【作者】 侯鑫；

【导师】 刘文剑；

【作者基本信息】 哈尔滨工业大学 ， 航空宇航制造工程， 2010， 博士

【摘要】 随着全球经济一体化进程和信息技术的不断发展,市场竞争日趋激烈。企业间竞争的核心是产品设计的竞争,其焦点是如何快速引入新技术,以最短的时间和最低的成本开新发出产品。经过几十年发展的积累,企业内部积累了大量的设计资源和设计知识,如何有效重用这些设计资源和设计知识已成为实现产品快速设计、提高产品质量、降低生产成本的重要手段。本文在分析国内外设计重用理论及相关技术研究成果的基础上,研究了基于本体的设计重用技术,为产品智能设计和创新设计提供了有益的解决方案,并结合CAFD(Computer-Aided Fixture Design)的特点以及实际需要,研究了设计重用技术在CAFD中的应用。主要研究内容如下:将设计重用理论和方法与计算机辅助夹具设计相结合,建立了基于设计重用的CAFD系统框架,充分利用了已有的夹具设计经验和知识,实现了夹具的快速设计。该框架以夹具设计本体为基础,通过夹具设计文档库构建了夹具设计领域本体,以此作为框架内部不同模块间通信的基础;设计了设计重用引擎,涵盖了基于本体的夹具设计知识检索与基于形状和本体的夹具模型检索,为基于设计重用的夹具设计技术提供了新方法。研究了设计重用中的知识获取技术,将领域知识获取问题转化为领域本体获取问题,提出了基于图的领域本体自动构建算法,实现了从领域文档语料库中自动获取领域知识。本体的核心是领域概念以及概念间的相互关系,该算法首先将领域文本文档映射为文档概念图,采用基于图上随机游走的词汇加权算法从全局和局部两方面衡量词汇的重要性,然后利用图顶点聚类算法对词汇进行聚类以产生候选概念;提出了约束条件下基于频繁信息子图挖掘的概念间任意关系提取算法,并引入了信息函数对子图的信息量进行衡量,得到的领域概念和概念间的关系通过本体评价后,采用OWL-DL描述为领域本体。研究了设计重用中的知识表达与检索技术,提出了基于本体的产品设计知识检索算法,实现了语义级别的信息检索,查询内容不再局限于孤立的关键词,也可以为用自然语言表达的句子。该算法首先对查询内容进行语义分析,将其映射为本体语义特征,提出了基于本体的查询语义扩展算法,得到与查询语义相关的扩展语义特征,使检索系统可以直接明确的理解用户检索意图;将领域文档用领域本体进行标记,提出了基于本体的查询语义检索算法,基于文档本体语义图对领域文档进行聚类,通过比较查询本体语义图与文档聚类簇中值图的距离,得到与查询内容语义相关的文档簇,最终通过语义匹配将文档簇中的相关知识返回给用户。研究了设计重用中产品模型的表达与检索技术,提出了基于网格特征临界点的三维工程模型检索算法,实现了与CAD系统无关的基于形状的三维模型检索。根据微分几何中的Morse理论,采用网格顶点处的离散平均曲率作为光滑实值函数,计算网格特征临界点,避免了采样不足对算法造成的影响;采用两临界点间近似测地距离和顶点法矢夹角余弦值作为联合形状函数,更好地表征了模型整体表面和局部区域的的变化;按照极大值点、极小值点和鞍点,分别计算同类临界点间的联合形状函数得到形状分布,而模型的整体描述由三者加权平均得到,从而将模型的比较转化为形状分布矩阵的比较。另外针对夹具模型检索问题,研究了基于本体的工件到夹具模型的映射关系,提出了通过工件模型和设计需求检索得到夹具模型的方法。将本文研究成果应用于组合夹具自动设计系统,开发了基于设计重用的组合夹具自动设计系统(DR-CAMFD)。该系统在组合夹具设计过程中充分考虑了已有的组合夹具设计结果及组合夹具设计中的经验知识,采用进化搜索算法在设计空间中搜索合适的夹具设计方案,以寻求夹具设计问题的可行解。以一套组合夹具的设计为例说明了原型系统的应用,并验证了本文研究关键技术的正确性。更多还原

