《计算机应用》唯一官方网站 ›› 2024, Vol. 44 ›› Issue (11): 3548-3555.DOI: 10.11772/j.issn.1001-9081.2023101505

• 多媒体计算与计算机仿真 • 上一篇    下一篇

基于3D几何特征深度表达学习的物理系统仿真

林滏1, 石稼晟1, 高泽2, 楚遵康2, 马琼敏3, 余海燕2, 饶卫雄1()   

  1. 1.同济大学 软件学院,上海 201804
    2.同济大学 汽车学院,上海 201804
    3.复杂系统仿真总体重点实验室,北京 100101
  • 收稿日期:2023-11-03 修回日期:2023-12-08 接受日期:2023-12-15 发布日期:2024-11-13 出版日期:2024-11-10
  • 通讯作者: 饶卫雄
  • 作者简介:林滏(2000—),男,福建宁德人,硕士研究生,CCF会员,主要研究方向:图神经网络、面向有限元的人工智能
    石稼晟(2000—),男,四川成都人,博士研究生,CCF会员,主要研究方向:图神经网络、面向有限元的人工智能
    高泽(1998—),男,江苏常州人,博士研究生,主要研究方向:图神经网络在汽车结构仿真中的应用
    楚遵康(1999—),男,山东济宁人,博士研究生,主要研究方向:机器学习在汽车结构设计中的应用
    马琼敏(1988—),女,广东普宁人,助理研究员,博士,主要研究方向:体系设计与评估、计算机仿真
    余海燕(1976—),女,安徽安庆人,教授,博士,主要研究方向:汽车结构优化、汽车制造工艺
  • 基金资助:
    国家重点研发计划项目(2022YFE0208000)

Physical system simulation based on deep representation learning for 3D geometric features

Fu LIN1, Jiasheng SHI1, Ze GAO2, Zunkang CHU2, Qiongmin MA3, Haiyan YU2, Weixiong RAO1()   

  1. 1.School of Software Engineering,Tongji University,Shanghai 201804,China
    2.School of Automotive Studies,Tongji University,Shanghai 201804,China
    3.National Key Laboratory for Complex Systems Simulation,Beijing 100101,China
  • Received:2023-11-03 Revised:2023-12-08 Accepted:2023-12-15 Online:2024-11-13 Published:2024-11-10
  • Contact: Weixiong RAO
  • About author:LIN Fu, born in 2000, M. S. candidate. His research interests include graph neural network, AI for FEM.
    SHI Jiasheng, born in 2000, Ph. D. candidate. His research interests include graph neural network, AI for FEM.
    GAO Ze, born in 1998, Ph. D. candidate. His research interests include application of graph neural network in automotive structural simulation.
    CHU Zunkang, born in 1999, Ph. D. candidate. His research interests include application of machine learning in automotive structural design.
    MA Qiongmin, born in 1988, Ph. D., research assistant. Her research interests include system design and evaluation, computer simulation.
    YU Haiyan, born in 1976, Ph. D., professor. Her research interests include automotive structural optimization, automotive manufacturing processes.
  • Supported by:
    National Key Research and Development Program of China(2022YFE0208000)

摘要:

针对现有深度学习方法在物理系统仿真中无法处理几何边界与初始条件同时变化的场景的问题,提出将几何边界约束的表达与物理系统仿真解耦的技术思路,设计了几何特征表达学习和物理系统仿真双步骤的技术路线。在构建与外部物理条件无关的几何特征提取模块之后,融合提取的几何特征与物理特征,最后设计基于神经网络的物理系统仿真方法。在应力场预测实验中,所提方法的预测时间为2.63 ms,远低于有限元法(FEM)的0.6 s,且平均绝对误差(MAE)仅为MeshNet的0.389倍。实验结果表明,所提方法能够保持较高仿真精度,同时能够较好地适应不同的几何边界与初始条件。

关键词: 有限元法, 物理系统仿真, 3D网格, 特征表达, 平均绝对误差

Abstract:

To address the limitations of the existing deep learning methods in handling scenarios where both geometric boundaries and initial conditions vary in physical simulation problems, a technical approach was proposed to decouple the representation of geometric boundary constraints from the physical system simulation, and a two-step technical route of geometric feature representation learning and physical system simulation was designed. After constructing an independent geometric feature extraction module which was unaffected by external physical conditions, the extracted geometric features were fused with physical features, and finally a neural network-based physical system simulation model was designed. In stress field prediction experiments, the proposed method achieves a prediction time of only 2.63 ms, which is much lower than 0.6 s of Finite Element Method (FEM), and has a Mean Absolute Error (MAE) only 0.389 times of that of MeshNet. Experimental results demonstrate that the proposed method maintains high simulation accuracy while effectively adapting to different geometric boundaries and initial conditions.

Key words: Finite Element Method (FEM), physical system simulation, 3D mesh, feature representation, Mean Absolute Error (MAE)

中图分类号: