文检测算法对新的损伤叶片图像识别结果与航线孔探仪目视结果可视化对比图

Figure 11. Visualization results between the proposed detection algorithm and bore scope inspector on new damaged blade images

下载: 全尺寸图片幻灯片

表 1 数据集分类统计信息

Table 1. Dataset classification statistics

类别	标注框/个	具体描述
裂纹	3959	叶片受机械应力、热应力、疲劳、腐蚀等影响产生的细小裂缝
烧蚀	647	叶片在高温环境下由于热应力等相关因素导致的叶片表面损伤
缺块	140	叶片由于机械应力或外部损伤导致的部分材料缺失或脱落
变形	347	叶片由于异常应力或外界载荷导致的叶片变形或打伤

下载: 导出CSV

表 2 实验环境配置

Table 2. Experimental environment configuration

硬件和软件	名称和版本
中央处理器	Intel Core i9-12900H@2.50 GHz
图像处理器	NVIDIA GeForce RTX 3060 Laptop GPU
操作系统	Windows 11 64 bit
程序设计语言	Python 3.9.7
深度学习框架	PyTorch 2.1.0+CUDA 11.6

下载: 导出CSV

表 3 与现有缺陷检测算法精度对比

Table 3. Accuracy comparison against existing object and defect detection algorithms %

算法类型精度指标	SW-YOLO^[22]	YOLOv8-EMA^[30]	本文方法
mAP@50	38.9	37.5	52.8
mAP@50-95	16.7	17.8	23.3

下载: 导出CSV

表 4 与通用目标检测算法的精度对比

Table 4. Accuracy comparison with general-purpose object detection algorithms %

通用目标检测算法模型	mAP@50	mAP@50-95
Faster-RCNN^[31]	28.0	11.5
FCOS^[32]	48.1^②	18.9^③
SSD^[33]	32.6	12.4
Deformable DETR^[34]	25.8	7.9
YOLOv3	38.3	12.1
YOLOv8	41.5^③	20.4^②
YOLOv11	41.3	18.8
YOLOv12	34.8	16.1
本文方法	52.8^①	23.3^①
注：数字上角标标注①、②、③表示检测算法模型对比中检测精度最高的前3名。

下载: 导出CSV

表 5 改进的YOLOv5s网络模型消融实验

Table 5. Improved YOLOv5 network model ablation experiments

模型编号	数据增强	MKF-C3	MSPF	mAP@50/%	mAP@50-95/%	检测速度/ （帧/s）	参数量/10⁶	浮点运算次数/10⁹
1				36.0	15.5	303.2	7.02	6.62
2	√			51.5	22.6	303.1	7.02	6.62
3	√	√		52.0	23.4	284.3	7.64	6.83
4	√	√	√	52.8	23.3	268.5	18.36	10.40

下载: 导出CSV

表 6 数据增强方法消融实验

Table 6. Ablation experiments on data augmentation method %

方法类型	mAP@50	mAP@50-95
YOLOv5s	36.0	15.5
无HSV	36.3（+0.3）	16.2（+0.7）
无随机仿射变换	39.3（+3.3）	18.5（+3.0）
无随机横向纵向翻转	43.1（+7.1）	20.0（+4.5）
本文方法	52.8（+16.8）	23.3（+7.8）

下载: 导出CSV

参考文献(34)

