基于小波神经网络和粒子群算法的铝合金板冲压回弹工艺参数优化  

Parameters optimization of stamping and springback for aluminum-alloy sheet based on wavelet neural network and particle swarm optimization algorithm

作　　者：孙新强[1] 谢延敏[1] 田银[1] 何育军[1] 

机构地区：[1]西南交通大学机械工程学院先进设计与制造技术研究所,四川成都610031

出　　处：《锻压技术》2015年第1期137-142,共6页Forging & Stamping Technology

基　　金：国家自然科学基金资助项目(51005193)

摘　　要：针对铝合金复杂件冲压后出现的较大回弹缺陷,同时为减少冲压成形工艺参数的优化时间,使用有限元仿真软件DYNAFORM对冲压成形及回弹过程进行数值模拟,在确保数值模拟与试验结果基本一致的基础上,利用代理模型对回弹进行了优化研究。以NUMISHEET'96 S梁为研究对象,凸模圆角半径、凹模圆角半径、压边力、板料厚度作为影响因素,成形后最大回弹值作为成形目标,运用拉丁超立方抽样,通过数值仿真获得样本数据,建立影响因素与成形目标之间的小波神经网络代理模型,利用粒子群算法对该模型迭代寻优获得最优工艺参数。结果表明:小波神经网络能较好地描述板料工艺参数与回弹之间的映射关系,优化后成形件的回弹量大大减小。For the large springback appeared after the stamping process of aluminum-alloy sheet, and for reducing the time of optimizing process parameters, the stamping process and springback were numerically simulated based on the finite element analysis software DYNAFORM. On the basis of the numerical simulation consisting with experimental results, the agent model was used to the optimization research of springback. The S-rail of NUMISHEET＇96 was taken into account, with the fillet radius of punch, the fillet radius of die, the blank holder force and the sheet thickness as influencing factors, and the maximum springback value after stamping was regarded as forming target. By using latin hypercube to sample, and the simulation was carried to get the samples, and the wavelet neural network agent model between influencing factors and forming target was built. Then the optimal solution was obtained by iterations of particle swarm optimization algorithm. The results show that the wavelet neural network agent model can describe the input-output relationship between the sheet forming process parameters and the springback, and the springback can be remarkably reduced after the optimization.

关 键 词：铝合金 回弹 小波神经网络 粒子群算法 

分 类 号：TG386[金属学及工艺—金属压力加工]