机器学习耦合有限元分析预测板料气弯回弹行为  

Springback behavior on sheet metal in gas bending predicted by machine learning coupled with finite element analysis

作　　者：徐承亮[1] 张祥林[2] 王大军[3] Xu Chengliang;Zhang Xianglin;Wang Dajun(Industry College,Guangzhou Vocational College of Technology and Busines,Guangzhou 511442,China;State Key Laboratory of Material Forming and Die Technology,Huazhong University of Science and Technology,Wuhan 430074,China;College of Automation,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]广州科技贸易职业学院产业学院,广东广州511442 [2]华中科技大学材料成形与模具技术国家重点实验室,湖北武汉430074 [3]重庆邮电大学自动化学院,重庆400065

出　　处：《锻压技术》2022年第6期107-112,共6页Forging & Stamping Technology

基　　金：广东省普通高校特色创新项目(自然科学类)(2018GKTSCX053);2021年度广州市基础研究计划基础与应用基础研究项目(2021-02-08-13-0018);材料成形与模具技术国家重点实验室基金资助项目(P2021-016)。

摘　　要：采用机器学习神经网络(NN)耦合有限元分析(FEA)的方法来构建弯曲成形过程的非线性回弹模型,并且考虑了不同材料、工艺参数和模具几何形状,可以有效和准确地预测工件的弯曲回弹行为。当模具开口量V=11 mm、板料厚度t=3 mm时,对于结构钢HC220材料,机器学习NN模型的预测值(Y_(NN))与回弹后分析解(y_(JBP))的均方根误差RMSE分别为0.28和1.70;对于双相钢DP590材料,Y_(NN)与y_(JBP)的RMSE分别为0.45和0.22。采用NN模型、回弹后分析解(y_(JBP))和FEA方法的CPU计算时间分别为3.1、6.3和278 s,NN模型的CPU计算时间最少,实验结果表明,NN模型可以在良好的预测精度和高效的求解速度之间达到一个最佳平衡。A nonlinear springback model of bending process was constructed by machine learning neural network(NN)coupled finite element analysis(FEA),and considering different materials,process parameters and die geometry shapes,the bending springback behavior of workpiece could be predicted effectively and accurately.When the die opening amount V=11 mm and the sheet thickness t=3 mm,for structural steel HC220 material,the root mean square errors RMSE of prediction value(Y_(NN))by machine learning NN model and analytical solution after springback(y_(JBP))were 0.28 and 1.70 respectively.For dual-phase steel DP590 material,the RMSE values of Y_(NN) and y_(JBP) were 0.45 and 0.22 respectively.The CPU calculation time of NN model,analytical solution after springback(y_(JBP))and FEA methods were 3.1,6.3 and 278 s respectively,and the NN model was of less CPU calculation time.The experimental results show that the NN model can achieve an optimal balance between good prediction accuracy and efficient solution speed.

关 键 词：空气辅助弯曲 弯曲回弹 机器学习 神经网络 有限元分析 

分 类 号：TG302[金属学及工艺—金属压力加工]