基于正交设计的神经网络训练样本的选择方法及其在冷挤压工艺设计中的应用研究  

Research on selecting approach of ANN training samples based on taguchi method and its applications to cold extrusion process design

作　　者：高军[1] 张承瑞[2] 季廷炜[1] 赵国群[1] 

机构地区：[1]山东大学材料科学与工程学院 [2]山东大学机械工程学院,济南250061

出　　处：《塑性工程学报》2006年第5期32-35,共4页Journal of Plasticity Engineering

基　　金：山东省优秀中青年科学家奖励基金资助项目(2005BS05005)

摘　　要：冷挤压件的形状是制定冷挤压工艺方案时需要考虑的最重要的因素之一,基于目的性和方便性原则,该文针对冷挤压件的形状特征进行分类,并在此基础上分别建立基于正交实验设计的人工神经网络训练样本子集,最后将所有子集进行整合,形成初始训练测试样本集。结合典型冷挤压件,分析了采用正交设计方法选取人工神经网络训练样本的过程,研究了基于正交设计的人工神经网络的训练、测试方法,实现了基于人工神经网络的冷挤压工艺设计。Part shape plays great role in setting down the process design proiect. Therefore, in order to simplify the problem, the cold extruded parts were classified according to their shapes first of all, taguchi-method-based training sample subset standing for different part sort was established, and then the original whole training set was formed through the combination of all the subsets. In the end of this paper, the selecting process of taguchi-method-based training sample subset was analyzed, the taguchimethod-based training and testing process of ANN was also discussed, ANN based cold extrusion process design was carried out, by applying typical cold extruded part.

关 键 词：冷挤压工艺设计 人工神经网络 训练样本 正交设计 

分 类 号：TG376[金属学及工艺—金属压力加工]