基于孔探视觉的航空发动机叶片计数算法

王严飞; 张吟龙; 刘章波; 邢燕好; 张佳

doi:10.13224/j.cnki.jasp.20240665

基于孔探视觉的航空发动机叶片计数算法

doi: 10.13224/j.cnki.jasp.20240665

王严飞^{1, 2, 3,},
张吟龙^{2, 3,*, ,},
刘章波⁴,
邢燕好¹,
张佳¹

1.
沈阳工业大学信息科学与工程学院，沈阳 110870
2.
中国科学院沈阳自动化研究所机器人学国家重点实验室，沈阳 110016
3.
中国科学院沈阳自动化研究所网络化控制系统重点实验室，沈阳 110016
4.
中国电子系统工程第四建设有限公司，北京 100071

基金项目: 国家自然科学基金（62273332）；广东省基础与应用基础研究基金（2023A1515011363）；辽宁省应用基础研究计划项目（2023JH26/10300028）

详细信息

作者简介:
王严飞（1999-），男，硕士，主要研究方向为机器视觉。E-mail：wangyanfei@sia.cn

通讯作者:
张吟龙（1988-），男，副研究员、硕士生导师，博士，主要研究方向为工业机器人多模态智能感知。E-mail：zhangyinlong@sia.cn

中图分类号: V263.6
计量
- 文章访问数: 734
- HTML浏览量: 397
- PDF量: 46
- 被引次数: 0
出版历程
- 收稿日期: 2024-09-27
- 网络出版日期: 2026-04-05

Borescope vision-based aero-engine blade counting method

Wang Yanfei^{1, 2, 3
,},
Zhang Yinlong^{2, 3,*
, ,},
Liu Zhangbo⁴,
Xing Yanhao¹,
Zhang Jia¹

1.
School of Information Science and Engineering，Shenyang University of Technology，Shenyang 110870，China
2.
State Key Laboratory of Robotics，Shenyang Institute of Automation，Chinese Academy of Sciences，Shenyang 110016，China
3.
Key Laboratory of Networked Control Systems，Shenyang Institute of Automation，Chinese Academy of Sciences，Shenyang 110016，China
4.
The Fourth Construction Company Limited of China Electronics System Engineering，Beijing 100071，China

摘要

摘要:
基于孔探视觉的航空发动机（简称航发）叶片计数是航发日常维护中不可或缺的环节。然而，叶片表面的反光及姿态变化等因素使得现有的叶片计数算法在实时性和准确性方面面临显著挑战。针对上述问题，提出一种基于叶片边缘检测的航发叶片计数算法。为实现叶片高效、准确检测，设计增强型叶片检测Transform（EB-DETR）模型，通过GS-ECA模块来提高叶片特征的表达能力和提取效率。在此基础上，创新性地提出了时空关联状态机（TSCSM）模型，通过捕捉叶片边缘特征点的时空信息来实现航发叶片的计数。实验表明：所提EB-DETR模型检测精度（AP₅₀）为93.17%，相较其他同类规模的先进模型提升了6.54%，同时模型计算复杂度（GFLOPs）降低了35.16%，检测速度达到61帧/s。此外，航发叶片平均计数精度（MCP）为97.06%，比现有计数算法提升了45.82%，可满足航发叶片计数的实际需求。
- 孔探视觉 /
- 航空发动机叶片 /
- 叶片计数 /
- 目标检测 /
- Transformer网络 /
- 状态机
Abstract:
Blades counting in aero-engines based on borescope vision is a critical component of routine maintenance. However, factors such as surface reflections and posture variations pose significant challenges to the real-time accuracy of existing blade counting methods. To address these challenges, a blade counting methods based on edge detection was proposed. To achieve efficient and accurate blade detection, an enhanced blade detection transformer (EB-DETR) model, augmented by the GS-ECA module, was presented to significantly enhance the representation and extraction efficiency of blade features. Furthermore, an innovative temporal-spatial correlation state machine (TSCSM) model was introduced for effectively capturing the temporal and spatial information of blade edge feature points for counting blades in aero-engines. Experimental results demonstrated that the proposed EB-DETR model achieved an average precision (AP₅₀) of 93.17%, reflecting a 6.54% improvement over other state-of-the-art (SOTA) models of similar scale, while reducing Giga floating point operations (GFLOPs) by 35.16% and achieving a detection speed of 61 frames per second (FPS). Additionally, the mean counting precision (MCP) for blades in aero-engines reached 97.06%, exceeding existing blade counting methods by 45.82%, thus satisfying the practical requirements for blade counting in aero-engines.
- borescope vision /
- aero-engine blade /
- blade counting /
- object detection /
- Transformer network /
- state machine

