【作者】 刘琪；

【导师】 霍炬；

【作者基本信息】 哈尔滨工业大学 ， 电气工程， 2025， 博士

【摘要】 飞行器柱形部件几何尺寸测量是飞行器柱形部件生产质量检测中的重要环节,其对于保证柱形部件的制造质量、降低故障风险、提升飞行器运行可靠性和安全性有重要意义。视觉测量技术以其非接触、适应性强、精度高、可视化效果好等优点在几何尺寸测量中得到广泛应用,但是现有的视觉测量方法难以同时实现飞行器柱形部件体整尺寸、关键结构几何尺寸等信息的测量。针对上述问题,本文开展基于多目视觉的飞行器柱形部件几何尺寸测量方法研究,主要研究工作包括以下几个方面:(1)视觉测量任务中相机参数的精确标定是实现准确测量的基础,针对圆形标定靶标成像时存在的圆心定位偏差及相机参数优化算法易陷入局部最优解的问题,提出基于适应度距离平衡混沌映射海洋捕食者算法的相机参数优化方法。设计二项异性中心对称标定靶标,一定程度上克服圆形靶标成像时的偏心误差。提出适应度距离平衡混沌映射海洋捕食者算法,在算法寻优过程中通过混沌映射和适应度距离平衡优化参数更新策略,增强算法全局寻优能力,实现相机参数的高精度标定。(2)飞行器柱形部件表面多个相似关键结构的提取是几何尺寸测量的前提和保障。针对传统边缘检测、顶点检测等方法难以区分相似特征的问题,提出基于检测区域线段筛选的飞行器柱形部件关键结构提取方法。构建轻量化目标检测网络,通过目标检测网络结构轻量化设计、模型剪枝、知识蒸馏等方法实现目标检测模型轻量化以及多个相似结构的准确定位。提出检测区域线段筛选方法,在目标检测区域内对关键结构边界进行筛选和提取,利用线段的连续性和延伸性克服极端光照、噪声在关键结构提取中的影响,实现飞行器柱形部件关键结构提取。(3)针对飞行器柱形部件几何尺寸测量时单个相机视野有限,难以同时准确获取飞行器柱形部件整体结构信息和关键结构位置信息的问题。提出基于多特征信息融合的图像拼接方法。设计混合深度特征提取匹配网络,实现待拼接图像中的特征点、线的精确提取匹配。建立基于多特征信息融合的能量方程,利用匹配的特征点、线重构匹配平面,通过特征点、线、平面联合约束构建图像变换模型,实现飞行器柱形部件的图像拼接,保持关键结构在图像中分辨率的同时还原飞行器柱形部件轴向关键结构的相对位置信息。(4)针对飞行器柱形部件几何尺寸测量的实际需求,设计飞行器柱形部件端面整体尺寸、关键结构几何尺寸测量方法,搭建飞行器柱形部件几何尺寸视觉测量系统。结合椭圆提取、空间圆成像理论及特征点三维重建技术,实现飞行器柱形部件端面内外径与长度的测量。通过关键结构提取、图像拼接及关键结构顶点三维坐标重建技术,结合欧氏几何计算方法,设计包含关键结构尺寸以及相对位置的测量方案,实现关键结构长、宽、高、相对距离以及相对角度的测量。更多还原

【Abstract】 The geometric dimension measurement for cylindrical components of aircraft is a critical part in the production quality inspection.It plays an essential role in ensuring the manufacturing quality of the components,reducing failure risks,and enhancing the operational reliability and safety of the aircraft.Visual measurement technology,with its non-contact,strong adaptability,high precision,and excellent visualization capabilities,has been widely applied in geometric dimension measurement.However,existing visual measurement methods struggle to simultaneously measure the overall dimension of the cylindrical component,as well as the geometric dimension of key structure.To address these challenges,this thesis investigates a geometric dimension measurement method for cylindrical components of aircraft based on multi-vision.The main research contributions include the following:(1)Accurate camera parameters calibration is fundamental for precise visual mea-surement.To address the challenges of center positioning deviation in circular calibra-tion targets and the susceptibility of camera parameters optimization algorithms to local optima,this thesis proposes an optimization method based on a fitness distance balance chaotic marine predator algorithm.A binomial centrosymmetric calibration target is de-signed to mitigate eccentricity errors in circular target imaging.The proposed algorithm enhances global optimization capability by integrating chaotic maps and a fitness distance balance strategy during parameters updates,thereby achieving high-precision camera cal-ibration.(2)The accurate extraction of multiple similar key structures on aircraft cylindrical components is essential for reliable geometric dimension measurement.Traditional edge and vertex detection methods struggle to distinguish similar features.To address this issue,a key structure extraction method based on detection region line segment filtering is pro-posed.A lightweight object detection network is designed using structural optimization,model pruning,and knowledge distillation to achieve accurate localization of multiple similar structures.Additionally,a detection region line-segment screening method is in-troduced to refine key structure boundaries within the object detection.By leveraging the continuity and extension properties of line segments,this approach mitigates the effects of extreme lighting and noise,ensuring robust extraction of key structure.(3)Due to the limited field of view of single camera,capturing both the overall struc-ture and key feature locations of aircraft cylindrical components simultaneously is chal-lenging.To address this,an image stitching method based on multi-feature information fusion is proposed.A hybrid deep feature extraction and matching network is designed to achieve precise extraction and matching of feature points and lines in images to be stitched.An energy equation is formulated based on multi-feature fusion,utilizing matched feature points and lines to reconstruct feature planes.By constraining the transformation model with feature points,lines,and planes,the method enables seamless image stitching while preserving the resolution of key structures and accurately restoring their axial relative po-sitions.(4)To meet the practical requirements of geometric dimension measurement of air-craft cylindrical components,a measurement system is developed for end-face dimension and key structure geometric dimension.By integrating ellipse extraction,spatial circle imaging theory,and 3D reconstruction techniques,the system enables precise measure-ment of the inner and outer diameters as well as the length of the component.Key struc-ture dimension and relative positions are obtained through key structure extraction,image stitching,and 3D coordinate reconstruction of key structure vertices.Using Euclidean geometry,the system measures structural length,width,height,relative distances,and angles,ensuring accurate and comprehensive geometric dimension assessment.更多还原