Machine learning for hot deformation flow stress prediction and dynamic recrystallization in Fe24.75Ni19.8Co9.9Cr14.85Al10.9Mn14.85Si4.95 high-entropy alloy

Abstract

In this study, a machine learning model based on a traditional support vector machine (BKA-SVR), optimized by the Black Kite Algorithm, was established for the first time and used to predict the flow stress value. Compared to the traditional SCAT constitutive model, the BKA-SVR model provides a more accurate flow stress prediction. The mean square correlation coefficient is 0.99799, the average absolute error is 5.4826, and the average absolute relative error is 12.02 %. In the context of small sample predictions, the BKA-SVR model still demonstrates high accuracy in flow stress prediction: R² is 0.99527, MAE is 6.0667, and MAPE is 12.76 %. According to the hot processing map that utilizes different stability criteria, the prediction based on Murty's instability criterion is found to be more applicable. Through a coupled analysis of energy dissipation and the instability criterion, the optimal hot working process was determined to be at a deformation temperature of 1100 °C and a strain rate of 0.001 s⁻¹. Under this process parameter, the energy dissipation efficiency (η) is approximately 56 %, and there is no risk of instability. Additionally, a microscopic analysis of different regions with varying η values reveals that as the η value increases, the degree of dynamic recrystallization (DDRX) gradually increases.