Exploration of Machine Learning to Predict Hot Ductility of Cast Steel from Chemical Composition and Thermal Conditions

Abstract

This study explores the use of machine learning (ML) as a data-driven approach to estimate hot ductility of cast steel from literature data. Four ML algorithms were used to predict hot ductility by considering elemental composition and thermal conditions. Experimentally-measured reduction of area (RA) values were converted to a low-temperature limit, center-temperature, and high-temperature limit, which were represented as Gaussian curves. The prediction accuracy of the four ML models was evaluated using RMSE for these three output variables. In a case study of three steels that had different contents of alloying elements, only the Neural-net model predicted the RA trough more accurately in all cases. These results demonstrate the utility of ML models to predict hot ductility of cast steels.