汽车工程 ›› 2022, Vol. 44 ›› Issue (10): 1600-1608.doi: 10.19562/j.chinasae.qcgc.2022.10.015

所属专题: 车身设计&轻量化&安全专题2022年

• • 上一篇    下一篇

基于深度学习的座椅抗挥鞭伤性能预测

张绍伟1(),朱大炜2,翟光照2   

  1. 1.法国ESI集团上海分公司,上海  200000
    2.上汽大众汽车有限公司产品研发车身研发部,上海  200000
  • 收稿日期:2022-07-30 修回日期:2022-09-02 出版日期:2022-10-25 发布日期:2022-10-21
  • 通讯作者: 张绍伟 E-mail:zhangganzi@126.com

Prediction on Seat’s Anti-whiplash-injury Performance Based on Deep Learning

Shaowei Zhang1(),Dawei Zhu2,Guangzhao Zhai2   

  1. 1.ESI-Group Shanghai Office,Shanghai  200000
    2.SAIC Volkswagen Automotive Co. ,Ltd. ,Shanghai  200000
  • Received:2022-07-30 Revised:2022-09-02 Online:2022-10-25 Published:2022-10-21
  • Contact: Shaowei Zhang E-mail:zhangganzi@126.com

摘要:

在传统仿真的基础上,结合深度学习,提出了一种快速预测座椅鞭打性能的方法。首先对上汽大众某车型座椅进行了一系列材料级、零部件级、分总成级和整椅级的静态与动态物理实验,其次利用实验结果对已有的仿真模型进行了标定,标定结果验证了仿真模型的有效性。然后,利用全因子法对所有影响座椅鞭打性能的因素进行了仿真。基于仿真结果,利用深度学习方法建立了长短记忆(LSTM)神经网络模型,对假人的挥鞭伤害响应进行快速预测。结果表明:基于LSTM的神经网络模型预测的假人响应曲线能与仿真得到的曲线较好地吻合,故可用于后续的座椅鞭打性能优化。

关键词: 座椅抗挥鞭性能, 深度学习, 神经网络, 有限元仿真, BioRID II, Pam-Crash

Abstract:

On the base of traditional simulation and combined with deep learning, a rapid prediction method of seat’s anti-whiplash injury performance is proposed. Firstly, a series material level, component level, subassembly level and seat level static and dynamic physical experiments are carried out on a Shanghai VW’s vehicle seat. Then using the results of experiments to calibrate the existing simulation model, resulting in the effectiveness of the simulation model verified. Next, a simulation on all factors affecting the seat’s whiplash performance is conducted by using full-factor method, and based on simulation results and using deep learning method, a long- and short-term memory (LSTM) neural network model is established to rapidly predict the whiplash injury response of dummy. The results show that the dummy response curve obtained from prediction by LSTM neural network model agrees well with simulated curve, so can be used in subsequent seat’s whiplash performance optimization.

Key words: seat anti-whiplash performance, deep learning, neural network, finite element simulation, BioRIDII, Pam-Crash