[1]	刘大响, 陈光. 航空发动机飞机的心脏[M]. 2版. 北京: 航空工业出版社, 2015. LIU Daxiang, CHEN Guang. The heart of an aero-engine airplane[M]. 2nd ed. Beijing: Aviation IndustryPress, 2015. (in Chinese LIU Daxiang, CHEN Guang. The heart of an aero-engine airplane[M]. 2nd ed. Beijing: Aviation IndustryPress, 2015. (in Chinese)
[2]	史广, 贺福强, 史红燕, 等. 基于深度学习的航空发动机叶片缺陷检测技术综述[J/OL]. 航空动力学报, (2024-08-30)[2025-07-05]. https://doi.org/10.13224/j.cnki.jasp.20240411. SHI Guang, HE Fuqiang, SHI Hongyan, et al. Review of deep learning-based defect detection techniques for aero-engine blade[J/OL]. Journal of Aerospace Power, (2024-08-30)[2025-06-01]. https://doi.org/10.13224/j.cnki.jasp.20240411. (in Chinese SHI Guang, HE Fuqiang, SHI Hongyan, et al. Review of deep learning-based defect detection techniques foraero-engine blade[J/OL]. Journal of Aerospace Power, (2024-08-30)[2025-06-01]. https://doi.org/10.13224/j.cnki.jasp.20240411. (in Chinese)
[3]	AUST J, MITROVIC A, PONS D. Assessment of the effect of cleanliness on the visual inspection of aircraft engine blades: an eye tracking study[J]. Sensors, 2021, 21(18): 6135. doi: 10.3390/s21186135
[4]	张宝诚. 航空发动机试验和测试技术[M]. 北京: 北京航空航天大学出版社, 2005. ZHANG Baocheng. Aero-engine test and testing technology[M]. Beijing: Beijing University of Aeronautics & Astronautics Press, 2005. (in Chinese ZHANG Baocheng. Aero-engine test and testing technology[M]. Beijing: Beijing University of Aeronautics & Astronautics Press, 2005. (in Chinese)
[5]	温立民. 发动机叶片裂纹检测自适应图像处理算法研究[D]. 西安: 西北工业大学, 2007. WEN Limin. Research on adaptive image processing algorithms for engine blade crack detection[D]. Xi’an: Northwestern Polytechnical University, 2007. (in Chinese WEN Limin. Research on adaptive image processing algorithms for engine blade crack detection[D]. Xi’an: Northwestern Polytechnical University, 2007. (in Chinese)
[6]	LEE D, YOON S, PARK J, et al. Demonstration of model-assisted probability of detection framework for ultrasonic inspection of cracks in compressor blades[J]. NDT & E International, 2022, 128: 102618.
[7]	程宇, 隋岩, 陈俊囡, 等. 航空发动机小径管焊接质量数字射线检测工艺仿真与测试[J]. 焊接技术, 2024, 53(10): 29-33. CHENG Yu, SUI Yan, CHEN Junnan, et al. Simulation and test of digital X-ray inspection process for welding quality of aero engine small tube[J]. Welding Technology, 2024, 53(10): 29-33. (in Chinese CHENG Yu, SUI Yan, CHEN Junnan, et al. Simulation and test of digital X-ray inspection process for welding quality of aero engine small tube[J]. Welding Technology, 2024, 53(10): 29-33. (in Chinese)
[8]	杨戈, 张翼东, 何喜. 航空发动机薄壁涡轮叶片涡流检测技术研究[J]. 无损探伤, 2024, 48(6): 37-40. YANG Ge, ZHANG Yidong, HE Xi. Research on eddy current testing technology of thin-walled turbine blades of aero-engine[J]. Nondestructive Testing Technologying Technology, 2024, 48(6): 37-40. (in Chinese YANG Ge, ZHANG Yidong, HE Xi. Research on eddy current testing technology of thin-walled turbine blades of aero-engine[J]. Nondestructive Testing Technologying Technology, 2024, 48(6): 37-40. (in Chinese)
[9]	宋强, 林国营, 马敬奇, 等. 基于渗透算法和改进型OPTA的裂纹检测算法[J]. 计算机工程与应用, 2017, 53(6): 200-203, 222. SONG Qiang, LIN Guoying, MA Jingqi, et al. Crack detection based on percolation algorithm and improved OPTA[J]. Computer Engineering and Applications, 2017, 53(6): 200-203, 222. (in Chinese SONG Qiang, LIN Guoying, MA Jingqi, et al. Crack detection based on percolation algorithm and improved OPTA[J]. Computer Engineering and Applications, 2017, 53(6): 200-203, 222. (in Chinese)
[10]	杨志军, 陈亮, 刘延雷, 等. 焊缝裂纹缺陷的磁粉检测自动识别方法研究[J]. 制造业自动化, 2016, 38(4): 22-25. YANG Zhijun, CHEN Liang, LIU Yanlei, et al. Research of welds crack defect automatic recognition method of magnetic particle testing[J]. Manufacturing Automation, 2016, 38(4): 22-25. (in Chinese YANG Zhijun, CHEN Liang, LIU Yanlei, et al. Research of welds crack defect automatic recognition method of magnetic particle testing[J]. Manufacturing Automation, 2016, 38(4): 22-25. (in Chinese)