HTML

图 1 航发叶片计数算法

Figure 1. Aero-engine blade counting algorithm

下载: 全尺寸图片幻灯片

图 2 EB-DETR模型

Figure 2. EB-DETR model

下载: 全尺寸图片幻灯片

图 3 普通卷积与Ghost卷积

Figure 3. Standard convolution and Ghost convolution

下载: 全尺寸图片幻灯片

图 4 ECA模块

Figure 4. ECA module

下载: 全尺寸图片幻灯片

图 5 GS-ECA模块

Figure 5. GS-ECA model

下载: 全尺寸图片幻灯片

图 6 ROI选取

Figure 6. ROI selection

下载: 全尺寸图片幻灯片

图 7 叶片计数状态转移图

Figure 7. Blade counting state transition diagram

下载: 全尺寸图片幻灯片

图 8 叶片计数平台

Figure 8. Blade counting platform

下载: 全尺寸图片幻灯片

图 9 数据集扩充示例

Figure 9. Dataset augmentation examples

下载: 全尺寸图片幻灯片

图 10 不同模型检测结果

Figure 10. Detection results of different models

下载: 全尺寸图片幻灯片

图 11 航发叶片计数

Figure 11. Aero-engines counting blades

下载: 全尺寸图片幻灯片

图 12 关键帧叶片计数过程

Figure 12. Key frame blade counting process

下载: 全尺寸图片幻灯片

图 13 叶片旋转角速度

Figure 13. Angular velocity of blade rotation

下载: 全尺寸图片幻灯片

表 1 网络结构对比

Table 1. Comparison of network architectures

输出尺寸	GEResNet	ResNet34
112×112	7×7, 64，步长2	7×7, 64，步长2
$ 56\times 56 $	3×3最大池化，步长2	3×3最大池化，步长2
$ 56\times 56 $	$ \left[\begin{matrix}3\times 3 & 64\\\mathrm{GS}{\text{-}}\mathrm{ECA} & 64\end{matrix}\right]\times 3 $	$ \left[\begin{matrix}3\times 3 & 64\\3\times 3 & 64\end{matrix}\right]\times 3 $
$ 28\times 28 $	$ \left[\begin{matrix}3\times 3 & 128\\\mathrm{GS}{\text{-}}\mathrm{ECA} & 128\end{matrix}\right]\times 4 $	$ \left[\begin{matrix}3\times 3 & 128\\3\times 3 & 128\end{matrix}\right]\times 4 $
$ 14\times 14 $	$ \left[\begin{matrix}3\times 3 & 256\\\mathrm{GS}{\text{-}}\mathrm{ECA} & 256\end{matrix}\right]\times 6 $	$ \left[\begin{matrix}3\times 3 & 256\\3\times 3 & 256\end{matrix}\right]\times 6 $
$ 7\times 7 $	$ \left[\begin{matrix}3\times 3 & 512\\\mathrm{GS}{\text{-}}\mathrm{ECA} & 512\end{matrix}\right]\times 3 $	$ \left[\begin{matrix}3\times 3 & 512\\3\times 3 & 512\end{matrix}\right]\times 3 $

