一种汽车空调支架力学性能快速仿真方法

doi:10.11871/jfdc.issn.2096-742X.2026.04.010

数据与计算发展前沿 ›› 2026, Vol. 8 ›› Issue (4): 136-148.

CSTR: 32002.14.jfdc.CN10-1649/TP.2026.04.010

doi: 10.11871/jfdc.issn.2096-742X.2026.04.010

一种汽车空调支架力学性能快速仿真方法

雷钧¹(),弓文娟²,曾多^2,^*(),石向成²,牛奔¹,刘敏学¹

¹ 北京理工大学深圳汽车研究院，广东深圳 518000
² 比亚迪汽车工业有限公司，广东深圳 518118

收稿日期:2026-01-08 出版日期:2026-08-20 发布日期:2026-08-21
通讯作者: *曾多（E-mail: zeng.duo@byd.com）
作者简介:雷钧，北京理工大学深圳汽车研究院（电动车辆国家工程实验室深圳研究院），硕士，工程师，主要研究方向为CAE仿真及结构优化。
本文承担工作为提出论文核心研究思路，主导研究方案设计与实施。
LEI Jun holds a master’s degree and is an engineer at the Shenzhen Automotive Research Institute, Beijing Institute of Technology. His main research interests include CAE simulation and structural optimization.
In this paper, he is mainly responsible for proposing the core research idea and leading the design and implementation of the research plan.
E-mail: leijun3013@sina.com|曾多，比亚迪汽车工业有限公司，博士，主要研究方向为AI+CAD/CAE。
本文承担工作为指导论文研究方案的优化与案例设计。
ZENG Duo, Ph.D., works at BYD Auto Industry Co., Ltd. His main research interests include AI+CAD/CAE.
In this paper, he is mainly responsible for guiding the optimization of the research proposal and the design of case studies.
E-mail: zeng.duo@byd.com
基金资助:
科学技术部汽车设计软件工具链开发及平台建设项目(S-KJ062022Y0002);深圳市高层次人才团队项目孔雀团队(KQTD20200820113110016)

A Rapid Simulation Method for Mechanical Properties of Automotive Air Conditioning Brackets

LEI Jun¹(),GONG Wenjuan²,ZENG Duo^2,^*(),SHI Xiangcheng²,NIU Ben¹,LIU Minxue¹

¹ Shenzhen Automotive Research Institute, Beijing Institute of Technology, Shenzhen, Guangdong 518000, China
² BYD Auto Industry Co., Ltd., Shenzhen, Guangdong 518118, China

Received:2026-01-08 Online:2026-08-20 Published:2026-08-21

摘要/Abstract

摘要：

【目的】 针对传统有限元分析在汽车空调支架设计中存在的计算耗时长、优化迭代效率低等问题，本文提出一种新的力学性能快速仿真方法。【方法】 该方法通过参数化建模与实验设计构建高保真数据集，在深度算子网络模型的基础上，结合傅里叶位置编码、蛇形激活函数与注意力机制，构建“参数—空间—全场—指标”的统一映射架构。以参数描述设计与工况，以空间坐标描述结构与边界，在同一网络中实现跨模态对齐与整场预测，并据此自动提取最大应力、模态频率等关键指标。【结果】 消融实验证明该模型具备高精度、高鲁棒性及高时效性；应用实例表明，本方法在保持工程精度的同时，可将汽车空调支架的随机振动与模态分析评估时间从小时级缩短至十秒级，最大应力与一阶模态频率预测相对误差均满足小于5%的企业整体误差预测精度要求，【结论】 显著提升了企业产品结构设计迭代效率与优化设计能力。

关键词: 汽车空调支架, 深度算子网络, 力学性能预测, 快速仿真, 优化设计

Abstract:

[Objective] To address the problems of time-consuming computation and low optimization iteration efficiency inherent in traditional Finite Element Analysis in the design of automotive air conditioning brackets, this paper proposes a new rapid simulation method for mechanical performance. [Methods] A high-fidelity dataset is constructed via parametric modeling and design of experiments. Based on the Deep Operator Network model, this method incorporates Fourier positional encoding, a Snake activation function, and an attention mechanism to establish a unified mapping framework for the ‘Parameter-Space-Full-Field-Indicator’ chain. Parameters are used to characterize design schemes and operating conditions, while spatial coordinates describe structural geometries and boundary conditions. Cross-modal alignment and full-field prediction are implemented within the same network, and key indicators including maximum stress and modal frequency are automatically extracted accordingly. [Results] Ablation experiments validate that the proposed model possesses high accuracy, high robustness, and high computational efficiency. Case studies show that this method can reduce the evaluation time of random vibration and modal analysis for automotive air conditioning brackets from hours to tens of seconds, while ensuring engineering-level precision. The relative prediction errors of the maximum stress and the first-order modal frequency both meet the enterprise’s overall precision requirement for predictive errors, with the allowable margin set at less than 5%. [Conclusions] The proposed method significantly improves the iteration efficiency of product structural design and enterprises’ optimal design capabilities

