【作者】 叶南；

【导师】 张丽艳；

【作者基本信息】 南京航空航天大学 ， 航空宇航制造工程， 2010， 博士

【摘要】 随着制造业的高速发展,采用工程应变比例软尺或者工具显微镜的手工应变测量方法在板料成形性能分析中逐渐暴露出测量效率低、精度差等弊端。为了克服手工应变测量方法的各种局限,满足板料冲压成形领域的技术发展要求,本论文系统而深入地开展了基于机器视觉的板料成形性能分析关键技术的研究,包括图像处理技术、相机标定技术、大变形下的位移跟踪技术、图像匹配技术、三维重建技术、板料成形性能参数的计算方法以及系统软、硬件实现技术等,自主研发了基于双目立体视觉的板料成形性能参数测量系统BOSAS(国家软件著作权2010SR008153),以及基于数码相机自由拍摄的复杂板件成形性能分析系统MOSAS(国家软件著作权2010SR025573)。本文的主要内容和创新点总结如下:1、详细讨论了双目立体视觉中的相机标定和三维重建技术;为了提高图像立体匹配和时序匹配的效率和稳定性,提出了基于区域生长法的初始匹配(RGIM);详细实验和分析了数字图像相关(DIC)方法中各种参数选择对重建精度的影响,并得出相应的结论;针对板料处于临界破裂状态下的大变形/大应变的测量难点,提出了时序图像匹配的分段位移传递法FDT。FDT方法可根据变形程度自动分段,对因分段而需要像素圆整所带来的像素精度损失,利用连续介质的变形连续性概念进行精度补偿。实验证明,在局部变形高达170%的情况下,时序图像匹配仍然稳定可靠,能够很好地实现板料成形过程中三维大变形的跟踪。2、对比分析了实验力学领域和计算机视觉领域中三种基于迭代优化的亚像素匹配算法——DIC,IGGA(Iterative Gray-Gradient Algorithm)和LSM(Least Square image Matching),并从理论上证明了这几种方法在数学上的一致性。这一结论不仅加深了对这几种方法的理解,也为不同学科的研究成果相互借鉴相互转换提供了有力的理论支持和技术指导。3、在深入研究了多视图几何理论的基础上,结合编码元识别技术和光束平差优化技术,实现了各次拍摄时相机的相对位置和姿态的精确定位,并提出了多视图几何约束下的三维重建方法,实验证明了该方法对提高重建精度的作用;针对自由拍摄的网格图像中网格节点准确识别和稳健匹配的难点问题,提出了网格节点的初、精两次定位法,并通过对网格节点进行编号同时利用网格拓扑关系和极线几何约束,实现了图像间同名网格节点的自动稳健匹配。在全面实现相机位姿确定、同名网格节点的准确匹配、多视图约束下的网格节点三维重建的基础上,给出了复杂板件网格三维重建的实例并进行了精度验证。4、对板料成形性能参数的测量与分析方法进行了研究和实现。在坐标网格技术中常用的方网格-三角形节点应变计算方法基础上,提出将其与面内形变梯度张量相结合,消除了初始网格边长不正交带来的应变计算误差。针对基于动态应变测量技术进行成形极限曲线FLC测定的问题,提出了根据破裂前应变分布的连续性剔除伪极限应变以及根据应变梯度的变化率选取破裂前临界状态应变分布的数据处理方法;利用动态应变测量技术的特点和优势,提出了一种新的测量塑性应变比r值的方法。该方法利用动态应变的计算结果,自动判断变形试件是否处于均匀塑性变形,并根据处于均匀塑性变形范围的最大/最小主应变计算r值。5、在基础理论和关键算法研究的基础上,分别构建了基于双目立体视觉的板料成形性能参数测量系统BOSAS和基于数码相机自由拍摄的复杂板件成形性能分析系统MOSAS。基于Visual C++平台实现了文中提出的所有算法。将BOSAS系统用于多种板料极限成形曲线和塑性应变比r值的测量,将MOSAS系统用于车门外板某复杂区域的成形性能分析,均取得了良好的测量和分析结果。更多还原

