基于修正J-C和BP神经网络模型的超细晶纯钛动态本构行为

Dynamic constitutive behavior of ultrafine-grained pure titanium based on modified J-C and BP artificial neural network model

摘要:

为研究超细晶纯钛在高温、高应变速率加载下的复杂力学行为，建立能够准确描述其动态力学行为的模型，对超细晶纯钛在温度为300~450 ℃，应变速率为2000~3000 s-1下进行了动态冲击实验，获得真应力-真应变曲线。结果表明：在所研究的条件下，真应力-真应变曲线均表现为明显的“双应力峰”特征，晶界处的位错湮灭、重新排列及后续绝热剪切带的形成是两次应力减小的主要因素，流动应力均表现出正应变速率敏感性和负温度敏感性。综合考虑应变硬化、应变速率硬化和热软化效应，提出一种修正J-C本构模型和BP人工神经网络模型，并对两种模型进行了准确性分析。结果表明BP人工神经网络模型能够更好地预测超细晶纯钛的动态力学行为，相关系数可达0.97065，平均相对误差仅为4.63%。

Abstract:

To study the intricate mechanical behavior of ultrafine-grained(UFG) pure titanium under high temperature and high strain rate loading, a model that can accurately describe its dynamic mechanical behavior was established. The dynamic impact test of UFG pure titanium was carried out at loading temperatures of 300~450℃ and strain rates of 2000~3000 s-1, the true stress-strain curves were also obtained. The results show that under the studied conditions, the true stress-strain curves show obvious double stress peaks, the annihilation and rearrangement of dislocations at grain boundaries and the subsequent formation of adiabatic shear bands(ASB) are the main factors for the two stress reduction. UFG pure titanium shows positive strain rate sensitivity and negative temperature sensitivity. Considering strain hardening effect, strain rate hardening and thermal softening effect, a modified Johnson-Cook(J-C) constitutive model and a BP artificial neural network(BP-ANN) model are proposed, and the accuracy of the two models is analyzed. The results show that the BP-ANN model can better predict the dynamic mechanical behavior of UFG pure titanium, the correlation coefficient can reach 0.97065, and the average relative error is only 4.63%.