ANN-enhanced determination and numerical model integration of activation energy and Zener–Hollomon parameter to evaluate microstructure evolution of AA6082 wheel forging

This study presents the integration of two Arrhenius-constitutive model parameters, activation energy (Q) and the Zener–Hollomon parameter (Z), into a numerical model to evaluate their correlation with the microstructural evolution of AA6082 wheel forging. Isothermal tests powered by a Gleeble machine were conducted to establish the constitutive model of AA6082 material, with deformation temperatures and strain rates varying between 350–560 °C and 0.05–15 s⁻¹, respectively. Two types of Arrhenius methods were employed: strain-compensated Arrhenius and artificial neural network (ANN)-enhanced Arrhenius. The key difference between the two methods is that the former ignores the effects of deformation temperature and strain rate when determining the activation energy (Q) value, while the latter considers these factors. Integrating activation energy and Zener–Hollomon parameters into a numerical model by directly inputting the mathematical equation from the strain-compensated Arrhenius method resulted in significant overfitting at certain nodes and elements. To address this issue, a new approach using trilinear interpolation and behavior-based clamping methods on Q values generated by the ANN–Arrhenius method proved effective. Additionally, the ANN–Arrhenius method demonstrated superior accuracy, reducing the prediction’s average absolute relative error (AARE) from 3.14% (strain-compensated Arrhenius method) to 1.10%. A comparative study of the distribution of Q and Z values in numerical model simulations, alongside average grain size and shape examined with an optical microscope, revealed that the Q and Z parameters are beneficial for predicting grain characteristics in final workpieces. This study aims to bridge the gap in implementing activation energy and Zener–Hollomon parameters in more realistic forging scenarios and with more complex workpiece designs.