【Abstract】 With high speed development in manufacturing, the traditional manual strain measurement methods which used mylar tape and tool microscope have gradually exposed some disadvantages such as low efficiency and poor accuracy in most sheet metal stamping situations. In order to overcome various limitations of the manual strain measurement methods and meet the requirements of strain measurement in the filed of sheet metal stamping, theory and application researches on non-contact steel strain measurement based on computer vision are deeply and systematically carried out in this thesis. The key techniques include image analysis, camera calibration, displacement tracking under large deformation, image matching, 3D reconstruction, strain calculation, software & hardware im-plementation, etc. The thesis’s main contributions are as follows.1. Key techniques of dynamic strain measurement based on binocular stereo vision have been studied. Binocular calibration and 3D reconstruction in stereo vision theory have been comprehensively discussed. In order to improve efficiency and stability in stereo image matching and tempo-ral-sequential matching, a new technique with Region Growth Initial Matching (RGIM) is proposed. Detailed results and analysis are figured out which demonstrate the effect of different interpolation method, shape function and size of subset on 3D reconstruction accuracy. The premise of DIC(Digital Image Correlation) is that these is no significant difference between two images, aim-ing at this limitation of DIC, a new algorithm call Fractionized Displacement Transfer (FDT) is proposed which can effectively and robustly implement image matching and tracking under large deformation. Experiment demonstrates that even local deformation has reached 170%, the sub-pixel matching still maintains accurate and stable.2. A detail comparative study is carried out on three representative iterative-optimization-based image registration methods, namely DIC using Newton-Raphson iteration, iterative gray-gradient algo-rithm and least square image matching, which are originated and developed in the fields of optical experimental mechanics and stereo vision respectively. The mathematical consistency of the three algorithms is demonstrated. This conclusion not only gives advanced understanding of these meth-ods, but also provides potent theoretical support and technical instruction when the research achievements among these different fields are used for reference and transferred for each other.3. Key techniques of large-scale complex sheet metal forming analysis based on digital camera free-dom taken have been studied. Accuracy 3D reconstruction under multi-view constraint in multiple view geometry theory is proposed. Camera poses calibration is carried out with the combination of coded targets identification and bundle adjustment. On the basis of improvement on the present grid image pre-processing, a coarse-to-fine algorithm is carried out so that the sub-pixel image coordi-nates of grid nodes are extracted and the topological relationship of each node is established. Aim-ing at particularity of grid matching, an automatic matching algorithm for establishing correspon-dence among the grid nodes in multiply images is proposed according to the topological relation and epipolar constraint on the basis of knowing camera position and orientation of every image shot. Experimental results show that the average epipolar distance of all matching points is approxi-mately 0.17 pixel, and the matching method is stable and reliable. Finally, 3D coordinates of grid nods are constructed with the combination of multiple view geometry theory and the proposed grid nodes matching algorithm.4. Determination of sheet mental forming parameters is researched and implemented. A strain calcula-tion with combination of deformation gradient tensor and Green deformation tensor is proposed, and grid node strain is introduced in order to decrease effect of noise. Aiming at FLC determination with dynamic strain measurement, some data-processing skills are proposed,such as elimination of pseudo limit strain according to continuous strain distribution before fracture and selection of criti-cal deformation stage before fracture according to strain gradient change rate. A new technique to determine plastic strain ratio (r value) of sheet metal is proposed utilizing the characteristics of dy-namic strain measurement. With major and minor strain, r value is calculated with conversion of strain in length and width direction be performed by using the property of approximately equality of local strains in the same time of deformation during uniform plastic deformation.5. Dynamic strain measurement system, namely BOSAS (binocular strain analysis measurement sys-tem) based on the binocular stereo vision and strain analysis system, namely MOSAS (monocular strain analysis system) based on multiple view geometry theory are designed and implemented re-spectively. BOSAS has been used for the determination of Forming Limit Curves (FLC) and plastic strain ratio (r value) of aviation aluminum sheet and favorable results have been obtained. MOSAS has been used for the sheet formability analysis of outer plate of the vehicle door, and reliable con-clusion has been achieved.更多还原