【作者】 刘金山；

【导师】 廖文和；

【作者基本信息】 南京航空航天大学 ， 航空宇航制造工程， 2007， 博士

【摘要】 提高夹具设计的质量和效率在产品生产制造中具有重要的意义,计算机辅助夹具设计技术(Computer Aided Fixture Design ,CAFD)的提出,为夹具设计提供了一条非常有效的途径。但是,当前基于CAD的夹具设计系统大多还停留在解决设计过程中的几何建模层次上,即使引入了一些智能机制,但在处理复杂夹具设计时仍然存在一些问题,主要表现在大多数CAFD系统忽略了工件的几何特征或者其求解机制很少结合这些几何特征,使得实现智能化的夹具详细设计困难。另一方面,工件几何拓扑特征直接决定了夹具结构,在应用广泛的基于实例推理的夹具设计系统中几何拓扑信息无法利用,限制了实例检索效率,也限制了实例重用效率。此外,夹具设计中常用的专家系统等推理方式,知识获取困难,在很大程度上限制了智能设计系统的应用。为了实现复杂夹具的智能化设计,本文结合特征识别技术、3D模型检索技术和传统的几何分析推理技术,提出了广义3D几何推理技术的概念;研究了面向夹具设计的3D模型几何分析和推理方法的关键技术,包括3D几何信息提取方法、几何推理策略、几何模型的拓扑表示、融合工程语义的几何模型相似性比较、基于几何相似性的智能变型设计、设计知识的提取和表示等方面的内容;结合实际工程需求开发了一套CAFD系统。论文的主要创新点如下:面向夹具设计的特征识别。提出了一种基于图的夹具特征识别技术,给出了一种通过最大凹边凝聚方法识别极限表面的方法,并定义了制造特征的方向,在此基础上给出夹具特征识别的完整算法。同时针对夹具设计中需要提取出特殊的特征的需要,提出了一种双链遗传算法,并将这种算法应用于特征识别过程中。可以在一定程度上克服传统基于属性邻接图的特征识别的不足。推广了Brost-Goldberg方法,提出了完整的面向三维夹具设计的定位夹紧规划方法。Brost-Goldberg方法,在一定程度上只能解决二维零件和棱柱形零件,论文结合夹具特征识别技术和三维几何分析技术,将该方法推广到了三维夹具设计,并提出了三维夹具设计中的夹紧方案的确定方法。针对夹具详细结构设计的需要,提出了一种面向智能设计的夹具元件完整描述,并给出了特征装配环的概念用以检验装配尺寸完备性,在自动选件的基础上实现夹具详细结构自动化设计,使得自动化夹具设计方法得到了一定的推广。在零件属性邻接图(AAG)的基础上提出了一种零件几何拓扑结构的特征关注度模型,并给出了特征关注度模型的生成方法,在此基础实现了结合几何推理的夹具实例检索算法。同时针对传统基于实例推理(CBR)技术设计实例修改自动化程度低的问题,论文在特征相似性的基础上,提出了一种夹具实例自动化变型设计的方法,进一步提高CBR处理的智能化程度,为实现CBR技术完整的解决方法探索了一条新路。针对夹具设计知识获取困难的问题,本文提出了一种在几何结构化数据挖掘和贝叶斯网络基础上,进行夹具设计知识学习方法。给出了夹具结构的图表示,以及自动获取夹具结构的方法和夹具结构聚类的完整描述。给出了描述工序件拓扑结构的扩展属性邻接图,在此基础上给出了工序件对夹具结构的支持子图的概念,并给出了支持子图的学习方法,实现了对几何信息和夹具结构内在联系的学习。使用贝叶斯网络方法实现了对于夹具宏观设计属性的学习,并给出了支持夹具设计的推理过程。更多还原

【Abstract】 It’s very important for manufacturing system to improve the quality and efficiency of fixture design.Computer Aided Fixture Design (CAFD) is effective to solve this problem, but the CAFD system based on CAD is mainly focus on the modeling of fixture. There exist some problems in the approach of fixture design, although the intelligent method is used. The geometry information is ignored in most of these systems, which make it difficult to realize the intelligent fixture structure design. On the other hand, the efficiency case-retrieval and case-reuse is limited without geometry information in case based fixture design, because the fixture structure mainly depends on the topology of workpiece. In the end, the expert system used in intelligent fixture design cannot extract knowledge, which confines the use of the intelligent system to some extent.A concept of general geometry reasoning is proposed and used in intelligent fixture design based on feature recognition, 3D model retrieval and traditional geometry reasoning in this dissertation. Extracting the geometry information from 3D models, geometry reasoning, representation of 3D models, the similarity of CAD models with engineering semantics, the intelligent variant design based on geometry similarity and the extraction of fixture design are the main direction of the dissertation. A CAFD system is developed based on these theories. The main innovation as follows:Fixturing feature recognition is proposed based on the attributes ajoined graph.The recognition method of terminal faces is proposed based on classify of all concave edge connected face, and the direction of manufacturing feature is defined. The complete algorithm of fixturing feature recognition is presented based on this method, which is the basis of automotive design of fixture. The feature recognition based on double-link genetic algorithm is proposed for the special feature request in the fixture design such as T-slot, which can overcome the shortcomings of the traditional feature recognition.The completed fixture planning algorithm faced to the 3D fixture design is presented based on the Brost-Goldberg Algorithm.The classic Brost-Goldberg mainly focus on the planning of the fixture of 2D parts or prismy parts. This algorithm is combined with the feature recognition in this dissertation and extended to the 3D fixture design. The clamping planning is discussed too.The complete description of fixture parts is proposed. The concept of assembly feature loop is described and used in the detail design of fixture. The automotive design of fixture is extended based on these methods.An attention-driven model of workpiece is proposed to fit the case retrieval of fixture design for overcoming the shortcoming of the case retrieval based on properties.The algorithm to get the attention-driven model is described in detail. The algorithm of similarity matching based on the attention-driven models of parts is described in detail.A new retrieval approach of fixture case is discussed based on the attention-driven model, which can improve the efficiency of the case retrieval. A new method of intelligent case revise is presented based on the similarity of feature, which can solve the problem of low level of case revise in case based reasoning.A new machining learning approach is provided based on geometry structure data-mining and Bayesian network. The description of the fixture structure is presented based on the graph data structure and the algorithm to get it is proposed too. Then the clustering algorithm of fixture is introduced detailedly. The extended attributes adjoined graph is presented to describe the topology of part, and the support graph of fixture structure is defined too.The learning method of the support graph is desicribed in the following sections. The Bayesian network is used to learn the contaction between the technical attributes and the fixture structure.The reasoning process using the learned knowledge is depicted at last.更多还原