基于冲压回弹控制的A柱上边梁零件智能设计方法研究  

Research on intelligent design method of A-pillar upper side beam part based on stamping springback control

作　　者：左哲 牛超 陈新平 胡志力 ZUO Zhe;NIU Chao;CHEN Xin-ping;HU Zhi-li(Hubei Key Laboratory of Advanced Technology for Automotive Components,Wuhan University of Technology,Wuhan 430070,China;Hubei Collaborative Innovation Center for Automotive Components Technology,Wuhan University of Technology,Wuhan 430070,China;Research Institute,BaoShan Iron&Steel Co.,Ltd.,Shanghai 201900,China;State Key Laboratory of Development and Application Technology of Automotive Steels(BaoSteel),Shanghai 201900,China)

机构地区：[1]武汉理工大学现代汽车零部件技术湖北省重点实验室,湖北武汉430070 [2]武汉理工大学汽车零部件技术湖北省协同创新中心,湖北武汉430070 [3]宝山钢铁股份有限公司研究院,上海201900 [4]汽车用钢开发与应用技术国家重点实验室(宝钢),上海201900

出　　处：《塑性工程学报》2021年第1期38-44,共7页Journal of Plasticity Engineering

基　　金：国家自然科学基金资助项目(51775397);高等学校学科创新引智计划项目(B17034);中国汽车产业创新发展联合基金资助项目(U1564202);教育部创新团队发展计划项目(IRT13087)。

摘　　要：针对汽车车身超高强钢结构件冲压成形回弹显著的问题,基于Tensor Flow机器学习框架,以A柱上边梁零件参数化特征为优化对象,建立总长、直弧比、半径等结构特征与扭转回弹之间的非线性函数模型,研究了A柱上边梁结构件智能设计方法。实验结果表明,当网络模型结构为13-12-1时,预测误差最小;可通过遗传算法优化神经网络参数提高神经网络的效率和准确率,将预测误差降低到3.4%;以扭转回弹为目标,基于遗传算法优化零件结构特征参数,并通过数值仿真实验验证了该优化方法的可靠性,实际仿真扭转回弹值与期望回弹值误差在5%以内,说明基于冲压回弹的A柱上边梁零件设计方法可行有效。Aiming at the problem that the stamping forming springback of ultra-high-strength steel structural parts of automobile body is significant,based on Tensor Flow machine learning framework,taking the parametric feature of the upper side beam part of A-pillar as the optimization object,a non-liner function model between structural features such as overall length,ratio of straight length and arc,radius and twist springback was established,and the intelligent design method of upper side beam structural part of A-pillar was studied.The experiment results show that the prediction error is the smallest when the network model structure is 13-12-1.The efficiency and accuracy of the neural network can be improved by optimizing neural network parameters using genetic algorithm,and the prediction error was reduced to 3.4%.Taking the twist springback as the target,the structural feature parameters of the part were optimized based on genetic algorithm,and the reliability of the optimization method was verified by numerical simulation experiments.The error between the actual simulated twist springback value and the expected springback value is within 5%,which indicates that the design method of upper side beam part of A-pillar based on stamping springback is feasible and effective.

关 键 词：BP神经网络 遗传算法 A柱上边梁 冲压成形 扭转回弹 

分 类 号：TG382[金属学及工艺—金属压力加工]