Comparison of Five Different Models Predicting the Hot Deformation Behavior of EA4T Steel

The modified Zerilli–Armstrong (mZ-A) model, optimized Zerilli–Armstrong (oZ-A) model, modified Johnson–Cook model, double multiple nonlinear regression (DMNR) model, and BP artificial neural network (BPANN) model were all used to compare the hot deformation behavior of EA4T steel in this paper. The Gleeble-3800 simulator was used to conduct thermal compression tests. The deformation temperature range was 1243~1443 K, the strain rate range was 0.01~1/s and the true strain was 0.8. The obtained stress–strain experimental data were used to calculate the material constants for the five constitutive models, and the established constitutive model was then thoroughly evaluated using the correlation coefficient (R), average absolute error (AARE), root-mean-square error (RMSE), and relative error statistical results (RESR). The experimental results showed that the R value of the mZ-A model is 0.9880, the AARE value is 6.6550%, the RMSE value is 5.8777 MPa, the variation range of the RESR value is − 40.3496 ~ 21.7640%, and the average value is 1.3098%. The mZ-A model cannot adequately depict the high temperature flow behavior of EA4T steel when compared to the other four models. With an R value of 0.9996, an AARE value of 1.1630%, an RMSE value of 1.0358 MPa, a variation range of 12.9101 to 10.3263%, and an average value of 0.0070%, the trained BPANN model has the best prediction performance. The other three models' predictions are in good agreement with the experimental results. The oZ-A model, however, can more accurately follow the deformation behavior of EA4T steel at high temperatures than the other two models. Therefore, when the physical situation of a material response needs to be known, the oZ-A model can be used.