下载: 导出CSV

表 2 叶片计数状态转换表

Table 2. Blade counting state transition table

状态	事件	动作	次态
S0	E1		S1
	E2		S2
	E0		S0
S1	E1		S1
	E2	A2	S2
	E0		S0
S2	E1	A1	S1
	E2		S2
	E0		S0

下载: 导出CSV

表 3 训练参数

Table 3. Training parameters

名称	参数及说明
初始学习率	0.001
优化器	Adam
批处理量	16
训练轮数	50
工作线程数	8

下载: 导出CSV

表 4 叶片检测方法对比结果

Table 4. Comparison results of blade detection methods

检测模型	A_p50/%	A_p50-95/%	GFLOPs/10⁹	检测速度/（帧/s）
SSD300（VGG-16）	78.34	63.75	35.2	39
YOLOV5-L	83.85	67.05	109.1	42
YOLOV10-L	87.45	72.79	120.3	38
RT-DETR-R50	87.14	72.13	136	42
EB-DETR	93.17	77.28	78	61

下载: 导出CSV

表 5 计数结果

Table 5. Counting results

方法	涡轮叶片		压气机叶片		风扇叶片		m_CP/%	m_ADE	R_MSDE
方法	计数结果	C_p/%	计数结果	C_p/%	计数结果	C_p/%	m_CP/%	m_ADE	R_MSDE
光流-统计联合方法	16	65.23	21	62.57	32	71.89	66.56	5.33	9.90
SSD+TSCSM	20	87.16	35	91.32	38	89.49	89.32	3.26	4.31
YOLOv10+TSCSM	22	92.75	36	94.56	43	93.85	93.72	2.34	3.67
RT-DETR+TSCSM	23	95.34	36	96.44	45	94.76	95.51	1.64	4.68
本文方法	23	95.83	38	97.44	47	97.92	97.06	1.00	1.00

下载: 导出CSV

参考文献(22)

