基于图卷积神经网络的加工特征识别方法  

Machining feature recognition method based on graph convolutional neural network

作　　者：陈阳焜 王华昌[1] 李建军[1,2] CHEN Yangkun;WANG Huachang;LI Jianjun(State Key Laboratory of Material Processing and Die&Mould Technology,Huazhong University of Science and Technology,Wuhan 430074,China;Hubei Huangshi Mold Industry Technology Research Institute,Huangshi 435000,China)

机构地区：[1]华中科技大学材料成形及模具技术国家重点实验室,湖北武汉430074 [2]湖北黄石模具产业技术研究院,湖北黄石435000

出　　处：《模具工业》2024年第9期6-13,共8页Die & Mould Industry

摘　　要：针对传统特征识别方法难以处理多变特征和相交干涉特征的局限性,提出了一种基于图卷积神经网络(GCN)的特征识别方法,为最大限度地利用加工特征属性邻接图中的信息,设计了加工特征的初始节点嵌入向量矩阵以作为模型训练的基础。通过将采集的各加工特征数据集用于图卷积神经模型的训练,并通过试验进行了模型参数调优,将GCN模型应用于加工特征的分类识别任务中,达到了约99%的整体识别性能。与经典的图匹配方法对比分析结果表明:该方法整体性能更为优越,具有良好的通用性和鲁棒性。Traditional feature recognition methods often struggle with the variability of features and the interference arising from intersecting features,an approach utilizing graph convolutional neural networks(GCN)was proposed.To levarage the attribute adjacency graph associated with the machining features,an initial node embedding vector matrix as a foundation for model training was designed.With thorough training on diverse datasets of machining features and experimental optimization of the model's parameters,the GCN model demonstrated proficiency in classifying machining features,achieving an overall recognition accuracy of approximately 99%.Comparative analyses have shown the method's superiority over classic graph matching techniques,highlighting its wide applicability and robustness in feature recognition tasks.

关 键 词：特征识别 图卷积神经网络 属性邻接图 

分 类 号：TG76[金属学及工艺—刀具与模具] O241.2[理学—计算数学]