SSGCN-混合式图卷积网络:用于三维CAD模型的加工特征识别  

SSGCN-hybrid Graph Convolutional Networks for 3D CAD Model Machining Feature Recognition

作　　者：王洪申[1] 王尚旭 强会英[2] WANG Hongshen;WANG Shangxu;QIANG Huiying(School of Mechanical and Electrical Engineering,Lanzhou University of Technology,Lanzhou 730050,China;School of Mathematics and Physics,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州理工大学机电工程学院,兰州730050 [2]兰州交通大学数理学院,兰州730070

出　　处：《机械科学与技术》2025年第1期30-39,共10页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(61962035)。

摘　　要：为解决CAD/CAPP/CAM集成过程中,三维CAD模型加工特征识别问题,提出了一种混合式图卷积网络(Hybrid spectral domain and spatial domain graph convolution networks, SSGCN)的特征识别算法。以三维模型的面为节点,边为节点间的连接关系,构建图数据结构。提取面的几何属性信息,自定义编码构建节点属性矩阵,作为网络的输入。提取图结构的邻接矩阵、度矩阵等构建混合式图卷积网络。通过Python-OCC相关算法以及布尔运算,设计了一种批量生成带有面标签的加工特征模型数据集算法。使用带有面标签的加工特征模型数据集对网络进行训练,对加工特征模型进行测试,得到很好的识别效果。A hybrid spectral domain and spatial domain graph convolution network(SSGCN)algorithm is proposed to solve the problem of 3D CAD model machining feature recognition in the CAD/CAPP/CAM integration process.The graph data structure is constructed with the surface of the 3D model as the node and the edge as the connection between the nodes.The geometric attribute information of the surface is extracted,and the node attribute matrix is constructed by custom coding as the input of the network.The hybrid graph convolutional network is constructed by extracting the adjacency matrix and degree matrix of graph structure.Through the Python-OCC related algorithms and Boolean operation,an algorithm for batch generation of machining feature model dataset with face labels is designed.The machining feature model dataset with face labels is used to train the network and test the machining feature model,and a good recognition effect has been obtained.

关 键 词：CAD模型 图卷积网络 加工特征识别 邻接矩阵 

分 类 号：TH164[机械工程—机械制造及自动化] TH166