基于图网络的汽车零部件应力场快速预测方法研究

doi:10.19562/j.chinasae.qcgc.2024.01.018

摘要/Abstract

摘要：

有限元分析作为一种重要的计算机辅助工程技术，在汽车零部件开发中起着极为重要的作用，但是在分析复杂问题时耗时长，影响开发周期。为此，本文提出了一种以有限元仿真提供样本数据、以图网络技术建立有限元输入与输出间的映射关系的神经网络方法，将该图网络方法应用于座椅骨架总成的应力场预测。该预测方法中采用图节点和图边模拟了有限元模型中节点间的连接关系，由此表达有限元模型中单元之间的拓扑关系，并将预测结果与有限元仿真结果进行了对比。结果表明该方法能准确预测座椅骨架总成的最大应力及其出现的位置，对应力分布一致性均有较好的预测能力，且计算速度比相应有限元求解度快3个数量级以上。

关键词: 有限元分析, 深度学习, 图网络, 汽车座椅

Abstract:

Finite Element Analysis （FEA）， as an important Computer-aided Engineering （CAE） technology， plays a significant role in the area of automotive part development. However， it costs too much time when solving complicated problems， which affects the development cycle. In this paper， a neural network method is proposed， in which sample data is provided by finite element simulation and the mapping relationship between finite element input and output is established by graph network technology. The graph network method is used to predict the stress field of the seat frame assembly. The prediction method simulates the connection relationship between nodes in the finite element model using graph nodes and graph edges， which can effectively express the topological relationship between elements in the finite element model. The prediction results are compared with the results of the finite element simulation. The results show that the method can precisely predict the maximum stress and its corresponding location of the seat frame assembly， with strong predictive capabilities for stress distribution consistency. Additionally， the model has a significant computational advantage， with a calculation speed three orders of magnitude faster than that of the corresponding finite element solver.

Key words: finite element analysis, deep learning, graph network, automotive seat

高泽, 楚遵康, 石稼晟, 林滏, 饶卫雄, 余海燕. 基于图网络的汽车零部件应力场快速预测方法研究[J]. 汽车工程, 2024, 46(1): 170-178.

Ze Gao, Zunkang Chu, Jiasheng Shi, Fu Lin, Weixiong Rao, Haiyan Yu. Research on Fast Prediction Method of Stress Field of Automotive Parts Based on Graph Network[J]. Automotive Engineering, 2024, 46(1): 170-178.

