【作者】 赵昌葆；

【导师】 魏生民；

【作者基本信息】 西北工业大学 ， 航空宇航制造工程， 2006， 博士

【摘要】 知识的价值早已被人们认识，尤其在解决工程领域问题中起着不可替代的作用。知识工程除了知识获取瓶颈外，缺乏对复杂问题求解用到的大范围异类知识的有效组织与表示方法，致使其处理的问题规模有限。本文结合应用广泛、求解困难的工程问题对工程知识进行了研究。解决一个复杂问题需用到若干不同的知识，它们通过问题求解中知识的运用发生联系，解决一类问题的知识自然地结合在一起，构成了知识系统。本文基于知识的系统本质，将知识系统的静态结构和动态行为作为知识系统的两个方面，建立了知识的表示与组织、获取与进化以及运用与管理的相关概念、性质与方法，形成了完整的知识系统描述体系。复杂问题的求解很难一步完成，往往需要将求解过程划分为若干阶段，各阶段依次构成递进关系，求解过程规划以及求解知识组织，都以构成递进关系的环节为基础。本文分析了问题求解与知识运用之间的关系，将问题求解中具有递进关系的单元作为知识系统的层次。问题求解中起作用的是知识中蕴涵的因果联系，它是知识的核心。知识中蕴涵的因果联系涉及诸多因素，对这些因素的认识程度的不同，体现为因素的不同表示形式，一则知识将具有不同形式的诸多因素逻辑上联结在一起，像是因素聚合的知识颗粒。因果联系涵盖的范围可以由知识颗粒的大小来比拟，即知识粒度。本文剖析了知识的粒度特性，给出了知识粒度的定义、性质与表示方法，将知识统一表示为知识前件、知识后件及二者之间的映射，知识前件与知识后件由不同的成分——要素、属性和约束来描述；提出了层次规划、粒度选择与解的生成三个求解策略，引入了基于匹配度、支持度及语义距离的知识检索方法，将知识推理与约束满足的方法结合起来完成解的生成与评价，并给出了约束调整规则和求解回溯方法，从而建立了基于粒度知识的综合求解方法；提出了基于任务分析的粒度知识获取方法，以及“知识数据增长—知识成分变化—知识粒度变化—知识层次调整”四个层级的粒度知识进化机制；揭示了粒度知识包含信息的分形特点，提出了粒度知识的分形编码方法，以简码和全码关联的方法准确而简洁地表示知识。作为示例与验证，本文以工艺设计为例，阐述了粒度技术在工艺设计中的应用，介绍了西飞工艺知识库系统，反映了本文的核心思想，验证了其效果。更多还原

【Abstract】 Knowledge is of remarkable value as is aware in a long history, especially in solution of engineering problems. Inclusive the bottleneck of knowledge acquirement, the scale of solving problems with knowledge engineering methods is limited for the lack of effective methods of organization and expression of large range different knowledge that’s used in solution of complicated problems. This thesis studies on the engineering knowledge in such engineering problems that are widely applied but difficult to solve.The solution of a complicated problem concerns knowledge of different kinds. The relations among different kinds of knowledge constructed through knowledge application in problems solving. Knowledge used in solving similar problems contacts spontaneously and forms a knowledge system. In the thesis, the static structure and dynamic action are considered as two aspects of a knowledge system based on the systematic nature of knowledge. Concepts, properties and methods are proposed for knowledge expressing, organizing, acquiring, improving, applying and managing. Consequently, a complete system describing knowledge is constructed.The solution of a complicated problem is usually divided into several steps other than one. The steps in the solving series are elements of problem solving process programming and problem solving knowledge organizing. In this thesis, relation between problem solving and knowledge application is analyzed, and the elements with increasing relations in problem solving are considered as levels of the knowledge system. In application, the factors taking effect on problem solution is the cause-result relation connotated in knowledge, whatever means the knowledge expressed in. The cause-result relation connotated in knowledge concerns various factors which are cognized in different level. A piece of knowledge connects various factors in different forms in logic, which seems to be a knowledge grain aggregated by various factors. Different knowledge reflects different cause-result relation. The covering range of cause-result relation can be simulated by the dimension of. knowledge grain, i.e. knowledge granularity.Knowledge granularity is analyzed in the thesis, including definition, properties and expression methods.A knowledge granularity is expressed with antecedent, consequent and mapping relations between them. The former component and the later component define by different elements, attributes and constraints.In this thesis, three solving strategies, i. e. level programming, granularity selecting and solution generating are proposed. With introduction of knowledge indexing methods based on matching scale, supporting degree and semantic distance, knowledge reasoning and constraint satisfying are combined to accomplish solution generating and judging. Rules of constraints adjusting and methods of solving back tracing are proposed. Consequently, a comprehensive solving method is presented.A granularity knowledge acquiring method based on task analyzing and an granularity knowledge evolving mechanism of four layers "knowledge data increasing-knowledge elements varying-knowledge granularity varying-knowledge levels adjusting" are proposed.In this thesis, the fractal properties of information connotated in knowledge granularity are disclosed. A fractal coding method with simplified code and complete code relating is proposed to express knowledge.As examples and validations, the expression, organization and application of process planning knowledge based on granularity technology are debated in detail. A process planning system developed based on granularity knowledge technology, i.e. XAC (Xi’an Aircraft Corporation) Process Planning Knowledge Database System is introduced.更多还原