【作者】 杨红娟；

【导师】 周以齐；

【作者基本信息】 山东大学 ， 机械电子工程， 2007， 博士

【摘要】 随着激光测量技术和几何造型技术的发展，以及新技术的不断引入，逆向工程已由最初的仿形制造，发展成为消化、吸收先进技术，实现产品开发和创新设计的重要技术手段。国内外有关逆向工程的研究和应用主要集中在产品的几何形状，即重建产品实物的CAD模型。逆向工程CAD建模的研究经历了以几何形状重构为目的逆向工程CAD建模、基于特征的逆向工程CAD建模和支持产品创新设计的逆向工程CAD建模等三个阶段。以现有产品为原型，还原产品设计意图，注重重建模型的再设计能力成为当前逆向工程CAD建模研究的重点。针对还原产品设计意图和支持产品创新设计的实际问题，国内外专家和学者提出了集成约束的逆向工程特征建模技术和基于特征和约束的的逆向工程CAD建模技术。但在具体研究过程中，集成约束的逆向工程CAD特征建模中复合曲面和复杂曲面特征的自动识别及大型非线性方程组的稳定求解问题仍然是研究的热点。因此，有必要结合CAD建模理论的发展，在现有的技术基础上，进一步完善现有的逆向工程CAD建模理论，从而指导逆向工程CAD建模技术的发展，建立还原产品设计意图和支持产品创新设计的逆向工程CAD模型。本文系统地综述了逆向工程CAD建模领域的研究现状，提出了基于变量化设计的逆向工程CAD建模方法；以激光扫描数据点云的几何特征理解和几何特征生成为手段，研究了相关的关键技术。通过实例验证了基于变量化设计的逆向工程CAD建模方法及其关键技术的正确性。论文的主要内容如下：在理论研究方面，重点剖析了变量化设计造型方法的特点，产品外形几何表示为基于特征的参数化形式，以约束驱动特征模型参数设计，满足当前逆向工程CAD建模反映产品设计意图和支持产品创新设计的需求。研究了变量化设计造型方法的实现技术，利用图论和可靠的数值求解技术可以实现集成约束的逆向工程CAD特征建模中大型非线性方程组的稳定求解。因此，本文将变量化设计造型方法引入逆向工程，提出了基于变量化设计的逆向工程CAD建模方法，为建立反映产品设计意图和支持产品创新设计的逆向工程CAD模型奠定了理论基础。详细分析总结了基于变量化设计的逆向工程CAD建模中的特征分类、表达和约束的分类、表达，进而系统地研究了基于变量化设计的逆向工程CAD建模方法的系统框架。从实物样件的测量数据点云中提炼出实物表面的特征组成以及实物表面特征间的几何约束关系；基于约束驱动的特征模型的优化重建。为建立反映产品设计意图和支持产品创新设计的逆向工程CAD模型提供了一个明确的建模思路。总结研究了基于曲率的特征识别方法，归纳出依据不同大小的窗口计算曲率，特征识别的精度和结果不同。提出了基于多尺度分析的自动特征分割技术思路，为捕捉高层次产品结构提供了新的技术途径。在逆向工程CAD建模中特征表达的基础上，探讨以带有明显几何意义的参数进行特征拟合的方法，为反映产品设计意图提供新的技术手段。研究约束驱动特征模型优化的实现过程，基于约束有向图表示和DSM表示的几何约束系统分解和稳定的数值求解方法。为重建支持产品创新设计的逆向工程CAD模型奠定了技术基础。在技术研究方面，在保形性较好的激光扫描数据点云预处理的基础上，基于多尺度分析提取截面特征，通过截面特征相似性度提取量曲面特征，基于图论和数值求解实现约束驱动特征模型优化重建。研究了保形性较好的数据点云预处理技术。在对激光扫描数据点云全局统计特性分析的基础上，提出了基于弦偏差的脉冲噪声自适应检测算法和对噪声点自适应地选择滤波窗口中的非脉冲噪声数据点进行中值滤波的脉冲噪声自适应滤除方法，有效解决了现有脉冲噪声滤波算法在处理大斜率区域和断线型激光扫描数据点云的多判和漏判的问题。分析了激光扫描数据点云的相关性，基于激光扫描数据点十字形3D邻域的模糊加权均值滤波实现激光扫描数据点云随机噪声滤波，对激光扫描数据点云沿扫描方向进行了平滑，改善了数据点云沿扫描线方向的平滑效果。在对激光扫描数据点的局部统计特性分析基础上，依据信号变化指标因子和信号平均变化率判断其位于平滑区域还是不平滑区域，以改进最小距离法和角度偏差法结合的两步精简法进行数据点云的精简，解决现有数据点云精简方法不能有效保留原始数据点云特征的问题，同时提高了模型重建的精度和效率。深入系统地研究了基于多尺度分析的截面特征提取技术，提出了基于曲率尺度空间的截面特征分割法，实现了激光扫描数据点云截面特征分割中主要曲线特征和次要曲线特征的自动分割。基于多尺度分析改进了区域增长法，以多尺度分析中大尺度下特征检测的结果作为种子区域，实现了种子区域的稳定自动选择。基于多尺度检测特征间的内在相关性指导种子增长，转角作为种子增长度量的区域增长法，避免了种子增长过程中重复的参数拟合。分析了激光扫描数据点投影高度函数的统计特性，改进了基于投影高度函数的截面特征直线和圆弧的自动识别准则。讨论了基于二次曲线不变量的圆锥曲线特征自动识别方法。系统地研究了截面曲线特征的表达和拟合方法，重点分析了圆锥曲线特征，平面上任意圆锥曲线可以通过标准圆锥曲线经过中心平移变换和旋转变换得到，变换矩阵参数可以作为圆锥曲线特征拟合的参数。推导了基于圆锥曲线标准表达形式参数，平移矩阵参数和旋转矩阵参数进行圆锥曲线特征拟合的理论。研究了基于截面特征相似性度量的曲面特征分割方法，在截面轮廓弧长和切倾角的形状描述基础上，通过截面特征相似性度量准则实现复杂曲面数据点云的曲面特征自动分割，能够处理分支和融合，较好地表达反映设计意图的曲面细节特征。详细地研究了曲面特征的识别方法，重点研究了点云切片算法提取扫掠曲面、旋转曲面和蒙皮曲面等简单自由曲面造型特征的特征线的方法。系统地研究了逆向工程曲面特征的表达和拟合方法。基于二次曲面的统一表示方程，采用最小二乘法进行参数拟合，通过对比其标准形式和二次曲面的统一表示方程，确定圆柱曲面特征带有明显几何意义的参数圆柱轴线法矢量、圆柱的中心和半径。对于圆锥曲面，二次曲面的统一方程中的参数没有明显的几何意义，无法表达设计的意图和设计的过程。基于圆锥曲面可以通过在其标准表达形式的基础上，经过中心平移变换和旋转变换得到的认识，将变换矩阵参数作为圆锥曲面特征拟合的参数。推导了以圆锥曲面标准表达形式的参数、平移矩阵参数和旋转矩阵参数为特征参数进行二次曲面中圆锥曲面特征拟合的理论。研究了约束驱动特征模型优化重建的实现过程。建立了几何约束系统的约束有向图和DSM矩阵，通过DSM矩阵分割算法消除几何约束系统中耦合约束，提出了基于多尺度特征的凝聚算法，实现逆向工程CAD建模中复杂曲面几何约束系统的简化和分解。分析了截面曲线特征和曲面特征的约束表达，建立了约束驱动特征模型优化的数学模型。讨论了罚函数乘子法和拟牛顿法中BFGS法求解逆向工程CAD建模中约束驱动特征模型优化问题的原理和步骤。讨论了实现基于变量化设计的逆向工程CAD建模方法的基本步骤，依据优化的特征模型参数在通用CAD软件中完成CAD模型重建。综上所述，本文提出的基于变量化设计的逆向工程CAD建模方法，完善了现有的逆向工程CAD建模理论。并对实现该方法的关键技术特征提取技术及基于约束驱动的特征模型优化技术进行了深入的研究。将多尺度理论引入逆向工程，提出了基于多尺度分析的截面特征分割方法，并以带有明显几何意义的参数进行了特征拟合，为从激光扫描数据点云提取反映原始设计意图的特征提供了新的技术途径。提出了基于多尺度特征的凝聚算法实现几何约束的简化，基于罚函数乘子法和拟牛顿法的中BFGS法进行约束驱动特征模型优化的数值求解，为实现产品的创新设计奠定技术基础。更多还原

