【作者】 方辉；

【导师】 殷国富；

【作者基本信息】 四川大学 ， 机械设计及理论， 2007， 博士

【摘要】 随着社会经济的发展和科学技术的进步，制造系统日益向智能化和柔性化方向发展，而智能化和柔性化是建立在对相关信息和知识快速有效的获取、处理和传递基础之上的。制造知识内涵丰富，具有多样性、复杂性、经验性和非规范性等特点，在相当程度上增加了对设计制造知识进行有效的归纳、整理、挖掘和应用的难度，同时也降低了其可重用性和可集成性，为企业信息交互带来了困难。如何减小和消除信息的非规范性的不利影响，最终实现面向智能制造的设计制造信息快速有效的获取、处理、传播和应用，是制造系统信息化和智能化研究的一项重要课题。从结构上来看，非规范制造信息是指半结构化和非结构化的制造信息；从形态上来看，非规范制造信息是指制造系统中的不确定、不精确、不完备、不协调和不稳定的制造信息。非规范信息是企业信息资源的重要组成部分，也是企业的宝贵知识财富，但非规范信息也使企业信息环境变得更为复杂，信息的快速无障碍交互更加难以实现，从而增加了信息的使用成本，降低了信息的应用价值。非规范信息处理的目标就是尽可能减小上述不利影响，并建立一种通用信息资源共享及应用的支撑环境，使得各类信息能够在企业范围内合理流动，将恰当的信息和知识在恰当的时间通过恰当的方式传递给相应的人员。全文的核心内容主要包括以下几个方面。(1)分析了智能制造对制造信息与知识资源的需求及现有的制造信息与知识资源系统的不足，在此基础上提出了面向智能制造的信息资源及知识库系统研究的总体目标，并给出了系统框架结构和功能模型；构建了网络环境下多模式知识形态的分类方法，并提出了制造知识的基元表示形式，以解决制造知识的统一描述问题。(2)分析了制造系统中的不确定性，归纳和阐述了不确定性制造信息产生的来源；对非规范性制造信息进行了定义，分析了其来源及表现形式，并在此基础上阐述了信息的非规范性与不确定性的联系与区别；分析和总结了非规范性制造信息的特点；对信息粒度概念进行了说明，阐述了对制造信息进行信息粒度分析的必要性，对制造信息颗粒的划分进行了讨论，并阐述了制造系统非规范性信息粒度计算方法；提出了非规范性信息的规范度概念，并以粗糙集为理论基础，提出了非规范性制造信息相关性描述方法。(3)以设计和工艺知识基元为例，建立了相应的制造知识基元体系，并提出了相关知识基元的基本格式；在上述工作基础上，提出了制造知识基元的元数据表示方法，并对其在企业信息环境和网络环境中的转换、处理技术进行了较为详细的说明。(4)讨论了智能检索与知识发现的联系与区别，提出了网络环境下非规范性制造知识的检索模式；构建了复杂信息环境下的非规范性制造知识智能检索系统，并分析了其工作模式、工作流程和关键技术，提出了包含语义匹配和语义推理机制的智能检索方法，通过相似度和相关度计算实现检索要求与检索对象的智能匹配，以提高制造知识检索的查全率和查准率。(5)建立了网络环境下个性化制造知识主动推荐服务体系结构，通过Web内容挖掘、结构挖掘和使用挖掘等Web数据挖掘方法，结合主动和被动学习方式，获取用户的制造知识兴趣或偏好。提出了相应的用户兴趣获取、Web页面相似性判断及相似Web页面收集算法。(6)针对工艺决策的非规范制造信息环境，提出运用不确定语言多属性决策方法进行多工艺方案评价：提出了工艺方案评价的多属性体系；应用不确定拓展有序加权平均算子进行工艺辅助决策，以充分利用工艺人员的经验型知识。(7)通过对简单遗传算法的改进实现算法的加速，同时将人工排样作为遗传算法初始种群的组成部分，使人的经验知识融合到算法中。上述改进不仅对算法有加速作用，以实现优化结果和优化效率的平衡，而且使得相关工作人员的经验知识融入实际优化过程，避免了完全脱离人工干预、单纯依靠算法获取优化方案带来的相关实用性问题。(8)以建立面向机械制造的信息资源共享及应用支撑环境为目标，采用统一的基于XML的数据交换和插件技术，开发由制造信息资源与知识库、制造知识发现、发布与应用等模块组成的通用制造信息资源和知识库软件原型系统。其中的制造信息资源与知识库系统包括尺寸公差与形位公差智能查询与选用、金属切削计算及查询等功能组件；制造知识应用包括基于不确定语言多属性决策方法的工艺方案评价和基于改进遗传算法的板材排样算法等组件。更多还原