Key words: automotive air conditioning bracket, DeepONet, mechanical performance prediction, rapid simulation, optimal design

雷钧, 弓文娟, 曾多, 石向成, 牛奔, 刘敏学. 一种汽车空调支架力学性能快速仿真方法[J]. 数据与计算发展前沿, 2026, 8(4): 136-148.

LEI Jun, GONG Wenjuan, ZENG Duo, SHI Xiangcheng, NIU Ben, LIU Minxue. A Rapid Simulation Method for Mechanical Properties of Automotive Air Conditioning Brackets[J]. Frontiers of Data and Computing, 2026, 8(4): 136-148, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2026.04.010.

图/表 13

图1

图2

图3

图4

图5

表1

表2

表3

图6

表4

图7

参考文献 20

[1]	陈树生, 贾苜梁, 林家豪, 等. 生成式模型赋能飞行器技术应用研究进展与展望[J]. 航空学报, 2025 46(11): 631194.
[2]	LIN J H, CHEN S H, JIN S Y, et al. A velocity-prediction lightweight diffusion model for three-dimensional aircraft aerodynamic inverse design[J]. Engineering Applications of Artificial Intelligence, 2026, 163: 113125. doi: 10.1016/j.engappai.2025.113125
[3]	RAISSI M, PERDIKARIS P, KARNIADAKIS G E. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear PDEs[J]. Journal of Computational Physics, 2019, 378: 686-707. doi: 10.1016/j.jcp.2018.10.045
[4]	GILMER J, SCHOENHOLZ S S, RILEY P F, et al. Neural Message Passing for Quantum Chemistry[C]. Proceedings of the 34th International Conference on Machine Learning, 2017, 70: 1263-1272.
[5]	MAURIZI M, GAO C, BERTO F. Predicting stress, strain and deformation fields in materials and structures with graph neural networks[J]. Scientific Reports, 2022, 12: 21834. doi: 10.1038/s41598-022-26424-3 pmid: 36528676
[6]	ZHAI H F. Stress Predictions in Polycrystal Plasticity using Graph Neural Networks with Subgraph Training[J]. Computational Mechanics, 2025, 75 (6): 147.
[7]	SÁNCHEZ-GONZÁLEZ A, GODWIN J, PFAFF T, et al. Learning to Simulate Complex Physics with Graph Networks[C]. ICML, 2020, 119: 8459-8468.
[8]	SITZMANN V, MARTEL J N P, BERGMAN A W, et al. Implicit Neural Representations with Periodic Activation Functions(SIREN)[C]. NeurIPS, 2020, 33: 7462-7473.
[9]	QI C R, SU H, MO K, et al. PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation[C]. CVPR, 2017: 652-660.
[10]	QI C R, LI Y, SU H, et al. PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space[J]. NeurIPS, 2017, 30: 5105-5114.
[11]	KOVACHKI N, LI Z Y, LIU B, et al. Neural Operator: Learning Maps Between Function Spaces[J]. Journal of Machine Learning Research, 2023, 24(89): 1-97.
[12]	LI Z Y, KOVACHKI N, AZIZZADENESHELI K, et al. Neural Operator: Graph Kernel Network for Partial Differential Equations[C]. ICLR 2020 Workshop on Integration of Deep Neural Models and Differential Equations, March 10, 2020: 1-15.
[13]	LI Z Y, KOVACHKI N, LIU B, et al. Multipole Graph Neural Operator for Parametric Partial Differential Equations[C]. NeurIPS, 2020: 6755-6766.
[14]	OSHER S, SETHIAN J A. Fronts propagating with curvature-dependent speed: Algorithms based on Hamilton-Jacobi formulations[J]. Journal of Computational Physics, 1988, 79(1): 12-49. doi: 10.1016/0021-9991(88)90002-2
[15]	ZENG C Y, ZHANG Y S, ZHOU J Y, et al. Point cloud neural operator for parametric PDEs on complex and variable geometries[J]. Computer Methods in Applied Mechanics and Engineering, 2025, 443: 118022. doi: 10.1016/j.cma.2025.118022
[16]	WANG H H, ZHOU W H, WANG H, et al. A DeepONets-based resolution independent ABC inverse method for determining material parameters of HAZ[J]. Engineering Fracture Mechanics, 2025, 315: 110843. doi: 10.1016/j.engfracmech.2025.110843
[17]	LU L, JIN P Z, PANG G F, et al. Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators[J]. Nature Machine Intelligence, 2021, 3(3): 218-229. doi: 10.1038/s42256-021-00302-5
[18]	HE J Y, KORIC S, ABUEIDDA D, et al. Geom-DeepONet: A point-cloud-based deep operator network for field predictions on 3D parameterized geometries[J]. Computer Methods in Applied Mechanics and Engineering, 2024, 429: 117130. doi: 10.1016/j.cma.2024.117130
[19]	WU W D, ZHANG Y, HAO L L, et al. Physics-informed partitioned coupled neural operator for complex networks[J]. Engineering Applications of Artificial Intelligence, 2025, 159: 111567. doi: 10.1016/j.engappai.2025.111567
[20]	LI Z Y, KOVACHKI N, CHOY C, et al. Geometry-Informed Neural Operator(GINO) for PDEs on irregular domains/large-scale 3D PDEs[C]. NeurIPS, 2023: 4567-4578.