【Abstract】 With development of measurement and geometric modeling and introduction of new techniques, reverse engineering has become an important means of digesting and absorbing advanced technology for new product design and innovation design instead of original product copy. Researches and applications in reverse engineering field mainly focus on reconstructing CAD model of geometric shape from physical product. Three stages of CAD modeling methodology in reverse engineering are investigated systematically, such as CAD model reconstruction for product geometric shape, feature based CAD model reconstruction and CAD model reconstruction supporting product innovation design. Based on the physical product, recovering original design intent and supporting product innovation design are highly regarded as the research emphasis of CAD modeling methodology in reverse engineering.In order to recover product design intent and support product innovation design, some techniques were proposed, such as feature modeling incorporating constraint and CAD modeling method based on feature and constraint. In the research of feature modeling incorporating constraint process, automatic feature identification of composite surface and complex surface and static solution of large scale non-linear equation set are still research hot topic. Based on developed feature modeling technology in reverse engineering, it is necessary to enrich CAD modeling theory in reverse engineering by integrating CAD modeling methodology. It is helpful and valuable to improve technologies of CAD model reconstruction in reverse engineering. CAD model, which supports product innovation design and design intent restoration, is reconstructedThe research situation of methods, techniques of CAD model reconstruction in reverse engineering is investigated systematically. Variation design based CAD modeling methodology in reverse engineering is proposed. Key techniques related to the proposed methodology are studied by geometric feature understanding of data point and geometric feature modeling. The correctness and validity of Variation design based CAD modeling methodology in reverse engineering and corresponding technology is proved by examples.The main contents are as follows:From the viewpoint of theory and methodology, characteristics of variation design modeling are analyzed. Product geometric shape is represented with feature-based parameters. Optimization and modification of feature model is driven by geometric Constraint. Characteristics of variation design modeling satisfy requirement of CAD model reconstruction on recovering design intent and supporting product innovation design. Techniques related to variation design modeling are investigated, such as constraint decomposition based on theory of graph and constraint numerical solving method. Stable numerical solution of large scale non-linear equations is achieved for feature model reconstruction by incorporating constraints in reverse engineering. Variation design modeling is introduced into reverse engineering. Variation design based CAD modeling methodology in reverse engineering is proposed, which lay the foundation in theory for reconstructing CAD feature model supporting product innovation design.Classification and presentation of features and constraints are analyzed and summarized thoroughly in variation design based CAD modeling methodology in reverse engineering. The architecture of variation design based CAD modeling methodologies is studied systematically. Surface feature and geometric constraints are extracted from laser scanning data point. CAD model is reconstructed with optimized feature parameters driven by constraint. The proposed methodology provides a clear modeling way for CAD modeling in reverse engineering, which support product design intent extraction and innovation design.Curvature based feature identification method are synthesized. Feature identification is different according to curvature calculated