摘要
以Si C添加量、搅拌转速、搅拌时间、模具预热温度和浇注温度为5个输入层神经单元参数,以抗拉强度为输出层神经单元参数,采用不同训练函数,构建3层结构的铝基复合材料搅拌铸造工艺优化神经网络模型。结果表明:与traingd训练函数相比,采用trainlm训练函数的神经网络模型平均相对训练误差从4.4%减小到2.6%、收敛时的迭代运算次数从7924次减少到5186次。以trainlm函数作为训练函数的神经网络模型平均相对预测误差为2.7%,可用于实际的铝基复合材料搅拌铸造工艺优化。
Taking Si C addition amount,stirring speed,stirring time,mold preheating temperature and pouring temperature as five input layer neural unit parameters,and tensile strength as output layer neural unit parameter,different training functions were adopted,an three-layer optimized neural network model of aluminum matrix composite agitation casting process was constructed.The results show that compared with the traingd training function,the average relative training error of the neural network model using the trainlm training function decreases from 4.4%to 2.6%,the number of iterations at convergence decreases from 7924 to 5186.The average relative prediction error of the neural network model with trainlm function as the training function is 2.7%,which can be used to optimize the stirring casting process of aluminum matrix composites.
作者
张亚敏
姜永亮
ZHANG Yamin;JIANG Yongliang(Luohe Medical College,Luohe 462002,China;Department of Materials Science,Hainan Normal University,Haikou 571158,China)
出处
《热加工工艺》
北大核心
2021年第18期91-94,共4页
Hot Working Technology
基金
河南省科技厅基础与前沿技术研究计划项目(142300410105)。
关键词
铝基复合材料
搅拌铸造工艺优化
神经网络算法
训练函数
抗拉强度
aluminum matrix composite
stirring casting process optimization
neural network algorithm
training function
tensile strength
