【作者】 郝敬宾；

【导师】 方亮；

【作者基本信息】 中国矿业大学 ， 机械制造及其自动化， 2011， 博士

【摘要】 模型分割技术可解决待加工模型或零件因结构复杂、体积过大(超出材料尺寸或设备加工范围)等造成的整体加工困难、材料利用率低等问题,同时可以有效地降低加工设备要求、提高生产效率。针对巨型三维模型的快速原型制造而言,模型分割的目标不是单纯地把大模型分割成尺寸符合加工要求的小模型,而是要将大尺寸复杂模型分割成形状简单、尺寸适中、方便加工和装配的子模型,以提高快速成型设备的加工能力和效率。针对快速成形系统的通用数据接口--STL模型,存在大量数据冗余且没有拓扑信息,无法直接进行表面曲率分析的问题,分析了三维模型的形体表达法与数据表示,使用Hash表算法对模型面片顶点进行了归并,研究了基于正向边结构的拓扑重构,构建了适用于边曲率分析的数据结构。针对点曲率分析计算复杂,聚类和迭代运算量大的问题,研究了基于边曲率分析的特征边界提取算法。使用二面角、周长比和凸凹性三种曲率参数对模型表面进行曲率分析,提取信息直指可用于模型分割的特征边界。在此基础上,研究了基于遗传算法的特征边阀值选取,以最大类间方差作为适应度函数,该方法提取的特征边数目比预设阀值法减少了一半以上,且有效的特征边界都得以保留。针对提取出的特征边在边链表中处于离散状态,必须链接起来才能用于模型分割的问题,研究了基于最优拟合平面的特征边链接算法。采用最小二乘法来生成特征边集的最优拟合平面,在拟合平面内合并孤立边,并使用Dijkstra最短路径法闭合特征边链。该方法可以有效闭合特征边界和去除噪声边,得到的特征环可以直接作为模型分割的边界。针对用户有特殊分割要求或是必须进行手动分割的情况,研究了基于归纳学习法的交互式分割算法。在OpenGL环境下实现了用户对三维模型的交互式操作,由用户手工选取所要分割的位置(可以是点、边、面,或是划线选取),根据特征环和分割位置之间的位置和相似度关系建立决策树,以指导机器进行归纳学习,生成分割位置上的最优分割边界,以辅助用户完成模型分割。针对模型自动分割中存在分割方案众多的问题,研究了基于多目标优化的智能化分割算法。该算法在保证子模型可被加工的前提下,综合考虑分割次数、子模型复杂度、材料使用率和分割面平整度等影响因素,建立了多目标优化函数,使用层次分割策略和多目标遗传算法来获取最优分割方案,以指导机器对模型进行自动分割。针对现有模型分割研究中缺少有效的子模型装配结构生成算法,提出了一种相似形装配结构的自动生成算法。根据模型的切割轮廓和自身结构,基于Voronoi图的二等分法和切平面投影法,自动构建形状相似且尺寸适中的公/母装配结构。相似形装配结构的生成不需要单独建模,也无需计算合并位置,而是直接与子模型的闭合操作一并完成的。装配表面的三角化精度一致,且留有配合公差,有效地保证了子模型的装配精度。基于以上模型分割和装配结构生成算法的研究,使用Visual C++编程软件,实现了快速原型制造的大尺寸模型分割和装配结构生成系统的开发。通过对大量实验模型进行的特征边界提取、交互式分割、智能化分割、以及相似形装配结构生成的实例,验证了模型分割和装配结构生成系统的有效性和通用性。该论文有图87幅,表3个,参考文献170篇。更多还原

【Abstract】 Model partitioning of rapid prototyping (RP) makes it possible to fabricate arbitrary large-scale models which are larger than the maximum dimension of the current RP machines. For the large-scale rapid prototype manufacturing, the objective of model partitioning is not just dividing a large model into a set of small sub-model, but decomposing the large complex model into several sub-models which are simple, moderate size and easy to fabricate and assembly.Aiming at the data redundancy and no topology information of STL model that is the common data interface of RP system, the shape representation and data structure of STL file are analyzed. The Hash-table algorithm was used to merge the vertices of the model. The Forward edge structure was proposed to reconstruct the data topology of the model. The model denoising based on two-sided filter was used to improve the quality of the model surface.Aiming at the complex clustering and iteration of the vertex-based curvature estimation, the edge-based curvature estimation was proposed to extract the feature edge. The calculation of the edge-based method is simpler than the vertex-based method. The three curvature parameters are dihedral angle, perimeter ration and convexity. The genetic algorithm (GA) was used to determinate the threshold of feature edge. The fitness function is the maximum variance of classes (MVC).Aiming at the discreteness of feature edges in the edge chain, the best-fit plane (BFP) was proposed to group and link feature edges. The similar and adjacent feature edges were firstly grouped to be feature edge sets. The BFP of a feature edge set was calculated by using the least square method (LSM). The isolated feature edges can be grouped or deleted based on BFP. The Dijkstra’s algorithm was used to close the incomplete feature boundaries.Aiming at the requirement of manual partitioning, the interactive partition method was researched. The interactive operation of STL model was achieved in the OpenGL environment. The partition positions (vertex, edge, facet, or scoring) can be selected by the operator using the mouse. The decision tree of cutting boundaries was constructed according to the relative position and similarity of the partition positions and the feature loops. Based on the inductive learing (IL), the optimal cutting boundaries were created on the partition positions for assisting the manual partitioning. Aiming at the requirement of automatic partitioning, the intelligent partition method was researched. Generally, the large complex model has a lot of partition schemes. It is different to select the best scheme by the enumeration method. The multiobjective optimization (MOO) algorithm was proposed to select the best partition scheme. On condition that all sub-models should be smaller than the maximum dimension of the RP machine, cutting times, complexity of sub-models, utilization ratio of materials and planarity of cutting boundaries were synthetically considered. GA was also used to generate the optimal partition scheme for guiding the automatic partitioning.Aiming at the effective construction of assembly features on the sub-models, the similar-shaped assembly feature was researched. According to the outside cutting contour, the similar-shaped contour was constructed by using the Voronoi diagram algorithm. The maximum assembly depth was calculated by using the projection method. Based on the similar-shaped contour and the maximum assembly depth, the similar-shaped assembly features were automatically created for facilitating the assembly of sub-models.Based on the above research of model partitioning and assembly feature, a prototype system of model partition and assembly feature construction for large-scale rapid prototype manufacturing was developed by using Visual C++. More than ten experimental models were used to validate the validity and generality of the proposed algorithms and the developed system.更多还原