Abstract: To enhance the automation of design calculation,a knowledge modeling method for design calculation based on the multi-layer knowledge graph was proposed that fused knowledge of different types,such as computational object structure,computational processes and computational formulas.A Multi-layer knowledge graph of design calculation was proposed,including a product structure layer,a computation event layer and a calculation formula layer,which respectively represented the knowledge of computation objects,computation processes and calculation formulas.To automatically extract event and formula entities according to the characteristics of design calculation knowledge from design manuals or standard documents,a deep convolutional neural network was constructed,which included feature extraction,candidate region generation and target classification.The correlation between example entities was established through the mutual location of example entities,while formula entities were transformed from text entities to calculation entities via a LaTeX intermediate representation.Finally,the feasibility of the proposed method was verified by automating a typical design calculation task for a crane product as an example.

Key words: knowledge graph, design calculation, knowledge model, semantic representation

摘要： 为了提升设计计算的自动化程度,提出了一种基于多层知识图谱的设计计算知识建模方法,用于融合计算对象结构、计算流程和计算公式等不同类型的知识。提出了包括产品结构层,计算事例层,计算公式层的设计计算多层知识图谱,分别用于计算对象、计算流程和计算公式的知识表示。根据设计计算知识来源于设计手册或者标准文档的特点,构建了包括特征提取、候选区域生成以及目标分类的深度卷积神经网络,用于事例和公式实体的自动抽取。通过事例实体间的相互位置关系建立了事例实体之间的关联关系,公式实体则经由LaTeX中间表示实现了从文本实体到计算实体的转换。最后,以起重机产品的设计计算为例,通过自动完成该产品的典型设计计算任务,验证了所提方法的可行性。

关键词: 知识图谱, 设计计算, 知识建模, 语义表示