西南交通大学学报2017，Vol.52Issue(5)：970-976,993,8.DOI:10.3969/j.issn.0258-2724.2017.05.018
基于SA-RBF神经网络的冲压成形拉延筋优化
Drawbead Optimisation in Stamping Using SA-RBF Neural Networks
摘要
Abstract
The structure of a radial basis function (RBF)neural network based on the k-means clustering algorithm was optimised by employing the simulated annealing algorithm for improving the prediction accuracy. The NUMISHEET 02 fender was considered as the object of research and six equivalent drawbead forces were used as input variables. Based on Spearman correlation analysis and Latin hypercube sampling,the data which had smaller correlation coefficient values were chosen as training samples for the simulated annealing-radial basis function (SA-RBF)neural network. The numerical simulations of training samples were performed by employing the Dynaform software package. The evaluation functions of forming quality were established based on the wrinkling defects and crack defects. The nonlinear relationship between the equivalent drawbead force and the associated objective function was established by incorporating a SA-RBF neural network. NSGA-Ⅱ algorithm was employed to achieve the Pareto frontier and the best equivalent drawbead forces were determined by applying grey correlation analysis theory. Finally,the numerical simulation of fender forming was performed based on the optimised drawbead forces. The resultant forming limit diagram (FLD)indicates decreased wrinkles in the optimised forming part and greater uniformity in the plastic deformation,thereby leading to improvement in the quality of fender forming.关键词
拉延筋/模拟退火算法/RBF神经网络/NSGA-Ⅱ算法Key words
drawbead/simulated annealing algorithm/RBF neural network/NSGA-Ⅱalgorithm分类
矿业与冶金引用本文复制引用
谢延敏,唐维,黄仁勇,熊文诚,卓德志..基于SA-RBF神经网络的冲压成形拉延筋优化[J].西南交通大学学报,2017,52(5):970-976,993,8.基金项目
国家色然科学基金资助项目(51005193) （51005193）
国家大学生创新创业训练计划项目(201710613033) （201710613033）
