Prediction flow behavior of Mg-12Gd-3Y-0.6Zr alloy during hot deformation based on Arrhenius and BP-ANN models: a comparative study

In this study, a Zwick 1484 machine was employed to conduct hot compression experiments on a Mg-12Gd-3Y-0.6Zr alloy (GW123 magnesium alloy) at deformation temperatures ranging from 360 to 480 °C and strain rates from 10⁻³ to 10 s⁻¹. A dataset comprising deformation temperature, strain, strain rate, and stress was built based on these experimental conditions, and the strain-compensated Arrhenius model and the backpropagation artificial neural network (BP-ANN) model were developed to investigate the hot deformation behavior of the GW123 magnesium alloy. The results show that the BP-ANN model has a correlation coefficient of 0.9986 and an average absolute error of 1.53, whereas the Arrhenius model has a correlation coefficient of 0.9729 and an average absolute error of 13. The reason for this result is that the BP-ANN model, by leveraging more data points (1500 in total) to optimize its adaptive parameters, can effectively capture complex nonlinear relationships between different deformation conditions simultaneously, thus providing superior accuracy in predicting the true stress–strain curve responses compared with the Arrhenius model. Furthermore, the optimal processing parameters for the GW123 magnesium alloy were determined through hot processing maps to be within the temperature range of 440–480 °C and strain rates of 10⁻³ s⁻¹ and 10⁻² s⁻¹.