【Abstract】 With the development of global economic and information techneques, the considerable competitions among enterprises have been bloomed, and product innovation has become the main subject of manufacturing industry. The competitiveness of enterprises lies in its capability of product design, the focus of which is to quickly introduce new technologies and develop new products with highest quality and at lowest costs. Generally, knowledge is the basis of modern product design. Information is the carrier of knowledge, therefore, the design activities can be regarded as the flowing processes of the knowledge and information, i.e., relevant information can be materialized into a solution by information collection, processing and synthesizing. After the several decades’development, enterprises have accumulated a large number of design resources and design knowledge. Therefore, how to effectively reuse these design knowledge has been the important means of achieving rapid product design, improving product quality and reducing product costs, in which knowledge acquisition and retrieval are playing the core roles. Based on the state-of-the-art design reuse theory and relevant technical achievements, this dissertation conducted deep and detailed researches upon ontology-based key technologies of design reuse, providing a valuable solution for intelligent product design. In addition, the outcomes of this disseatation have also been applied in CAFD in terms of its characteristics and practical requirements. The main contributions of this dissertation are:The CAFD framework based on design reuse is proposed by combining the theory of design reuse with the characteristics of fixture design, so that existing fixture knowledge can be made good use of to achieve the rapid design of fixtures. This framework is based on fixture design ontology produced from domain corpus, which facilitates the commutations between different modules of this system. It encompasses ontology-based fixture design knowledge retrieval and ontology-based fixture model retrieval. The similarity between fixture model and fixture design requirements is assessed by the weighted average of the similarity of workpiece structural information and workpiece processing plan. Finally, a new fixture that meets the design requirements can be obtained by adapting the retrieved fixture model.On knowledge acquisition in design reuse, a graph-based approach for automatic construction of domain ontology from domain corpus, including concept extraction and relation extraction, has been proposed for knowledge and information management. First, each document in the collection is represented by a document graph. Then random walk term weighting is employed to estimate the relevance of the term information to the corpus from both local and global perspectives. Next, graph clustering algorithm is used to disambiguate terms with different meanings and group similar terms to produce concepts. Next, an improved frequent subgraph mining algorithm constrained by both vertices and informativeness is exploited to find arbitrary latent relations among these concepts. For ontology evaluation purposes, a method for adaptive adjustment of concepts and relations with respect to its practical effectiveness is conceived. Finally the domain ontology is formed by describing domain concepts and relations using OWL-DL, which is the basis of this dissertation.On the knowledge representation and retrieval, an ontology-based production information retrieval approach has been proposed. The semantics of the queries are first analyzed, and then converted into domain ontology concepts. The corresponding semantic features with regards to query semantics are expanded by ontology-based query expansion so that the retrieval system can understand the query intention of the users more clearly. The clustering between expanded queries and domain ontology concepts is performed during query process to find the proper ontology concepts closest to the queries, and then through the domain corpus labeled with domain ontology, the corresponding domain documents related to the query semantics are obtained. Finally, the relevant results are feedback to users by semantic matching.On the 3D shape descriptor and retrieval of product models for design reuse, a novel shape retrieval algorithm independent of CAD systems based on mesh salient critical points has been proposed. According to the Morse theory, the discrete mean curvature at vertices of the model mesh is used as the smooth real function to calculate salient critical points of the mesh, which reduces the impacts of insufficient sample points on algorithm performance. Then the approximate geodesic distance between two critical points and the corresponding cosine value of the intersection angle of the two vertex-normals are employed as the united shape function (USF) to describe the variation of global shape and local surfaces. Shape distributions are generated by computing the USF within each class of critical points, i.e. maximum, minimum, and saddle, respectively. The 3D shape is represented by the weighted average of these three sub-descriptors. Thus the matching of models is transformed into the comparison of corresponding shape distribution matrixes. This algorithm can assess the similarity of product models from the model itself without the requirements of other further information. In addition, a novel approach of retrieving fixture models by workpiece model is also addressed. In this method, a mapping from workpiece to its fixtures is established to achieve the retrieval of fixture models.The outcomes of this dissertation have been applied to computer-aided fixture design system, and the design-reuse based computer-aided modular fixture design (DR-CAMFD) system has been developed. In DR-CAMFD system, existing modular fixture design results, including fixture models and design knowledge and experiences, has been considered for a new fixture design. An evolutionary searching algorithm is used to find feasible solutions to the new design requirements in the design space. The fixture retrieved can be directly employed as the new solution if it meets the new design demands, or can be adapted to meet the new design requirements. We also took a modular fixture design as an example to verify the key techniques addressed in this dissertation.更多还原