Prediction of mechanical property and costs of high-strength steel using machine learning

Mechanical properties of high-strength steels (HSSs) are complex and influenced by steel composition and rolling parameters. Thus it is complicated to achieve the prediction of the yield strength of steel and the cost based on traditional mechanism models. A machine learning model was developed to predict mechanical properties and costs of steel. Compared with other algorithms, the XGBoost algorithm model exhibited the smallest error in current calculations, which was used to predict the yield strength with various steel composition and rolling parameters. Random Forest was used in feature selection regression to evaluate the correlation between input parameters and output yield strength of HSSs by assigning an importance score. The yield strength of HSSs was mainly determined by the basic alloy elements of C, Cr, Ni and Mo, while it was also obviously related to precious elements of Nb, V and Ti. There was a maximum yield strength of steel with 0.08% Nb. Thus the addition of Nb was more beneficial in enhancing the strength of steel than the addition of V and Ti, but the increased yield strength of this HSSs suffers from the highest price Nb. Additionally, the larger the thickness of steel plates, the larger the yield the strength of HSSs. Moreover, it was suggested to increase the Ti addition instead of Nb and V with the target of lowering the production costs of HSSs. The model can be widely used to enhance design efficiency and better understand the relationship between material performances, manufacturing processes and production costs of HSSs.