基于人工神经网络的高精度Ti6242s合金热变形本构模型

High-precision constitutive model of Ti6242s alloy hot deformation based on artificial neural network

摘要:

使用Gleeble-3800对锻态Ti6242s钛合金在温度950~1010℃、应变速率0.01~10s-1的条件下进行了75%变形量的热压缩模拟实验。基于实验取得的真应力-真应变曲线，分别使用人工神经网络（ANN）和Arrhenius方程建立Ti6242s合金本构模型，研究其热变形行为。结果表明：流变应力在变形开始后迅速上升至峰值应力，随后硬化与软化达到动态平衡，在真应变达到0.6后加工硬化逐渐占据主导，硬化幅度随应变速率的增大而提高；人工神经网络本构模型预测值的平均相对误差（AARE）为2.5%，相关系数（R）为0.999；Arrhenius方程本构模型预测值的AARE为14.5%，R为0.955，精度在参数范围内波动较大；ANN本构模型精度远高于Arrhenius本构模型，且在整个参数范围内具有一致的精度；ANN本构模型具有良好的泛化能力，在实验参数范围外预测流变应力仍具有较高的精度。

Abstract:

The forged Ti6242s titanium alloy was subjected to a thermal compression simulation experiment with 75% deformation at a temperature of 950~1010℃ and a strain rate of 0.01~10s-1 by Gleeble-3800. Based on the true stress-true strain curve obtained from the experiment, the artificial neural network (ANN) and Arrhenius equation were used to establish the constitutive model of Ti6242s alloy, and study its thermal deformation behavior. The results show that the flow stress rapidly rises to the peak stress after the deformation begins, and then the hardening and softening reach a dynamic balance. After the true strain reaches 0.6, the work hardening gradually dominates, and the hardening amplitude increases with the increase of the strain rate; artificial neural network The average relative error (AARE) of the predicted value of the constitutive model is 2.5%, and the correlation coefficient (R) is 0.999; the AARE of the predicted value of the Arrhenius equation constitutive model is 14.5%, R is 0.955, and the accuracy fluctuates greatly within the parameter range; The accuracy of the ANN constitutive model is much higher than that of the Arrhenius constitutive model, and it has consistent accuracy across the entire parameter range; the ANN constitutive model has good generalization ability, and it still has high accuracy in predicting flow stress outside the range of experimental parameters.