[1]	Zhao Bingfeng, Xie Liyang, Li Haiyang, et al. Reliability analysis of aero-engine compressor rotor system considering cruise characteristics[J]. IEEE Transactions on Reliability, 2020, 69(1): 245-259. doi: 10.1109/TR.2018.2885354
[2]	Ballal D R, Zelina J. Progress in aeroengine technology (1939: 2003)[J]. Journal of Aircraft, 2004, 41(1): 43-50. doi: 10.2514/1.562
[3]	Aust J, Pons D. Taxonomy of gas turbine blade defects[J]. Aerospace, 2019, 6(5): 58. doi: 10.3390/aerospace6050058
[4]	胡明辉, 高金吉, 江志农, 等. 航空发动机振动监测与故障诊断技术研究进展[J]. 航空学报, 2024, 45(4): 630194. Hu Minghui, Gao Jinji, Jiang Zhinong, et al. Research progress on vibration monitoring and fault diagnosis for aero-engine[J]. Acta Aeronautica et Astronautica Sinica, 2024, 45(4): 630194. (in Chinese doi: 10.7527/S1000-6893.2024.30194 Hu Minghui, Gao Jinji, Jiang Zhinong, et al. Research progress on vibration monitoring and fault diagnosis for aero-engine[J]. Acta Aeronautica et Astronautica Sinica, 2024, 45(4): 630194. (in Chinese) doi: 10.7527/S1000-6893.2024.30194
[5]	何超, 陈果, 王雨薇. 基于深度学习的航空发动机内部损伤实时检测方法[J]. 航空动力学报, 2023, 38(8): 1857-1864. He Chao, Chen Guo, Wang Yuwei. Real-time detection method of aero-engine internal damage based on deep learning[J]. Journal of Aerospace Power, 2023, 38(8): 1857-1864. (in Chinese doi: 10.13224/j.cnki.jasp.20210381 He Chao, Chen Guo, Wang Yuwei. Real-time detection method of aero-engine internal damage based on deep learning[J]. Journal of Aerospace Power, 2023, 38(8): 1857-1864. (in Chinese) doi: 10.13224/j.cnki.jasp.20210381
[6]	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
[7]	Wong C Y, Seshadri P, Parks G T. Automatic borescope damage assessments for gas turbine blades via deep learning[R]. AIAA-2021-1488, 2021.
[8]	Zou Fuqun. Review of aero-engine defect detection technology[C]//2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference. Piscataway, US: IEEE, 2020: 1524-1527.
[9]	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
[10]	蔡舒妤, 闫子砚. 基于改进YOLOv4的航空发动机小目标损伤检测研究[J]. 航空动力学报, 2023, 38(2): 445-452. Cai Shuyu, Yan Ziyan. Research on small target damage detection of aero-engine based on improved YOLOv4[J]. Journal of Aerospace Power, 2023, 38(2): 445-452. (in Chinese doi: 10.13224/j.cnki.jasp.20220557 Cai Shuyu, Yan Ziyan. Research on small target damage detection of aero-engine based on improved YOLOv4[J]. Journal of Aerospace Power, 2023, 38(2): 445-452. (in Chinese) doi: 10.13224/j.cnki.jasp.20220557
[11]	Harding K G, Gu Jiajun, Tao Li, et al. Blade counting tool with a 3D borescope for turbine applications[J]. Interferometry XVII: Advanced Applications, 2014, 9204: 92040Q. doi: 10.1117/12.2063742
[12]	Von Wrangel M, Kästner M, Reithmeier E. Robust counting of engine blades in aircraft engine video endoscopy[C]//Dimensional Optical Metrology and Inspection for Practical Applications XI. Orlando, US: SPIE, 2022: 16.
[13]	郭立宇, 钟晨杰, 李志鹏. 基于3D卷积神经网络的金属滚轮叶片转动计数算法设计与实现[J]. 计算机科学与应用, 2023(1): 34-48. Guo Liyu, Zhong Chenjie, Li Zhipeng. Design and implementation of metal roller blade rotation counting algorithm based on 3D convolution neural network[J]. Computer Science and Application, 2023(1): 34-48. (in Chinese doi: 10.12677/CSA.2023.131004 Guo Liyu, Zhong Chenjie, Li Zhipeng. Design and implementation of metal roller blade rotation counting algorithm based on 3D convolution neural network[J]. Computer Science and Application, 2023(1): 34-48. (in Chinese) doi: 10.12677/CSA.2023.131004
[14]	Li Jinna, Yang Mingwei, Lewis F L, et al. Compensator-based self-learning: optimal operational control for two-time-scale systems with input constraints[J]. IEEE Transactions on Industrial Informatics, 2024, 20(7): 9465-9475. doi: 10.1109/TII.2024.3384621
[15]	Han Kai, Wang Yunhe, Tian Qi, et al. GhostNet: more features from cheap operations[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, US: IEEE, 2020: 1577-1586.
[16]	Wang Qilong, Wu Banggu, Zhu Pengfei, et al. ECA-net: efficient channel attention for deep convolutional neural networks[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, US: IEEE, 2020: 11531-11539.
[17]	He Kaiming, Zhang Xiangyu, Ren Shaoqing, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, US: IEEE, 2016: 770-778.
[18]	Villa T, Kam T, Brayton R K, et al. Synthesis of finite state machines: logic optimization[M]. New York: Springer Science & Business Media, 2012.
[19]	Everingham M, Ali Eslami S M, Van Gool L, et al. The pascal visual object classes challenge: a retrospective[J]. International Journal of Computer Vision, 2015, 111(1): 98-136. doi: 10.1007/s11263-014-0733-5
[20]	Jiang Peiyuan, Ergu D, 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
[21]	Zhao Yian, Lv Wenyu, Xu Shangliang, et al. DETRs beat YOLOs on real-time object detection[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, US: IEEE, 2024: 16965-16974.
[22]	Liu Wei, Anguelov D, Erhan D, et al. SSD: single shot MultiBox detector[C]//Computer Vision-ECCV 2016. Cham: Springer, 2016: 21-37.