神经网络在热轧铬镍钢变形阻力预测的应用  

Study on the applications of neural networks for predicting the flow stress during hot rolling of nickel-chrome steel

作　　者：杜勇[1] 孟令启[1] 

机构地区：[1]郑州大学机械工程学院,郑州450001

出　　处：《塑性工程学报》2010年第1期129-133,共5页Journal of Plasticity Engineering

基　　金：国家自然科学基金资助项目(50175031)

摘　　要：以凸轮式高速形变试验机进行的热模拟压缩实验得到的实验数据为基础,建立了热轧铬镍钢的变形阻力与其化学成分、变形温度、变形速度及变形程度对应关系的BP、Elman、RBF和GRNN 4种神经网络预测模型;并对4种网络的训练过程和预测精度进行比较分析。结果表明,神经网络有很强的预测能力;4种神经网络相比较,RBF网络具有更高的预测能力和好的泛化能力。On the basis of the data obtained from the cam-type plastometer, four prediction models including the BP, Elman, RBF and GRNN neural network prediction models were established, which corresponded to the relationship between the deformation stress of nickel-chrome steel with chemistry elements, temperature, strain rate and deformation strain of the steel respectively. A comparative analysis for the prediction accuracy of the four models was obtained. The results show that neural network is capable for prediction, and RBF possess higher capability in prediction and better adaptability in comparing with other three neural networks.

关 键 词：人工神经网络 变形阻力 热轧:预测 泛化 

分 类 号：TF3[冶金工程—冶金机械及自动化]