基于人工神经网络和遗传算法的封头成形工艺参数多目标优化  

Multi-objective optimization on head forming process parameters based on artificial neural network and genetic algorithm

作　　者：李雷[1] 赵柏森[2] Li Lei;Zhao Baisen(Department of Vehicle Engineering,Chongqing Industry Polytechnic College,Chongqing 401120,China;Department of Mechanical Engineering and Automation,Chongqing Industry Polytechnic College,Chongqing 401120,China)

机构地区：[1]重庆工业职业技术学院车辆工程学院,重庆401120 [2]重庆工业职业技术学院机械工程与自动化学院,重庆401120

出　　处：《锻压技术》2021年第5期39-45,共7页Forging & Stamping Technology

基　　金：重庆市教委科学技术研究重点项目(KJZD-K201803201);重庆市教委科学技术研究项目(KJ1503105)。

摘　　要：以DYNAFORM数值模拟、人工神经网络和NSGA-Ⅱ多目标遗传算法为研究手段,以奥氏体不锈钢大型封头作为研究对象,将上下模间隙、下模圆角半径、拉延筋高度、拉延筋相对位置作为优化变量,将起皱、厚度不均匀性、回弹量作为优化指标,利用人工神经网络代表优化变量与优化指标之间的关系,采用NSGA-Ⅱ多目标遗传算法优化人工神经网络,获得帕累托前沿解集。最后,从帕累托前沿解集中选取比较合适的解为:上下模间隙为10.99 mm、下模圆角半径为44.96 mm、拉延筋高度为39.97 mm、拉延筋相对位置为0.4,此时的起皱、厚度不均匀性、回弹量3个指标均较小且均衡。将这组最优工艺参数进行试验试制,得到了表面光滑、厚度均匀、无折皱以及回弹较小的封头构件。Based on numerical simulation software DYNAFORM,artificial neural network and NSGA-Ⅱ multi-objective genetic algorithm,for the large austenitic stainless steel head,taking clearance between upper and lower dies,lower die fillet radius,drawbead height and relative position of drawbead as the optimization variables,and taking wrinkle,thickness unevenness and springback amount as the optimization indexes,the relationship between the optimization variables and the optimization indexes was represented by the artificial neural network which was optimized by NSGA-Ⅱ multi-objective genetic algorithm,and the Pareto front solution set was obtained finally.Then,the reasonable solutions were selected from the Pareto front solution set with the clearance between upper and lower dies of 10.99 mm,the lower die fillet radius of 44.96 mm,drawbead height of 39.97 mm and the relative position of drawbead of 0.4,and the three indexes of wrinkle,thickness unevenness and springback amount were relatively smaller and balanced.Finally,this set of optimal process parameters were tested and trial-produced,and the head parts with smooth surface,uniform thickness,no wrinkle and less springback were obtained.

关 键 词：大型封头 NSGA-II遗传算法 人工神经网络 冲压成形 帕累托前沿解集 多目标优化 

分 类 号：TG386.41[金属学及工艺—金属压力加工]