名称	单位	下限	初始值	上限
变量1	mm	0.5	2	4
变量2	mm	-10	0	10
变量3	mm	-5	0	5
变量4	mm	-5	0	5

ID	变量1/mm	变量2/mm	变量3/mm	变量4/mm	总质量/t	一阶频率/Hz	最大RMS应力/MPa
1	2.7753	-3.5920	4.3019	-0.9347	0.0004	87.1755	8.064947
2	1.7001	1.7338	-3.8618	-1.0602	0.0003	42.8172	18.2087
3	1.9518	1.3868	-1.4713	0.5024	0.0003	53.2871	13.43852
4	1.5710	3.2669	4.7207	0.6496	0.0003	39.0019	34.2015
5	1.7436	4.3659	-3.7474	-1.5175	0.0003	43.9257	16.95614
6	1.2421	1.9387	3.7436	-0.1697	0.0003	27.5663	132.4294
7	2.6258	-4.7100	-3.2766	0.8793	0.0004	82.4137	8.809083
8	1.4680	1.2715	4.5587	-1.7636	0.0003	34.4639	76.8751
9	1.2556	1.3049	2.8629	-0.7124	0.0003	27.8433	130.6451
10	1.6696	-3.0785	1.8566	1.5530	0.0003	43.5215	19.21172
11	2.0919	0.0016	-3.2303	0.2457	0.0004	58.7661	12.10574
12	2.1431	-2.4072	-2.0578	-1.6226	0.0004	59.6862	12.1288
13	2.6196	3.3918	0.6393	-0.2790	0.0004	80.1414	8.115237

模型	R²	MAE	RMSE	L₂	预测时间（s）
Baseline	0.99919	0.02455	0.04585	0.01952	0.1268
only_Snake	0.99920	0.02408	0.04539	0.0193	0.1271
only_Fourier	0.99929	0.01596	0.04288	0.01826	0.1237
only_Attention	0.99898	0.02346	0.05135	0.02186	0.1422
no_Snake	0.99858	0.01741	0.06079	0.02588	0.2830
no_Fourier	0.99910	0.02419	0.04839	0.02061	0.2744
no_Attention	0.99926	0.01599	0.04387	0.01868	0.2822
FAS-DeepONet	0.99943	0.01389	0.03821	0.01630	0.1253

ID	变量1/mm	变量2/mm	变量3/mm	变量4/mm	总质量/t	一阶频率/Hz	最大RMS应力/MPa
1	1.6	-1.1	-5.0	2.0	0.0003156	40.93	14.35
2	1.6	-4.8	-4.9	2.0	0.0003156	41.20	14.34
3	1.6	-4.9	3.7	2.0	0.0003156	41.31	14.12

一种汽车空调支架力学性能快速仿真方法

A Rapid Simulation Method for Mechanical Properties of Automotive Air Conditioning Brackets

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 20

相关文章 0

编辑推荐

Metrics

本文评价