with laser scanning data point in different size window. A technical idea of automatic feature segmentation is proposed based on multi-scale analysis. A new technical approach is provided for capturing high level product structure.Based on feature representation in CAD model reconstruction, feature fitting method with distinct geometric significance parameters is discussed. A new technical way is developed for recovering product design intent.The process of feature model optimization driven by constraints is implemented. Constraint is decomposed with constraint directional graph and DSM. Stable numerical solving method of geometric constraint systems is discussed. Technological basis is provided for reconstructing CAD model supporting product innovative design.In technical research, preprocessing techniques of laser scanning data point are developed for preserving shape. Sectional feature extraction is achieved based on multi scale analysis of laser scanning data point. Surface feature extraction is accomplished with similarity measure of sectional curve feature. Feature model optimization driven by constraint is studied, including constraint decomposition and numerical solution.Global statistical characteristics of laser scanning data point are investigated. Adaptive detection method of impulse noise is proposed based on the chord deviation. The impulse noise filtering is achieved with median filter by adaptively choosing the data that is not impulse noise in filter window. The proposed method can effectively solve the problem of previous impulse noise filtering methods on processing laser-scanning data of sharp area and data of broken-line area with unwanted detection and skipped detection. A random noise filtering method is presented within 3D neighbors of laser scanning data, which outperforms other filters in noise smoothing along the scan-line way and achieves good result in noise smoothing along the scan way. Local statistical characteristics analysis of laser scanning data point is surveyed. According to signal change factor and signal mean change rate, area property of laser scanning data point is determined with sharp area or smooth area. A two-steps data reduction method including modified least distance method and angle deviation is presented to improve the precision and efficiency of model reconstruction. The presented method can solve the problem of previous data reduction method that can not effectively preserve feature information of data point.Theory of multi scale is introduced into section feature curve segmentation. Based on curvature scale space, automatic feature segmentation of sectional composite curve is presented to obtain segmentation of primary curve primitives and secondary curve primitives. A seed growing segmentation method is developed based on multi scale analysis. Automatic selection of seed region is achieved with feature detection at large scale. According to information correlations among multi scale feature detection, a seed growing algorithm is studied with the homogeneity criteria relative angle to obtain feature segmentation. The statistic characteristic of projection height function is analyzed. Improved criterions of sectional curve feature classification are presented to identify line and arc. Automatic identification method of conics is studied based on the curve geometric invariants. Representation and fitting method of curve feature are systematically studied. A general conic curve can be modeled by translating and rotating the standard conic. The transformation matrix parameters can