Constructing processing map of M50NiL steel by artificial neural network model

Abstract

The processing map for M50NiL steel was established by hot compression tests at temperatures of 950–1150 °C and strain rates of 0.002–1.0 s−1. Based on the experimental results of hot compression tests, the predictability in both reproducing experimental flow stresses and predicting flow stresses using the Arrhenius, physical-based, and artificial neural network (ANN) models was compared. The results showed that the average absolute relative errors of Arrhenius, physical-based, and ANN models in both reproducing and predicting flow stresses were 6.04 % and 8.01 %, 6.61 % and 7.78 %, and 1.91 % and 4.74 %, respectively. The ANN model had a considerably higher accuracy in reproducing and predicting flow stresses than the other two models. In addition, a processing map of M50NiL steel was established using the predicted flow stresses by the ANN model. This processing map indicated that the optimized processing parameters were 975–1050 °C/0.01–0.002 s−1. Instability occurred during deformation at 950–975 °C at 1.0 s−1 and 1075–1150 °C at 0.01 s−1. The instability prediction was verified by the microstructure evolution.