基于图神经网络的机匣特征自动识别方法  

Method on Automatic Feature Recognition of Casing Based on Graph Neural Network

作　　者：邓岩 王波兴[1] Deng Yan;Wang Boxing(School of Mechanical Science and Engineering,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]华中科技大学机械科学与工程学院,武汉市430074

出　　处：《工具技术》2023年第12期150-156,共7页Tool Engineering

基　　金：国家重点研发计划(2020YFB1708901)。

摘　　要：特征自动识别技术是数控加工编程中的重要组成部分,针对航空发动机机匣的特征自动识别问题,提出一种基于图神经网络的特征识别方法,在机匣模型的属性邻接图基础上,通过边界关系对属性邻接图进行分解得到特征子图,利用图神经网络的拓扑学习机制将特征子图的识别问题转换为图分类问题,避免传统BP神经网络编码特征子图的复杂性。实验表明,本文方法能够准确识别机匣特征,具有可行性。Automatic feature recognition technology is an important part of automatic programming of CNC machining.In order to solve the automatic feature recognition problem of aero-engine casing,a feature recognition method based on graph neural network is proposed.Based on the attribute adjacency graph of the casing model,the attribute adjacency graph is decomposed in accordance with the boundary relationship,thus to obtain the feature subgraph.The topology learning mechanism of graph neural network is used to transform the recognition problem of feature subgraph into a graph classification problem,which avoids the complexity of the traditional BP neural network coding feature subgraph.Experiments show that the proposed method can accurately identify the characteristics of the casing and is feasible.

关 键 词：机匣 数控加工 特征自动识别 属性邻接图 图神经网络 

分 类 号：TG659[金属学及工艺—金属切削加工及机床] TH164[机械工程—机械制造及自动化]