be seen as parameters of conic curve feature fitting. A general conic curve fitting is achieved with the standard conic curve parameters and translation matrix parameters and rotation matrix parameters.Surface feature segmentation method is studied based on similarity measure of sectional curve feature. The shape description of arc length and rotation angle is discussed. Complex surface is segmented into individual surfaces according to similarity measure rules. Automatic surface identification method is investigated. Feature extraction of simple free form modeling surface is achieved by slicing the laser scanning data point, such as sweeping surface, revolution surface and lofting surface. Surface representation and surface fitting method with explicit geometric parameter are systematically studied. Cylinder is fitted by least square method with the general surface representation. Feature parameters are achieved by comparing the coefficient of standard representation and general representation. As for cone, parameter of general representation has no explicit geometric meaning and can not reflect design intent. A general cone can be modeled by translating and rotating the standard conic surface. The transformation matrix parameters can be seen as parameters of cone feature fitting. A general cone feature fitting is achieved with the standard conic cone parameters and translation matrix parameters and rotation matrix parameters.Feature model optimization driven by constraint is studied, including constraint decomposition and numerical solution. Representation of directed graph and Design Structure matrix (DSM) of geometric constraint system are discussed. Coupled constraints of complex surface are eliminated by DSM partitioning algorithm. A new clustering method based on multi scale feature is proposed to reduce and decompose the geometric constraint system for identifying the constraint subset. Constraint representation between feature curves of composite curve is investigated. And constraint representation between feature surfaces of complex surface is discussed. Mathematical models of CAD model optimization are built with exponential penalty to translate constraint optimization into unconstraint optimization. The principle and steps of BFGS in Quasi-Newton method are studied for feature model optimization problems of variation design based CAD model reconstruction in reverse engineering. Steps of variation design based CAD modeling methodology are summarized. According to the optimized feature parameters, CAD model is reconstructed with the general CAD software UG.As stated above, variation design based CAD modeling methodology is proposed to improve the theory of CAD model reconstruction in reverse engineering. Key techniques related to this methodology are studied, such as feature extraction and constraint driving feature optimization for CAD model reconstruction. The theory of multi scale is introduced into reverse engineering. Based on curvature scale space, a section feature automatic segmentation method is presented. Feature fitting method with explicit geometric meaning parameters is systematically studied. Some innovation techniques are presented to recover the original design intent feature from laser scanning data point. Based on multi scale feature, a new clustering method is proposed to reduce the geometric constraint system. The constraint driving feature model optimization is achieved by exponential penalty and BFGS in Quasi-Newton method. All these researches provide the basis for product innovation design.更多还原