集成改进半监督分割方法的航空发动机短舱声衬孔检测系统  

Aero-engine nacelle acoustic hole detection system integrating improved semi-supervised segmentation method

作　　者：董清钰 梅标 傅云 羊荣金 朱伟东 DONG Qingyu;MEI Biao;FU Yun;YANG Rongjin;ZHU Weidong(Polytechnic Institute,Zhejiang University,Hangzhou 310015,China;Quanzhou Institute of Equipment Manufacturing,Haixi Institutes,Chinese Academy of Sciences,Quanzhou 362100,China;Xizi Spirit Aerospace Industry(Zhejiang)Ltd.,Hangzhou 311222,China)

机构地区：[1]浙江大学工程师学院,浙江杭州310015 [2]中国科学院海西研究院泉州装备制造研究中心,福建泉州362100 [3]浙江西子势必锐航空工业有限公司,浙江杭州311222

出　　处：《光学精密工程》2024年第23期3457-3468,共12页Optics and Precision Engineering

基　　金：杭州市重大科技创新项目资助(No.2022AIDZ0026);浙江省“尖兵”“领雁”研发攻关计划资助(No.2022C01134)。

摘　　要：针对飞机发动机短舱复合材料声衬上大规模制孔需求,以及复合材料表面纹理复杂、声衬孔尺寸小、数量大的问题,开发了一套视觉检测系统,用于短舱声衬机器人多主轴制孔系统。为解决复合材料声衬孔缺乏标注数据的问题,提出改进的半监督分割方法,对复合材料声衬孔进行准确分割。基于半监督分割结果提出一种基准孔检测方案,利用由粗到精的几何参数拟合算法,实现了制孔前的基准孔准确检测。利用穿孔率计算公式和分割结果,实现了复合材料声衬的穿孔率检测。最后,基于Labview和Python开发了可实现声衬孔穿孔率和基准孔自动检测的视觉检测系统。对声衬样件进行检测,结果表明,改进半监督方法的mIoU达到95.70%,节省了70%的数据标注量,并且减少了训练检测所需的参数量。集成该半监督分割方法的视觉检测系统对穿孔率的检测相比人工检测结果相差仅0.023%,对基准孔检测的准确度也较高。该视觉检测系统可以满足航空发动机短舱高质量制造的需求。该系统对基准孔和穿孔率的检测耗时较短,能够满足大规模声衬孔制孔和检测所需的高效率要求。将半监督方法引入到视觉检测系统中,对于缺少标注数据的声衬机器人多主轴制孔工业场景具有重要意义。To address the need for large-scale drilling of acoustic holes in aero-engine nacelle composite liners,as well as challenges posed by complex material surfaces and the small size and high density of holes,a visual detection system was developed for a robotic multi-spindle drilling system.To overcome the lack of labeled data for composite acoustic holes,a semi-supervised segmentation method was introduced for precise segmentation.Based on these results,a reference hole detection scheme was designed using a geometric parameter fitting algorithm to accurately identify reference holes prior to drilling.Porosity detection was achieved using a porosity calculation formula combined with segmentation outcomes.A visual detection system,integrating LabVIEW and Python,was developed to automate the detection of acoustic hole porosity and reference holes.Tests on composite liner samples demonstrated that the improved semi-supervised method reduces labeled data requirements by 70%while achieving an mIoU of 95.70%.Training and detection parameters were significantly reduced.The visual detection system,incorporating the semi-supervised method,achieved a porosity detection variance of only 0.023%compared to manual results and demonstrated higher accuracy in reference hole detection.The visual detection system satisfies the high-quality manufacturing standards of aero-engine nacelles and meets efficiency demands for drilling and detection.The integration of semi-supervised methods into the system is a significant advancement for multi-spindle robotic drilling in scenarios with limited labeled data.

关 键 词：视觉检测 航空发动机短舱 复合材料声衬孔 半监督语义分割 机器人制孔 

分 类 号：V260.0[航空宇航科学与技术—航空宇航制造工程] TH691.9[机械工程—机械制造及自动化]