Modelling strain-induced precipitation kinetics of Nb (C, N) by symbolic regression machine learning

Because strain-induced precipitation (SIP) behavior in Nb micro-alloyed steels directly affects recovery and recrystallization processes during hot rolling, it is crucial to establish high-precision models for SIP kinetics. In this work, based on experimental results with Nb micro-alloyed steels, symbolic regression machine learning (SRML) was used to construct the models for precipitation start ( P s ) and finish times ( P f ) with concentrations of Nb, C, N, Mn, and Si in solution, deformation temperature ( T def ), strain ( ε ), strain rate ( ε ˙ ), Zener-Hollomon parameter ( Z ), and the supersaturation ratio ( K s ) being the model inputs. Compared to the Pereda model for P s and Medina model for P f , root mean square errors ( RMSEs ) were reduced by about 89% and 91%, respectively. Furthermore, the effects of Si and Mn contents, strain, and strain rate on the SIP kinetics were analyzed, showing good consistency with the experimental results. The precipitated volume fractions were calculated based on the JMAK equation with the precipitation start times by using different models, which further demonstrates that the SRML model outperforms conventional models. Finally, the static recrystallization critical temperatures (SRCT) were determined for different Nb micro-alloyed steels, which can guide the designs for new thermomechanical processing routes.