摘要
以始锻温度、终锻温度、模具预热温度和变形速度作为输入层神经元,并以力学性能(抗拉强度)作为输出层参数,构建了4×24×1三层拓扑结构的汽车用新型镁合金锻压工艺的神经网络模型,并对该模型进行了预测和验证。结果表明,采用traingd函数、traingda函数、trainlm函数的镁合金锻压工艺的神经网络模型的平均相对训练误差分别为4.51%、4.58%、3.65%。对于选择trainlm函数的模型进行了样本验证,结果是平均相对预测误差为4.02%。该模型预测能力强、预测精度高。AZ80-0.3Ti镁合金的锻压工艺优选为:始锻温度390℃,终锻温度280℃,模具预热温度260℃,变形速度6.5 mm/s。
Taking the initial forging temperature, final forging temperature, die preheating temperature and deformation speed as the input layer neurons, and taking the mechanical properties(tensile strength) as the output layer parameters, the neural network model of a new magnesium alloy forging process with 4 ×24 ×1 three-layer topology was constructed. The model was predicted and verified. The results show that the average relative training error of the neural network model for the magnesium alloy forging process using the tradingd function, the tradingda function and the trainlm function is 4.51%, 4.58%and 3.65%, respectively. The sample verifi cation was carried out for the model using trainlm function. The result is that the average relative prediction error is 4.02%. The model has strong prediction ability and high prediction accuracy. The forging process of AZ80-0.3 Ti magnesium alloy is the initial forging temperature of 390℃, the final forging temperature of 280℃, the die preheating temperature of 260℃ and the deformation speed of 6.5 mm/s.
作者
田云霞
高燕
林海霞
TIAN Yunxia;GAO Yan;LIN Haixia(School of Information Technology,Hebei Polytechnic Institute,Shijiazhuang 050091,China;Shijiazhuang Vocational College of Technology&Information,Shijiazhuang 050091,China)
出处
《热加工工艺》
北大核心
2021年第3期92-94,99,共4页
Hot Working Technology
关键词
神经网络
新型镁合金
锻压工艺
传递函数
力学性能
neural network
new magnesium alloy
forging process
transfer function
mechanical properties
