Modeling the Double-Pass Flow Curve of Nb Micro-Alloyed Steel by Machine Learning and its Extrapolation to Static Softening Kinetics

This paper presents a method to predict the double-pass flow stress behavior of hot-rolled niobium (Nb) micro-alloyed steel by combining the constitutive model and back-propagation neural network (BPNN) model (the machine learning (ML) model). The constitutive model is used to predict the first-pass flow stress behavior, and the BPNN model is used to predict the second-pass flow stress behavior. The results show that the ML model can accurately predict the double-pass flow stress behavior at different temperatures and holding time. The second-pass flow stress predicted by the BPNN model is compared with those predicted by the mathematical model in the previous literature. The results show that the BPNN model can more accurately predict the second-pass flow stress than the mathematical model in the previous literature. In addition, the static softening fraction during the interval time is calculated by using the predicted double-pass flow stress curve in this paper. The results show that the precision of the static softening fraction calculated in this paper is not only higher than that of the traditional Avrami kinetic model but also higher than that of the traditional model which couples recovery, precipitation and recrystallization. Therefore, the ML model for predicting the double-pass flow stress behavior has high accuracy and provides a theoretical basis for predicting the double-pass flow stress behavior of other alloys.