[11]	候小建. 基于深度学习的PCB表面缺陷检测技术研究[D]. 江苏徐州: 中国矿业大学, 2023. HOU Xiaojian. Research on PCB surface defect detection technology based on deep learning [D]. Xuzhou Jiangsu: China University of Mining and Technology, 2023. (in Chinese HOU Xiaojian. Research on PCB surface defect detection technology based on deep learning [D]. Xuzhou Jiangsu: China University of Mining and Technology, 2023. (in Chinese)
[12]	周华锋. 基于计算机视觉的PCBA元器件缺陷检[D]. 广州: 广东工业大学, 2023. ZHOU Huafeng. PCBA component defect detectionbased on computer vision [D]. Guangzhou: Guangdong University of Technology, 2023. (in Chinese ZHOU Huafeng. PCBA component defect detectionbased on computer vision [D]. Guangzhou: Guangdong University of Technology, 2023. (in Chinese)
[13]	GAO L, ZHANG J, YANG C, et al. Cas-VSwin transformer: a variant swin transformer for surface-defect detection[J]. Computers in Industry, 2022, 140: 103689. doi: 10.1016/j.compind.2022.103689
[14]	黄克腾, 王玉奇, 王清, 等. 基于深度学习的表面缺陷检测算法文献研究[J]. 青岛大学学报(工程技术版), 2025, 40(2): 106-113. HUANG Keteng, WANG Yuqi, WANG Qing, et al. Literature review of surface defect detection algorithms based on deep learning[J]. Journal of Qingdao University (Engineering and Technology Edition), 2025, 40(2): 106-113. (in Chinese HUANG Keteng, WANG Yuqi, WANG Qing, et al. Literature review of surface defect detection algorithms based on deep learning[J]. Journal of Qingdao University (Engineering and Technology Edition), 2025, 40(2): 106-113. (in Chinese)
[15]	马作江, 高峰, 刘乃刚. 孔探检测在航空发动机维护中的应用[J]. 黑龙江科学, 2015, 6(6): 131. MA Zuojiang, GAO Feng, LIU Naigang. Application of borescope detection in maintenance of aero engine[J]. Heilongjiang Science, 2015, 6(6): 131. (in Chinese MA Zuojiang, GAO Feng, LIU Naigang. Application of borescope detection in maintenance of aero engine[J]. Heilongjiang Science, 2015, 6(6): 131. (in Chinese)
[16]	SHANG Hongbing, SUN Chuang, LIU Jinxin, et al. Deep learning-based borescope image processing for aero-engine blade in situ damage detection[J]. Aerospace Science and Technology, 2022, 123: 107473. doi: 10.1016/j.ast.2022.107473
[17]	刘宜轩, 程志江, 吴动波, 等. 基于改进YOLOv5的航空发动机叶片表面缺陷检测方法研究[J]. 激光杂志, 2023, 44(7): 57-61. LIU Yixuan, CHENG Zhijiang, WU Dongbo, et al. Research on surface defect detection method of aero-engine blade based on improved YOLOv5[J]. Laser Journal, 2023, 44(7): 57-61. (in Chinese LIU Yixuan, CHENG Zhijiang, WU Dongbo, et al. Research on surface defect detection method of aero-engine blade based on improved YOLOv5[J]. Laser Journal, 2023, 44(7): 57-61. (in Chinese)
[18]	ARTHUR D, VASSILVITSKII S. k-means++: the advantages of careful seeding[R]. San Francisco, California: Stanford University, 2006.
[19]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the Eu-ropean conference on computer vision (ECCV). Cham: Springer International Publishing, 2018: 3-19.
[20]	ZHANG Yifan, REN Weiqiang, ZHANG Zhang, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146-157. doi: 10.1016/j.neucom.2022.07.042
[21]	ZHENG Zhaohui, WANG Ping, LIU Wei, et al. Distance-IoU loss: faster and better learning for bounding box regression[C]//Proceedings of the AAAI Conference on Artificial Intelligence, Palo Alto, California, US: AAAI Press, 2020, 12993-13000.
[22]	何宇豪, 曹学国, 刘信良, 等. 基于SW-YOLO模型的航空发动机叶片损伤实时检测[J]. 推进技术, 2024, 45(2): 2302058. HE Yuhao, CAO Xueguo, LIU Xinliang, et al. Realtime detection of aircraft engine blade damage based on SW-YOLO model[J]. Journal of Propulsion Technology, 2024, 45(2): 2302058. (in Chinese HE Yuhao, CAO Xueguo, LIU Xinliang, et al. Realtime detection of aircraft engine blade damage based on SW-YOLO model[J]. Journal of Propulsion Technology, 2024, 45(2): 2302058. (in Chinese)
