基于BS-TabNet和LSSA的车架智能轻量化设计  

Intelligent Lightweight Chassis Design Based on BS-TabNet and LSSA

作　　者：聂昕[1] 刘文涛 陈少伟[1] 张承霖[1] 陈勇[1] 杨昊 NIE Xin;LIU Wentao;CHEN Shaowei;ZHANG Chenglin;CHEN Yong;YANG Hao(Hunan University,State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body,Changsha 410082,China;Zhuzhou CRRC Times Electric Co.,Ltd.,Zhuzhou 412000,China;Aisheng Automotive Technology Development Co.,Ltd.,Changsha 410205,China)

机构地区：[1]湖南大学整车先进设计制造技术全国重点实验室,湖南长沙410082 [2]株洲中车时代电气股份有限公司,湖南株洲412000 [3]湖大艾盛汽车技术开发有限公司,湖南长沙410205

出　　处：《湖南大学学报（自然科学版）》2024年第2期163-176,共14页Journal of Hunan University:Natural Sciences

基　　金：广西区科技计划重大专项(桂科AA23062072);柳州市科技计划重大专项(2022ABA0104;2022ABB0101);国家重点研发计划资助项目(2019YFB1706504)。

摘　　要：为解决传统牵引车车架轻量化设计中设计周期长、设计难度较大和过分依赖工程师经验等问题,提出了一种智能轻量化设计方法.首先,通过试验设计(Design of Experiments,DOE)联合仿真获取车架性能表格数据.其次,基于TabNet算法、贝叶斯优化算法和沙普利增量解释理论(SHapley Additive exPlanation,SHAP)构建出BS-TabNet模型,用于学习车架性能表格数据,生成车架代理模型.最后,采用莱维飞行策略对麻雀搜索算法(Sparrow Search Algorithm,SSA)进行改进,得到莱维麻雀搜索算法(Levy Sparrow Search Algorithm,LSSA),用于求解车架轻量化任务,找到最优车架结构参数.相较于传统机器学习算法,BS-TabNet模型在准确性、稳定性和可解释性三个方面都有着更高的评价,其准确度达到了0.98左右,稳定性提高50%以上,而且具有更强的可解释能力,解决了深度学习在表格型数据上表现较差的问题.相较于传统群智能优化算法,LSSA算法能够寻找到更好的优化结果,在满足其他性能要求的同时,实现了车架质量减轻5.64%的轻量化效果.智能轻量化设计方法将人工智能与车架轻量化设计相结合,能够节省大量设计时间,提高设计效率.In order to address the issues of long design cycles,high design complexity,and excessive reliance on engineer experience in traditional lightweight design of tractor chassis,an intelligent lightweight design method is proposed.Firstly chassis performance data is obtained through Design of Experiments(DOE)combined simulation.Then,the BS-TabNet model is constructed based on the TabNet algorithm,Bayesian optimization algorithm,and SHapley Additive exPlanation(SHAP)theory.This model is used to learn from the chassis performance data and generate a surrogate model for the chassis.Finally,the Levy flight strategy is applied to improve the Sparrow Search Algorithm(SSA),resulting in the Levy Sparrow Search Algorithm(LSSA),which is used to solve the lightweight design task and find the optimal structural parameters for the chassis.Compared with traditional machine learning algorithms,the BS-TabNet model shows higher accuracy,stability,and interpretability.Its accuracy reaches around 0.98,stability is improved by over 50%,and it has stronger interpretability,addressing the poor performance of deep learning on tabular data.Compared with traditional swarm intelligence optimization algorithms,the LSSA algorithm can obtain better optimization results.While meeting other performance requirements,it achieves a 5.64%reduction in chassis weight.The intelligent lightweight design method combines artificial intelligence with chassis lightweight design,and it can save a significant amount of design time and improve design efficiency.

关 键 词：深度学习 贝叶斯优化 车架设计 TabNet SHAP SSA 

分 类 号：U463.32[机械工程—车辆工程] TP181[交通运输工程—载运工具运用工程]