燃油泵航插与锁紧保险的缺陷检测算法研究  

Research on Defect Detection Algorithm for Fuel Pump Aviation Socket and Locking Fuse

作　　者：王宇鑫 陆永华[1] 刘江伟 朱赟 WANG Yuxin;LU Yonghua;LIU Jiangwei;ZHU Yun(College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China;China Aviation Engine Control System Research Institute,Wuxi 214000,China)

机构地区：[1]南京航空航天大学机电学院,江苏南京210016 [2]中国航发控制系统研究所,江苏无锡214000

出　　处：《测试技术学报》2024年第6期668-677,共10页Journal of Test and Measurement Technology

基　　金：国家自然科学基金资助项目(51975293);航空科学基金资助项目(2019ZD052010)。

摘　　要：燃油泵是航空发动机燃油控制系统的重要组成部件,其表面形态复杂、缺陷种类繁多。航空插座和锁紧保险是燃油泵上广泛应用的零件,由于工艺水平有限、工人操作不当等因素,在工业生产中容易出现多种表面缺陷,因此对其进行缺陷检测具有重要意义。针对传统检测方法检测精度低、效率低的问题,提出了基于特征提取的表面缺陷检测算法和基于深度学习的小目标检测网络、旋转目标检测网络,以满足不同场景的检测需求。实验结果显示,改进的小目标检测网络相较于传统方法准确率提高了11.34%,相较于原始YOLOv8s网络,mAP50、mAP50:95分别提升了6.6%、4.7%,参数量下降34%。改进的旋转目标检测网络相较于传统方法准确率提高了10.38%,相较于原始ShuffleNet-V2网络,mAP50、mAP50:95分别提升了4.6%、2.6%,参数量下降17.6%。实验结果表明,提出的缺陷检测算法具有较高的检测精度,能够实现燃油泵航插、锁紧保险的快速准确检测。The fuel pump is an important component of the aviation engine fuel control system,with complex surface morphology and a wide variety of defects.Aviation sockets and fuses are widely used parts on fuel pumps,which are prone to multiple surface defects in industrial production due to factors such as limited craftsmanship and improper operation by workers,so it is of great significance to detect defects on them.Aiming at the problems of traditional inspection methods such as low detection accuracy and low efficiency,a surface defect detection algorithm based on feature extraction,a small object detection network based on deep learning,and a rotating object detection network is proposed to meet the detection needs of different scenarios.The experimental results show that the improved small object detection network has an accuracy increase of 11.34%compared to traditional methods.Compared to the original YOLOv8s network,the mAP50 and mAP50:95 has increased by 6.6%and 4.7%respectively,and the number of parameters has decreased by 34%.Compared to traditional methods,the improved rotating object detection network has an accuracy increase of 10.38%.Compared to the original ShuffleNet-V2 network,the mAP50 and mAP50:95 has increased by 4.6%and 2.6%respectively,and the number of parameters has decreased by 17.6%.The experimental results indicate that the defect detection algorithm has high detection accuracy and can achieve fast and accurate detection of fuel pump aviation sockets and locking fuse.

关 键 词：燃油泵表面缺陷检测 特征提取 小目标检测 旋转目标检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]