[23]	冯雄博. 基于深度学习的航空发动机叶片DR图像缺陷识别[D]. 南昌: 南昌航空大学, 2022. FENG Xiongbo. Defect recognition of aero-engine blade DR image based on deep learning[D]. Nanchang: Nanchang Hangkong University, 2022. (in Chinese FENG Xiongbo. Defect recognition of aero-engine blade DR image based on deep learning[D]. Nanchang: Nanchang Hangkong University, 2022. (in Chinese)
[24]	魏永超, 刘嘉欣, 朱泓超, 等. 改进YOLOv7的航空发动机叶片损伤检测方法[J]. 航空发动机, 2025, 51(1): 133-139. WEI Yongchao, LIU Jiaxin, ZHU Hongchao, et al. Improved YOLOv7 damage detection method for aeroengine blade[J]. Aeroengine, 2025, 51(1): 133-139. (in Chinese WEI Yongchao, LIU Jiaxin, ZHU Hongchao, et al. Improved YOLOv7 damage detection method for aeroengine blade[J]. Aeroengine, 2025, 51(1): 133-139. (in Chinese)
[25]	JIANG Peiyuan, ERGU Daji, LIU Fangyao, et al. A review of yolo algorithm developments[J]. Procedia Computer Science, 2022, 199: 1066-1073. doi: 10.1016/j.procs.2022.01.135
[26]	黎静. 航空发动机涡轮叶片表面裂纹缺陷检测方法[D]. 重庆: 重庆理工大学, 2023. LI Jing. Detection method of surface crack defect of aero-engine turbine blade[D]. Chongqing: Chongqing University of Technology, 2023. (in Chinese LI Jing. Detection method of surface crack defect of aero-engine turbine blade[D]. Chongqing: Chongqing University of Technology, 2023. (in Chinese)
[27]	马超, 王玉娜, 武耀罡, 等. 航空发动机风扇叶片硬物冲击损伤特征[J]. 航空动力学报, 2017, 32(5): 1105-1111. MA Chao, WANG Yuna, WU Yaogang, et al. Hard object impact damage characteristics of aero engine fan blade[J]. Journal of Aerospace Power, 2017, 32(5): 1105-1111. (in Chinese MA Chao, WANG Yuna, WU Yaogang, et al. Hard object impact damage characteristics of aero engine fan blade[J]. Journal of Aerospace Power, 2017, 32(5): 1105-1111. (in Chinese)
[28]	陈纪宗. 基于深度学习的航空发动机涡轮叶片烧蚀检测研究[D]. 四川德阳: 中国民用航空飞行学院, 2024. CHEN Jizong. Research on ablation detection of aero-engine turbine blades based on deep learning[D]. Deyang Sichuan: Civil Aviation Flight University of China, 2024. (in Chinese CHEN Jizong. Research on ablation detection of aero-engine turbine blades based on deep learning[D]. Deyang Sichuan: Civil Aviation Flight University of China, 2024. (in Chinese)
[29]	马瑞阳. 基于深度学习孔探图像的航空发动机缺陷识别研究[D]. 四川德阳: 中国民用航空飞行学院, 2021. MA Ruiyang. Research on defect recognition of aero-engine based on deep learning borescope images[D]. Deyang Sichuan: Civil Aviation Flight University of China, 2021. (in Chinese MA Ruiyang. Research on defect recognition of aero-engine based on deep learning borescope images[D]. Deyang Sichuan: Civil Aviation Flight University of China, 2021. (in Chinese)
[30]	李文龙, 王欣威, 慕丽. 基于改进YOLOv8的航空发动机叶片表面缺陷检测[J]. 组合机床与自动化加工技术, 2024(12): 46-50. LI Wenlong, WANG Xinwei, MU Li. Detection of surface defects on aircraft engine blades based on improved YOLOv8[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2024(12): 46-50. (in Chinese LI Wenlong, WANG Xinwei, MU Li. Detection of surface defects on aircraft engine blades based on improved YOLOv8[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2024(12): 46-50. (in Chinese)
[31]	REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Trans Pattern Anal Mach Intell, 2017, 39(6): 1137-1149. doi: 10.1109/TPAMI.2016.2577031
[32]	TIAN Zhi, SHEN Chunhua, CHEN Hao, et al. FCOS: fully convolutional one-stage object detection[C]//2019 IEEE/CVF International Conference on Computer Vision. Seoul, Korea: IEEE, 2019: 9627-9636.
[33]	LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]//Computer Vision-ECCV 2016. Cham: Springer International Publishing, 2016: 21-37.
[34]	CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[C]//European conf-erence on computer vision. Cham: Springer Internat-ional Publishing, 2020: 213-229.