【Abstract】 With the development of socioeconomic and advancement of science and technology, manufacturing system becomes more and more intelligentized and flexible, and these characters are based on speediness and availability of information and knowledge acquirement and disposal. Manufacturing knowledge has characters of diversity, complexity, empirical and unnormalized, and has abundance connotation. These characters make it difficult to induce, coordinate, mining and apply, as well as reduce the possibility of re-use and integration, and make it harder for enterprises’ information interactive. How to diminish and eliminate the disadvantages of unnormalized information, and achieve the goal of acquire, dispose, spread and apply information rapidly and efficiently, is a key problem should be researched.From the aspect of structure, unnormalized information refers to half-structuring and non-structuring information, and from the aspect of conformation, unnormalized information refers to uncertain, un-precise, imperfection, un-correspond and unstable information. On the one hand, unormalized information is an important part of enterprise information resource and valuable fortune, on the other hand, it makes information environment much more complicated, increases information’s use-cost, and decreases information’s use-value. The goal of disposal of unnormalized information is to diminish the disadvantages mentioned above, and achieve the target of pass the right information to the right staff at the right time. The main contents of this research are as follows.(1) Demands toward manufacturing information and knowledge resource and deficiencies of existing information resource system are analyzed, and put forward the general goal of the research of information resource and knowledge base oriented intelligent manufacturing, and bring forward the system’s skeleton structure and function model. The classify method of multi-model knowledge form within web environment is constructed, and put forward manufacturing knowledge’s elementary express model to resolve the problem of unified description.(2) The uncertainty within manufacturing system is analyzed and the source of uncertainty is inducted and expatiated. Defined and analyzed unnormalized information and its source and represent, and based on the work mentioned above, explained the relation and distinguish between unnormalize and uncertainty. Sum up characters of unnormalize manufacturing information. Based on the explain of information granularity, set forth the necessity of information granularity computing, and bring forward the concept of unnormalized information’s degree of criterion, and based on theory of rough sets, put forward describe method of relativity for unnormalized manufacturing information.(3) Aimed at peculiarities of unnormalized manufacturing information, constructed describe mechanism of manufacturing knowledge based on knowledge elementary cells. Take design and process elementary cells knowledge, construct corresponding manufacturing elementary cells knowledge system and bring forward knowledge cells’ basic format. Based on these work, put forward meta-data expression method for knowledge cells and explain how to transfer and disposal it within enterprises’ information environment.(4) The connection and difference of intelligent retrieve and knowledge discovery are discussed, and put forward the retrieve model of unnormalized manufacturing knowledge under web environment as well as its function model, workflow and key technology. Bring forward a intelligent retrieve method include semantics match and semantics reasoning, through the degree of similarity and correlation to achieve intelligent match of retrieve demands and objects, and to get the aim of improve retrieve method’s recall ratio and precision ratio.(5) An architecture of individuation manufacturing knowledge initiative recommend service is put forward, and through Web content mining, usage mining and structure mining, combine with initiative and passivity learning method to capture users’ interesting and preference toward manufacturing knowledge. Put forward corresponding algorithm such as users’ information acquire, Web pages comparability judge and similar Web pages’ collection.(6) Applying uncertain language multiple attribute decision making method for process evaluation, and discussed methods to get multi process plan and the multi attribute system of process evaluation. Based on these work, applying uncertain expansion order weighting average operator for process evaluation, and using uncertain language evaluation method for process aided decision-making to make the best of technologists’ work experience. (7) Take manual nesting plan as part of genetic algorithm’s initial population so as to mix human experience with the algorithm, and take the transformation interval of finity excellence individual as new initial evolution interval and run standard genetic algorithm again, so the evolvement and accelerate inheritance go along at the same time and get the best result step by step. The improvements mentioned above not only has the effect of accelerate the algorithm, but also make use of workers’ experience and avoid the problem that the algorithm be divorced from any human intervention, at the same time, the improved genetic algorithm has better computing performance and lend itself to get the balance of better optimize results and higher optimize efficiency.(8) Adopted uniformed local data exchange and plug-in technology to develop knowledge base software prototype system, and the system mainly including manufacturing information and knowledge base, knowledge mining, issue and application models. Among them, the knowledge base including function modules such as dimensional tolerance and form and position tolerance intelligence query and choose module, metal-cutting computing and query module, and the knowledge application model including function modules such as process plan evaluation module based on uncertain language multi-attributes decision making method and cutting stock module based on improved genetic algorithm.更多还原