Artificial neural network modified constitutive descriptions for hot deformation and kinetic models for dynamic recrystallization of novel AZE311 and AZX311 alloys

Abstract

Two novel alloys of AZE311 and AZX311 were designed through adding 1 wt% Gd and Ca to the AZ31 alloy, respectively. To study the hot deformation behavior, compression tests were conducted at temperatures of 200–350 °C and the strain rates of 0.001–0.1/s. The conventional strain-related Arrhenius-type constitutive equations were developed based on the true stress-strain curves. The activation energy for hot deformation of AZE311 and AZX311 were determined about 137.04 kJ/mol and 155.57 kJ/mol, respectively. Remarkable deviations existed between the predicted and the experimental results for the flow curves with obvious stress peaks. A dimensionless correction factor λ was introduced to the conventional constitutive model based on the artificial neural network (ANN) method. After the modification, the average absolute relative error (AARE) of the models decreased significantly from 10.331% to 1.428% for AZE311 and from 8.555% to 0.703% for AZX311. Kinetic models for the dynamic recrystallization (DRX) process were established. The average values of εc/εp of AZE311 and AZX311 were 0.54 and 0.58, respectively. The addition of Gd and Ca promoted the initiation and process of DRX and Gd had a larger effect. DRX occurred at the grain boundaries of AZX311 while more uniformly in AZE311. The actual DRX fraction at large strain was slightly lower than the predicted results. This was because the grain orientation factor on the deformation and the DRX process was not considered in the establishment of the Avrami-type DRX kinetic models. The effect of grain orientations on the DRX process was also discussed.