离合器片锻压工艺的神经网络优化研究  

Study on Clutch Plates Forging Process Optimized by Neural Network

作　　者：张玲[1] 曹青华 ZHANG Ling;CAO Qinghua(Jiangsu Vocational College of Business, Nantong 226000, China;School of Mechanical Engineering, Nantong University, Nantong 226000, China)

机构地区：[1]江苏商贸职业学院,江苏南通226000 [2]南通大学机械工程学院,江苏南通226000

出　　处：《热加工工艺》2018年第7期151-153,158,共4页Hot Working Technology

摘　　要：以离合器片材料、模具温度、始锻温度、终锻温度和锻压速度为输入层参数,以磨损性能为输出层参数,采用不同训练函数构建出5×40×8×1四层结构的离合器片锻压工艺优化神经网络模型。结果表明,当训练函数选用trainlm、traingd和traingdm函数时,神经网络的相对训练误差分别在2.6%~4.7%、3.1%~5.6%、1.9%~3.4%。以traingdm函数作为训练函数的离合器片锻压工艺优化神经网络相对预测误差在2.1%~3.3%,具有较强的预测能力和较高的预测精度。Taking the clutch plate material, die temperature, initial forging temperature, final forging temperature and forging speed as input layer parameters, and taking the abrasion performance as the output layer parameter, the neural network model（four layers topology structure of 5×40×8×1）of forging process optimization of the clutch plate was built by using the different training functions. The results show that when the training function uses trainlm, traingd and traingdm, the relative training errors of the neural network are 2.6%-4.7%, 3.1%-5.6% and 1.9%-3.4%, respectively. Taking traingdm function as training function, the neural network relative prediction error of forging process optimization for clutch plate is 2.1%-3.3%,which has strong prediction ability and high prediction accuracy.

关 键 词：训练函数 锻压工艺优化 神经网络 离合器片 磨损性能 相对训练误差 

分 类 号：TG301[金属学及工艺—金属压力加工]