Abstract: With the widespread application of digital manufacturing systems,the amount of process data generated by manufacturing companies has been continuously increasing.To achieve effective reuse,learning and mining of existing process data,a knowledge graph and deep learning-based approach for part machining process design was proposed.A process knowledge graph model based on features,parts,feature process plan and part processes was constructed to achieve a structured multi-level representation of process data.On this basis,a BiLSTM+Attention deep learning model was developed to reveal the mapping patterns between parts and typical process plans,and a Seq2Seq+Attention deep learning model was developed to generate effective sequences of part process steps.A part process reasoning method based on the fusion probability of feature process plans and macro process sequences of parts was proposed,achieving effective generation of part process plans with complete process context.Finally,a prototype system was developed and validated using pin parts as an example to demonstrate the effectiveness of the proposed approach.

Key words: process data, process design, knowledge graph, deep learning, fusion

摘要： 随着数字化制造系统的广泛应用,制造类企业产生的工艺数据数量持续增多。为了实现对已有工艺数据的有效复用、学习和挖掘,提出一种基于知识图谱与深度学习的零件机加工艺设计方法。首先构建以特征、零件、特征工步方案、零件工艺为基础的工艺知识图谱模型,实现工艺数据的结构化多层次表示。在此基础上,构建一种BiLSTM+Attention深度学习模型揭示零件与典型工艺方案之间的映射模式,以及一种Seq2Seq+Attention的深度学习模型实现零件工序序列的有效生成。其次,提出一种基于特征工步方案与零件工序方案融合概率的零件工艺方案决策方法,实现具有完整工艺情境的零件工艺方案有效生成。最后,以销轴类零件为例,开发原型系统验证了所提方法的有效性。

关键词: 工艺数据, 工艺设计, 知识图谱, 深度学习, 融合