摘要
针对兵器行业关键零部件加工工艺自动生成所需样本数量不足的问题,提出了一种基于小样本知识学习的工艺生成方法,重点研究基于样本变换和样本集扩展的几何模型增强解析技术,以及基于数字孪生的工艺知识增量学习技术,以实现工艺样本的整体增强。同时,建立面向增强样本分析的卷积神经网络模型和适于异构知识融合的知识图谱,形成工艺知识的高效分析与有效利用方法。采用深度神经网络模型解决复杂零件三维设计模型的工艺结构解析问题;基于知识图谱的工艺知识表征方法和工艺生成技术以克服工艺信息碎片化的难题;通过数字孪生模型与工艺验证技术解决理论工艺方案与现场生产要素脱节的问题。实现了基于零部件结构解析、基于知识图谱的工艺方案生成和基于数字孪生的工艺验证和优化的工艺知识闭环分析。
Aimed at the problem that the number of samples needed for automatic generation of key parts processing technology in ordnance industry was insufficient,a process generation method based on small sample knowledge learning was proposed.Mainly focused on the geometric model enhancement analysis technology based on sample transformation and sample set expansion,and process knowledge incremental learning technology based on digital twin to realize the overall enhancement of process samples.At the same time,a convolution neural network model for enhanced sample analysis and a knowledge map suitable for heterogeneous knowledge fusion was established to form an efficient analysis and utilization method of process knowledge.The deep neural network model was used to solve the problem of process structure analysis of 3D design model of complex parts.Process knowledge representation method and process generation technology based on knowledge map were used to overcome the problem of process information fragmentation.Digital twin model and process verification technology were used to solve the problem of disconnection between theoretical process plans with field production factors.It was realized that the process plan generation based on component structure analysis,process plan generation based on knowledge map and process knowledge closed-loop analysis based on digital twin.
作者
应小昆
金阳
孙林
关杰
张晓华
刘芳
YING Xiaokun;JIN Yang;SUN Lin;GUAN Jie;ZHANG Xiaohua;LIU Fang(China North Advanced Technology Generalization Institute,Beijing 100089,China)
出处
《新技术新工艺》
2020年第12期6-11,共6页
New Technology & New Process
关键词
小样本
工艺知识图谱
工艺生成
数字孪生
few sample
process knowledge graph
process generation
digital twins
