基于认知语言学原理的工艺知识发现方法研究

Knowledge acquisition from the manufacturing process data is key and urgent for manufacturing enterprises to improve the intelligent level and innovation capability of process planning. As empirical process data is with characteristics of uncertainty and diversity, and to overcome limitations in the existing research whose knowledge representation were hard to be reused and shared, a new way exploring cognitive linguistics theory for process knowledge discovery is proposed. The research mainly consists of the following two aspects. Firstly, the cognitive mechanism of knowledge discovering of process planning will exposure by learning from the use of cognitive linguistics. The actions of extracting the process knowledge from the process data will be regarded as the cognitive process from the realistic interactive experiences to the image schema to the concept structure. Secondly, the computing model will be established to simulate the cognitive process of knowledge discovering, which will comprehensively apply various knowledge discovery methods including latent semantic analysis, spectral clustering, manifold learning, ontology and other technologies. The model will conform to relevant cognitive regulations and characteristics of process data. As a result, the concepts and their relationships will be mined from huge process documents and will be represented as the ontology, which will provide the foundation for a continuing development framework of process knowledge. The proposed method will promisingly lead to change and enrich the process of knowledge discovery in both methodological theory and the mechanism for process planning. It also has important theoretical value to the knowledge discovery from the data with empirical and uncertainty characteristics.

工艺知识的有效获取是提高企业整体工艺设计水平和创新能力迫切需要解决的问题。项目针对工艺知识发现过程中存在的认知方向不确定、挖掘结果不易重用和数据类型复杂等问题展开研究，提出一种基于认知语言学原理的工艺知识发现新方法。主要研究内容包括:1）借鉴和运用认知语言学原理，揭示工艺知识发现内在认知机理，建立从现实经验到意象图式到概念体系的工艺知识发现认知过程；2）从信息处理的角度出发，综合运用潜在语义分析、谱聚类、流形学习等知识发现技术，建立一个可计算的知识发现认知过程模型，使之符合人类认知规律和工艺数据特点，实现从海量工艺规程文档中进行概念及概念结构的提取，形成初步的工艺知识本体，从而为实现工艺知识的可持续积累和演化奠定基础。研究成果预计能够从理论基础和机理上丰富和突破工艺知识的发现方法，对于拓展具有经验性、不确定性特征的知识发现方法具有重要的理论价值。

工艺知识的有效获取是提高企业整体工艺设计水平和创新能力迫切需要解决的问题。针对工艺知识发现过程中存在的认知方向不确定、挖掘结果不易重用和数据类型复杂等问题，本项目借鉴认知语言学对人类概念系统和句法结构形成的认知规律，提出了一种基于认知语言学原理的工艺知识发现方法，并通过融合多种知识发现技术建立了可计算模型来模拟该过程的运行。主要研究进展及创新点：1)从认知心理学角度研究了工艺知识学习的内在认知机理，提出了将工艺知识概念化过程用表达概念的词汇的认知规律来近似，建立了一个以“互动经验信息采集->常规意向图式建立->意向图式提取->概念间结构形成”四个具有密切连贯性和协作性认知活动为核心的工艺知识发现模型；2）以工艺要素作为工艺设计的元认知，完成了工艺设计信息建模，实现了工艺设计互动经验信息的采集；3）基于潜在语义分析方法模拟创建意向图式的认知心理特征，实现了工艺设计决策结果与工艺要素之间隐含语义结构的提取，最大程度保留了工艺知识的不确定性。将先验知识与数据相结合，提出了基于贝叶斯网络学习的工艺设计过程模糊因果关系知识挖掘方法；4）提出了流形学习和关联规则相结合的意象图式提取算法，获取高维常规意象空间中低维嵌入流形，即同类工艺过程的本质工艺要素及关系，形成概念的原型，成为构建具有普遍性的工艺知识的出发点；5）通过融合认知心理学关于概念结构的特征表说和原型说,建立了多层次变维结构概念模型，由工艺要素维、工艺实例维和意向图式维组成变维向量空间，实现了具有经验语境和动态性的工艺概念描述。研究团队按照研究计划完成了既定的研究内容，共发表学术论文14篇，其中SCI索引2篇、EI索引9篇；申获发明专利1项，软件著作权1项；培养博士生1名，硕士生4名。