基于神经网络算法的汽车用高强铝合金铸造性能优化  

Casting Performance Optimization of High Strength Aluminum Alloy for Automobile Based on Neural Network Algorithm

作　　者：叶进宝[1] 陈建华[1] 李相军 YE Jinbao;CHEN Jianhua;LI Xiangjun(Handan Polytechnic College,Handan 056001,China;School of Materials Science and Engineering,Henan University of Technology,Jiaozuo 454003,China)

机构地区：[1]邯郸职业技术学院,河北邯郸056001 [2]河南理工大学材料科学与工程学院,河南焦作454003

出　　处：《热加工工艺》2022年第11期73-75,78,共4页Hot Working Technology

基　　金：河北科技支撑计划项目(11212138)。

摘　　要：为了进行汽车用高强铝合金铸造性能优化,本文以合金元素、元素含量、熔炼温度、浇注温度为输入层参数,以合金的流动性作为输出层参数,选用Purelin函数、Tansig函数和Trainlm函数分别作为神经网络模型的输出层传递函数、隐含层传递函数、训练函数,采用神经网络算法构建了4×16×4×1四层拓扑结构的神经网络模型。进行了神经网络模型的学习训练、预测分析以及未经学习训练样本的验证。结果表明,神经网络模型经过8892次迭代运算后收敛,模型的相对训练误差是3.50%~6.41%,平均相对训练误差是4.76%;相对预测误差是4.25%~5.56%,平均相对预测误差4.88%。神经网络模型预测能力较强,预测精度较高,可以用于汽车用高强铝合金铸造性能优化预测。In order to optimize the casting performance of high-strength aluminum alloy for automobile, taking the alloy element, element content, melting temperature and pouring temperature as the input layer parameters, the fluidity of the alloy as the output layer parameter, and selecting the Purelin function, Tansig function and Trainlm function as the output layer transfer function, implicit layer transfer function and training function,the 4 ×16 ×4 ×1 four-layer neural network model was constructed by using the neural network algorithm. Learning training, prediction analysis, and validation of untrained samples of neural network models were performed. The results show that the neural network model converges after 8892 iterations,and the relative training error of the model is 3.50%-6.41%, and the average relative training error is 4.76%. The relative prediction error is 4.25%-5.56%, and the average relative prediction error is 4.88%. The neural network model has strong prediction ability and high prediction accuracy, which can be used to optimize and predict the casting properties of high strength aluminum alloy for automobile.

关 键 词：铸造性能 流动性 神经网络算法 高强铝合金 优化模型 

分 类 号：TG292[金属学及工艺—铸造]