基于人工神经网络的超细晶纯钛热变形本构模型研究

A Constitutive Model of Ultrafine Grained Pure Titanium at ElevatedTemperature Based on Artificial Neural Network

摘要:

对等通道转角挤压(ECAP)制备的超细晶纯钛,在温度为250~450 ℃、应变速率为10_-5~1s_-1的条件下进行热压缩实验。基于真应力和真应变实验数据,分别使用人工神经网络(ANN)和Arrhenius方程建立超细晶纯钛的热变形本构模型,研究其热变形行为。实验结果表明：在变形初期,流变应力随应变的增大而升高,随后趋于平缓,最终流变应力达到一个稳定值。人工神经网络训练和预测结果表明：人工神经网络最佳结构为3×12×1,人工神经网络模型预测的平均相对误差(AARE)为2.1%,相关系数(R)为0.9979,Arrhenius方程模型预测的AARE为11.54%,R为0.9464。即人工神经网络模型能够更加精确的描述超细晶纯钛的本构关系。通过对比不同温度下两种模型的误差,人工神经网络模型在高温条件下具有更好的稳定性。

Abstract:

Ultrafine grained (UFG) pure titanium was prepared by ECAP up to four passes. The hot compression tests were conducted in the different temperatures (250~450 ℃) and the strain rates of 10_-5~1s_-1.The artificial neural network (ANN) and Arrhenius constitutive equation were used for establishing constitutiveSmodel of UFG pure titanium, respectively. The experimental results show that the flow stress increased with the increase of strain at the beginning of the deformation, then increased slowly. Finally, the stress reached a stable value. The experimental value and the predicted value of flow stress showed that the average absolute relative errors obtained from the artificial neural network model and Arrhenius constitutive equations were 2.1% and 11.54%, respectively. The correlation coefficient of the artificial neural network model and Arrhenius constitutive equations were 0.9979 and 0.9464, respectively. It means that the artificial neural network model can more accurately describe the constitutive relations of UFG pure titanium. By comparing the error of the two models under different temperatures, it can be find that artificial neural network model has better stability under the condition of high temperature.