【作者】 岳小莉；

【导师】 谭建荣； 陆国栋；

【作者基本信息】 浙江大学 ， 机械设计及理论， 2002， 博士

【摘要】 产品信息的描述、传递和获取是计算机辅助设计和计算机图形学所要解决的一个基本问题，工程图样理解是基于工程图样的产品信息获取行为，是一个工程图学、计算机图形学、认知行为学以及计算机辅助设计技术等多个研究领域相互交叉的课题，也是近三十年来计算机应用技术研究中的一个重要方向。 本文通过对已有工程图样理解算法的分析，探讨了其中存在的问题及其原因，将研究内容重点放在基元体的识别、理解方面。在矢量数据的基础上，对工程图样图形信息进行高层分析和理解，挖掘出其中所蕴涵的工程语义，用于基元体相关信息的获取。所产生的结果，可作为形体三维重构的基础，同时也为CAD／CAPP／CAM系统提供所需要的产品信息。 论文的主要工作包括以下几个方面： 将粒度的思想引入到工程图样理解的分析和研究中，提出基于知识的工程图样信息粒度分析方法，该方法分为三个步骤，基于认知的图样理解目标分解、面向理解子目标的知识划分、基于知识引导的图样信息分类。图样信息分类涉及图样理解的各个层次，为实现产品信息的智能化获取提供了具有合理组织结构的图样信息资源。 提出了基于分类的工程语义获取技术。在工程语义分类基础上，探讨了三类工程语义在基于工程图样的产品信息获取中的作用和关系，即表达方式语义是前提，投影关系语义是核心，尺寸约束语义是引导。重点介绍了表达方式语义、投影关系语义和尺寸约束语义的分类获取技术，以及相切组合关系、相贯组合关系、截交组合关系语义的整合获取技术，为后续基元体信息的识别、理解和获取奠定基础。应用该方法可以简化信息获取的复杂程度，提高计算机信息处理的速度和可靠性。 提出了基于形状相似的非相交基元体识别方法。从分析图形的相似性特征、相似性的内涵以及相似性的性质入手，给出了图形相似度的计算方法；借助图的方式对图形的各相似性特征进行描述，并提出了统一的图形特征编码方式，以此实现图形的由“形”到“数”的转换；提出了一种具有层次性的、以图形特征编码为基础、以相似度为依据的图形相似性判别算法。与传统算法相比，该方法在识别非相交基元体上，具有更好的灵活性、可靠性和更高的效率。 提出了基于形体组合关系的相交基元体信息获取技术。根据已经获取的工程语义，将隐含基元体之间组合关系的相关投影定义为组合因子，由此建立组合关系表，最后借助组合关系表，采用回溯的方法，即从二维的组合因子出发，找出它所隐含的两两基元体在三维空间中的组合关系，进而对该组合关系的主体—基元体的信息进行识别和获取。由于整个过程是以基元体之间的组合关系为切入点，对基元体的形状类型不做限制，因而具有较广的基元体识别范围。 设计了产品信息获取的策略，包括基于尺寸约束信息引导的图形搜索策略、以目标为驱动的控制策略、以及基于多种约束信息的合成策略。该策略基于实际工程图样，融合了人读图的行为特点，在信息获取的方式和过程方面，具有较为显著的智能化特征。最后，实现了一个基于工程图样的产品信息获取原型系统，并给出了一些实例以说明本文所论述方法的应用情况。更多还原

【Abstract】 The description, transfer and acquisition of product information is a basic problem in the area of computer-aided design and computer graphics. Engineering drawing interpretation (EDI) is a kind of acquisition behavior of product information on the base of engineering drawing. It emerged as a vibrant area of interdisciplinary research, attracting the efforts of computer graphics scientists, artificial intelligent scientists, cognitive psychologists and etc. This research can be traced back to sixties in twenty century, whereas the result is still far from the desired purpose up to date.After analyzing the current algorithms, the problems existed in them and the root reasons lying behind are discussed. Then the primitive recognition and interpretation are taken as the research emphasizes of this dissertation. The main process includes performing high-level analysis of the drawing information on vector data in 2D orthographic form, mining engineering semantics from the drawing information, and obtaining the corresponding information of primitive by using engineering semantics. The result could be used as a basis for 3D reconstruction, and to provide product information for CAD (computer-aided design), CAPP (computer-aided process planning) and CAM (computer-aided manufacturing). The main work of the dissertation is as follows:The knowledge-based granularity analysis method of engineering drawing information is put forward by introducing the concept of granularity into the process of EDI. This method is composed of three main steps, namely cognition-based decomposing of drawing interpretation goal, object-oriented partitioning of knowledge and knowledge-directed classification of drawing information. The drawing information classification is expanded into the total procedure of EDI, which could be treated as the information resource for product information acquisition with reasonable organization structure.The classification-based acquisition technique of engineering semantics is presented. On the base of engineering semantics classification, the contribution of different kind of semantics and the relationships among them in EDI are analyzed. The class acquisition of expression mode semantics, projection relationship semantics and dimension constraint semantics, and integration acquisition technology of tangency combined relationship, interpenetrate combined relationship and intersectedcombined relationship are emphatically introduced, which lay the foundation for further primitive recognition. By using this method, information processing could be simplified, and interpretation reliability could also be enhanced.The shape similarity-based recognition method of non-intersected primitive is proposed. By analyzing the features, the implication and the property of graphic similarity, the computation method of similarity degree is provided. Then the features of graphic are described by means of graph, while the method of graphic coding mode is presented. In the end, a hierarchical discrimination algorithm of graphic similarity is put forward based on above methods. Compared with the old ones, some improvements are obtained by using this method, such as pattern recognition scope, speed of pattern searching and matching, etc. Furthermore, the new algorithm has stronger flexibility and stability.The information acquisition technique of intersected primitive based on combined relationship is provided. With the help of obtained engineering semantics, the correlative projections implying combined relationship among primitives are defined as combined factor. Then combined relationship table is established. Finally, the combined relationship table is used to recognize primitives with different kinds of combined relationships. Because the technique establishes the connection between the 2D combined factors and their homologues 3D combined relationships, the search space could be limited and the recognition shape could be extended.The strategy for product information acquisition is designed, including the更多还原