【作者】 郑军红；

【导师】 叶修梓； 张三元；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2010， 博士

【摘要】 工件的装夹是加工过程的一个重要方面,对生产质量、时间和成本都有极大的影响。随着计算机技术的发展,特别是CAD/CAM技术的发展,人们开始用计算机来继承设计知识和经验,通过系统化的方法辅助夹具的设计与分析,并重点研究自动化的方法以支持夹具方案的概念设计和详细设计。在这一背景下,论文围绕智能夹具设计系统的构建问题,以知识工程方法学为指导,展开计算机辅助夹具设计(CAFD)系统和相关技术的研究,主要工作包括：为了提高夹具设计系统的实用性和改进设计过程,基于知识工程进行智能夹具设计系统的构建。论文引入CommonKADS方法来指导CAFD任务的分析和知识获取,通过它提供的CML语言构建了夹具设计知识模型,刻画了知识的上下文和相互依赖关系。在该模型中,领域知识将夹具设计涉及的知识项概念化,任务知识描述夹具设计任务的分解和控制策略,推理知识描述基本推理步骤和推理结构。这一分析与构建过程把不同类型的知识区分层次,引导我们以类似软件工程的方法来探索知识密集型系统的开发。所获取的知识包含静态知识以及夹具设计问题求解的组织与推理控制知识,为后续的研究和知识系统构建工作提供了一个坚实的基础。针对CAFD领域存在的设计知识重用困难问题,论文提出了基于本体的夹具设计知识表示与推理方法。遵循CommonKADS知识模型的结构化构建方法,采用OWL来构建分层的夹具设计知识本体,借助Protege系统定义了该领域的基本概念,如装夹特征、元件、关系与约束等。该本体也定义了“规则类型”概念,用于收集具体的设计规则,并用SWRL来描述,建立一个语义清晰的夹具设计知识库,分离了事实和推理。该本体采用了层次结构,便于扩充和修改,具有较强的推理能力。由于夹具规划中工件形状和加工工艺的复杂性,夹具规划的优化较少研究,往往延迟到校验阶段才进行,存在不少弊端。论文提出基于Kohonen自组织神经网络的装夹面分组与选择算法和基于遗传算法的定位点优化算法,利用人工神经网络来处理装夹面选择中各种复杂的影响因素,选择最佳的装夹表面；在此基础上,参考夹具校验的一些结论,通过一些参数来模拟工件的稳定性与变形特性,利用遗传算法进行组合优化,确定最优定位点。夹具结构设计是CAFD系统的核心,但现有的CBR方法和基于生成的方法设计效果还不理想。为此论文引入“典型结构”概念以扩展夹具结构设计的概念模型,它融合了CBR方法和基于生成方法的概念模型的优点,不仅包含设计意图,也包含结构信息。在此基础上,完善基于图论表示的分层夹具结构模型,以利于设计问题的分解和设计元素的重用。借助该分层结构模型,本文进一步提出了基于图规划算法的夹具结构设计方法。在设计过程中,先确定夹具方案所用的典型结构及其构建顺序,形成了夹具结构的骨架,然后在功能组件层从目标状态出发进行图扩展,达到初始状态后搜索规划图生成功能组件。由于本方法是在局部结构上进行目标导向图规划,并以启发性信息指导搜索过程,缩小了规划图规模,改善了设计效果,提高了设计效率。在上述研究的基础上,开发了智能夹具设计原型系统。介绍了该系统的体系结构和开发平台,以及系统支持的半自动设计方法和交互式设计方法。原型系统的应用验证了本文基于知识工程方法构建智能化夹具设计系统的有效性。更多还原

【Abstract】 Fixturing is an important manufacturing operation which contributes greatly to the production quality, cycle time, and cost. As computer technology, especially computer-aided design and manufacturing (CAD/CAM), developed, computer is utilized to inherit the knowledge and experiences and some systemic method is implemented to assist fixture design and analysis. The current research is focused on the approaches to automate the task of conceptual and detailed fixture design. Under this background, the thesis makes a point of building an intelligent fixture design system. The related research on fixture design ontology, fixture planning and fixture configuration design in a computer-aided fixture design (CAFD) system are also surveyed as follows:In order to enhance the practicability of fixture design system and improve the design process, knowledge engineering is utilized to develop the intelligent system for fixture design. CommonKADS methodology is imported to analyze the task of the fixture design and to retrieve the knowledge. Using the conceptual modeling language (CML), the knowledge model of fixture design is constructed, including domain knowledge, task knowledge and inference knowledge. The domain knowledge describes schematically the knowledge and information types concerning fixture configuration design. The task knowledge describes the functional decomposition and control strategy. The inference knowledge specifies the basic reasoning steps and inference structure. Such analysis and construction process distinguishes the knowledge types in a hierarchy, which guides us to probe into the development of the knowledge intensive system by a process comparable to software engineering. The knowledge retrieved contains not only the static knowledge, but also the organization knowledge and reasoning control knowledge concerning fixture design problem solving, which provides us with a foundation for algorithms research and knowledge system building.Reuse of design knowledge has become an urgent problem for computer-aided fixture design. Ontology is widely used to represent the conceptual model of information systems at the semantic knowledge level and makes the model reusable. Therefore an ontology based representation and inference approach on fixture design is proposed. Following the structural modeling method of CommonKADS, the fixture design knowledge ontology is established and represented by ontology modeling language OWL. The ontology is corresponding to the domain knowledge model of CommonKADS and comprises a detailed specification of the types and relations of the data involved in fixture design process. Assisted by the Protege system, the basic concepts such as fixturing feature, fixture component, relation and constraint, are defined, amongst which the layered conceptual structure are constructed. Based on the ontology, the concept of Rule-type is defined to accommodate specific design rules. The rules of CAFD are expressed in Semantic Web Rule Language (SWRL), which improves the interoperability between the rules and the ontology. The facts and rules are captured into the knowledge base with explicit semantic. This ontology has an advantage of good expansibility and can be easily transferred for reasoning.Due to the complexity of workpiece and machining process, the optimization of fixture planning is seldom researched and is often forced in fixture verification phase. The thesis presents the algorithm of fixturing surfaces grouping and selection based on Kohonen Self-Organizing Neural Network and the algorithm of optimizing locating points based on Genetic Algorithm. Artificial Neural Network is utilized to deal with the complicated influencing factors in fixturing surfaces selection to select the best fixturing surfaces. Based on the results of fixture verification, the stiffness and deformation of the workpiece is simulated. Genetic Algorithm is then adopted to optimize the combination of locating points.Fixture configuration design is the kernel of CAFD system. But the effect of existing case-based reasoning (CBR) method and generating-based method has been far from ideal. The thesis introduces the concept of typical structure to extend the conceptual model of fixture configuration design, which contains design intention and configuration information. The layered fixture configuration model is improved therefrom, so as to decompose the design problem and reuse the design elements. Furthermore, with the layered model, an approach based on graphplan algorithm for fixture configuration design is presented, which transforms the design problem into a planning problem. During design process, the typical structures and their construction order are firstly determined, forming the skeleton of fixture configuration. Then in functional unit layer underneath, planning-graph is expanded from goal states, and is searched forward to generate functional units. Because the goal-directed graphplan is applied on localized fixture configuration and is heuristically guided, the method reduces the scale of the planning graph, improves the efficiency.Based on the study above, a prototype intelligent fixture design system has been developed, which supports sub-automated mode and interactive mode of fixture design. The architecture and the developing platform are introduced. The application of the prototype system verifies the feasibility and effectiveness of the building method based on knowledge engineering.更多还原