图/表 22

图 1

图 2

图 3

图 4

图 5

图 6

表1

表2

图 7

表3

图 8

表4

表5

图 9

图 10

图 11

表6

表7

图 12

图 13

图 14

表8

参考文献 20

1	JIN J. The finite element method in electromagnetics［J］. Journal of the Japan Society of Applied Electromagnetics， 2002.
2	卢荡，杨文豪，吴海东，等.轮胎力学特性仿真高精度有限元建模方法研究［J］. 汽车工程， 2022， 44（10）： 1556-1562，1590.
	LU D， YANG W H， WU H D， et al. Research on high-accuracy finite element modeling method for tire mechanics characteristics simulation［J］. Automotive Engineering， 2022， 44（10）： 1556-1562，1590.
3	ZHAO M， ZHANG D Z， LIU F， et al. Mechanical and energy absorption characteristics of additively manufactured functionally graded sheet lattice structures with minimal surfaces［J］. International Journal of Mechanical Sciences， 2020，167：105262.
4	SENER A S. Fatigue life resolution of the steering wheel tie rod of a LCV with FEA［J］. Mechanika， 2017，23（5）：622-629.
5	刘长钊，张铁，宋健，等.纯电动汽车电驱动系统耦合动力学研究［J］. 汽车工程， 2022， 44（12）：1896-1909.
	LIU C Z， ZHANG T， SONG J， et al. Study on coupling dynamics of electric drive system of pure electric vehicle［J］. Automotive Engineering， 2022， 44（12）：1896-1909.
6	BLACK N， NAJAFI A R. Learning finite element convergence with the multi-fidelity graph neural network［J］. Computer Methods in Applied Mechanics and Engineering， 2022， 397：115120.
7	OISHI A， YAGAWA G. Computational mechanics enhanced by deep learning［J］. Computer Methods in Applied Mechanics and Engineering， 2017， 327： 327-351.
8	DESHPANDE S， LENGIEWICZ J， BORDAS S. Probabilistic deep learning for real-time large deformation simulations［J］. Computer Methods in Applied Mechanics and Engineering， 2022， 398： 115307.
9	LIANG L， LIU M， MARTIN C， et al. A deep learning approach to estimate stress distribution： a fast and accurate surrogate of finite-element analysis［J］. Journal of The Royal Society Interface， 2018， 15（138）： 20170844.
10	NIE Z， JIANG H， KARA L B. Stress field prediction in cantilevered structures using convolutional neural networks［J］. Journal of Computing and Information Science in Engineering， 2020， 20（1）： 011002.
11	ZHOU H， XU Q， NIE Z， et al. A study on using image-based machine learning methods to develop surrogate models of stamp forming simulations［J］. Journal of Manufacturing Science and Engineering， Transactions of the ASME， 2022， 144（2）： 1087-1357.
12	FENG Y T， FENG Y F， YOU H X， et al. MeshNet： mesh neural network for 3D shape representation［C］. AAAI Conference on Artificial Intelligence， 2019： 8279-8286.
13	张绍伟，朱大炜，翟光照.基于深度学习的座椅抗挥鞭伤性能预测［J］. 汽车工程， 2022， 44（10）：1600-1608.
	ZHANG S W， ZHU D W， ZHAI G Z. Prediction on seat’s anti-whiplash-injury performance based on deep learning［J］. Automotive Engineering， 2022， 44（10）：1600-1608.
14	ATWOOD J， TOWSLEY D， LEE D D， et al. Diffusion-convolutional neural networks［J］. Advances in Neural Information Processing Systems， 2016， 29.
15	BATTAGLIA P W， HAMRICK J B， BAPST V， et al. Relational inductive biases， deep learning， and graph networks［J］. Statistics， 2018： 1467-5463.
16	SANCHEZ-GONZALEZ A， GODWIN J， PFAFF T， et al. Learning to simulate complex physics with graph networks ［C］. International Conference on Machine Learning. PMLR， 2020： 8459-8468.
17	汽车座椅、座椅固定装置及头枕强度要求和试验方法： GB 15083—2019 ［S］. 国家市场监督管理总局，中国国家标准化管理委员会， 2019.
	Strength requirement and test method of automobile seats， their anchorages and any head restraints： GB 15083—2019 ［S］. State Administration for Market Regulation， Standardization Administration， 2019.
18	MCKAY M D， BECKMAN R J， CONOVER W J. A comparison of three methods for selecting values of input variables in the analysis of output from a computer code［J］. Technometrics， 1979， 21（2）：239-245.
19	MOUSTAPHA M， MARELLI S， SUDRET B. Active learning for structural reliability： survey， general framework and benchmark［J］. Structural Safety， 2022， 96： 102174.
20	SHEN Y. Loss functions for binary classification and class probability estimation［M］. University of Pennsylvania， 2005.

编号	1	2	3	4	5	6	7	8	9	10	11	12
初始厚度/mm	1.4	1.2	2.0	1.8	3.4	1.8	1.4	3.4	4.5	7	2.0	2.0
采样下界/mm	1.0	0.84	1.4	1.26	2.4	1.26	1.0	2.4	3.15	4.9	1.4	1.4
采样上界/mm	1.8	1.56	2.6	2.34	4.4	2.34	1.8	4.4	5.85	9.1	2.6	2.6

硬件/系统/IDE	说明/版本
CPU	Inter（R） 13900K
GPU	Nvidia GeForce RTX 4090
显存	24 GB
CUDA	12.0
Windows	Windows11

传递步数	MSE	RMSE
15	1.100 842	0.277 184
20	0.755 519	0.190 884
25	0.558 019	0.141 125
30	0.543 721	0.140 798

编号	FEM仿真应力/MPa	预测应力/MPa	绝对误差/MPa	相对误差/%
1	437.799 7	431.977 45	5.822 3	1.329 9
2	330.791 7	331.800 72	1.009 0	0.305 0
3	383.638 9	385.096 37	1.457 5	0.379 9
4	476.107 9	477.871 03	1.763 1	0.370 3
5	464.164 3	470.417 97	6.253 7	1.347 3
6	411.926 5	413.097 35	1.170 9	0.284 2
7	355.065 1	354.502 30	0.562 8	0.158 5
8	369.533 5	369.539 15	0.005 7	0.001 5
9	393.871 8	394.924 13	1.052 3	0.267 2
10	412.051 8	413.097 44	1.045 6	0.253 8
平均	403.495 1	404.232 39	2.014 3	0.469 8

验证集编号	FEM预测的最大应力节点id	GNN预测的最大应力节点id	最大应力位置的距离/mm
1	44180	44183	4.079
2	42928	42928	0
3	42928	42928	0
4	44183	44180	4.079
5	44180	44180	0
6	42928	42928	0
7	42928	42928	0
8	42928	42928	0
9	44180	44180	0
10	42928	42928	0