摘要
采用不同的渗钒温度、渗钒时间和渗剂使用次数进行了40Cr钢的表面固体渗钒试验,并以这些实验数据作为训练样本和验证样本,建立了一个四层神经网络系统,对40Cr钢表面同体渗钒试样的性能进行了预测。结果表明.经过训练的四层神经网络,对未学习的40Cr钢表面固体渗钒样本数据具有较好的识别能力和较高的预测精度:硬度的预测值与试验值之间的相对误差在O.31%~2.62%之间;耐磨失重的相对误差在0.35%~2.96%之间:渗层深度的相对误差在0.75%~2.35%之间。
The die steel 40Cr was treated by superficial solid vanadizing under different vanadizing temperature, vanadizing time, consumption fi'equency of vanadizing agent, and a four-layer neural network system was established taking the experimental data as training sample and verification sample. And the performance of the 40Cr steel samples after superficial solid vanadizing were forecast. The results show that the network has better recognition ability and high precision of prediction to the samples, the relative error of hardness is 0.31%-2.62% between predicted value and test value, the relative error of mass-loss of wear resistant is 0.35%-2.96%, the relative error of vanadizing layer depth is 0.75% -2.35%.
出处
《铸造技术》
CAS
北大核心
2013年第9期1151-1153,共3页
Foundry Technology
关键词
神经网络
模具
表面固体渗钒
预测
neural network
die
superficial solid vanadizing
prediction