A data-driven machine learning approach to predict the hardenability curve of boron steels and assist alloy design

摘要

Boron steel is one of the most valuable lightweight steels for automobile due to its high strength after hot stamping and low cost. In order to ensure service performance of automobile parts, the steel is required to have good hardenability. A novel data-driven machine learning (ML) model has been established by using relevant material descriptors, including chemical composition and distance along the Jominy bar, to predict the hardenability curve of boron steel. By evaluating and comparing prediction results of different ML methods on training and test sets, random forest is found to be the optimal model with high correlation coefficient and low error. Moreover, the ML model performs better than JMatPro and empirical formula in terms of prediction accuracy and variation trend of hardenability curve. The optimal ML model combined with orthogonal design is employed to successfully design a press-hardening steel with good hardenability, i.e., 0.04 V-added boron steel. Therefore, this study demonstrates that ML can predict accurately and efficiently the hardenability curve of boron steel and guide the material design and heat treatment process of advanced boron steel. GRAPHICS .