45钢的温变形行为及本构模型

Thermal deformation behavior and constitutive model of 45 steel

在变形温度为600~700℃,应变速率为0.001~1 s^(-1)及变形量为60%的条件下,采用Gleeble-3500热模拟试验机对45钢进行了等温压缩试验,研究了其温变形行为。在对45钢真应力-真应变曲线进行摩擦修正及绝热修正的双修正基础上,建立了45钢基于Arrhenius方程的本构方程、基于应变补偿的Arrhenius本构方程和BP(Back-propagation)神经网络的本构模型。结果表明:45钢的流变应力对变形温度和应变速率十分敏感,随变形温度的升高而降低,随应变速率的增大而增加。通过计算本构方程的预测值和试验值的相关系数与平均相对误差发现:基于Arrhenius方程的本构方程,相关系数为0.98537,平均相对误差为3.1197%;基于应变补偿的Arrhenius本构方程,相关系数为0.97923,平均相对误差为4.42948%;基于BP神经网络的本构方程,相关系数为0.99682,平均相对误差为2.28368%,表明基于BP神经网络的本构方程的整体预测性较好,具有较低的误差及较好的预测性。

Under the conditions of deformation temperature of 600-700℃,strain rate of 0.001-1 s-1 and deformation amount of 60%,isothermal compression tests were conducted on 45 steel using Gleeble-3500 thermal simulation testing machine to study its thermal deformation behavior.On the basis of friction correction and adiabatic correction on the true stress-true strain curves of the 45 steel,the constitutive equation based on Arrhenius equation,the Arrhenius constitutive model based on strain compensation and the BP(Back propagation)neural network constitutive model were established for the 45 steel.The results show that the flow stress of the 45 steel is highly sensitive to deformation temperature and strain rate,decreasing with increasing deformation temperature and increasing with increasing strain rate.By calculating the correlation coefficient and the average relative error between the predicted values of the constitutive equation and experimental values,it is found that the correlation coefficient of the constitutive equation based on Arrhenius equation is 0.98537,and the average relative error is 3.1197%;the Arrhenius constitutive equation based on strain compensation has a correlation coefficient of 0.97923 and an average relative error of 4.42948%;the constitutive equation based on BP neural network has a correlation coefficient of 0.99682 and an average relative error of 2.28368%,indicating that the overall predictive performance of the constitutive equation based on BP neural network is good,with lower errors and better predictive performance.