Abstract: The assembly process documents contain semantic information with multiple levels of assembly elements and complex relationships.It exists in natural texts without standard or specific constraints,and faces great challenges in evaluating assembly process enforceability and guiding assembly operations.In view of this,a semantic recognition method based on Long-Short-Term-Memory (LSTM) network was proposed.The semantic elements of assembly process documents were extracted and the knowledge description of assembly process semantic information was established.The assembly semantics and the description method of its knowledge graph form were defined.Through assembly morpheme decomposition,assembly morpheme classification and relationship construction,the assembly process knowledge graph was obtained.Taking the assembly process documents of a certain type of aero-engine rotor component as an example,the semantic recognition rate of assembly entities using the proposed method could reach 93.7%,the semantic recognition rate of relationships could reach 94.8%,and the assembly process knowledge graph was obtained.Therefore,the effectiveness of the proposed method was proved.

Key words: long-short-term-memory, assembly process, semantic recognition, knowledge graph

摘要： 装配工艺文档蕴含要素层次多样、关系复杂的工艺语义信息,它们以自然文本存在,无标准或规范约束,在评估装配工艺可执行性和指导装配作业时面临巨大挑战。鉴于此,提出一种基于长短期记忆网络(LSTM)的装配工艺语义识别方法,抽取装配工艺文档语义要素,建立装配工艺语义信息的知识化描述。首先定义装配语义并规定其知识图谱的描述形式。然后通过装配语素分解、装配词素分类与关系构建,得到装配工艺知识图谱。以某型航空发动机压气机转子部件的装配工艺文档为例,自动识别其装配语义并得到知识化描述。其中,对装配实体的语义识别率为93.7%,装配关系的语义识别率为94.8%,得到装配工艺知识图谱,证明了方法的有效性。

关键词: 长短期记忆网络, 装配工艺, 语义识别, 知识图谱