@Article{cmc.2026.086653,
AUTHOR = {Gi-Hoon Kwon, Byoungho Choi, Kyong Jun An, Hyosoo Jeon, Kyoung Il Moon},
TITLE = {A Physics-Guided Machine Learning Framework Linking Carbon Concentration, Phase Constitution, and Wear Behavior in Carburized SCR420 Steel},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/cmc/online/detail/28366},
ISSN = {1546-2226},
ABSTRACT = {A physics-guided machine-learning framework was developed to investigate the relationships among carbon concentration, phase constitution, hardness, and wear behavior in carburized SCR420 steel. Surface carbon concentrations of 0.5–0.9 wt.%C were produced by vacuum carburizing, followed by hardness, XRD, friction, and wear measurements. Experimentally measured descriptors were integrated with physics-based descriptors derived from JMatPro thermodynamic and phase-transformation simulations to predict wear behavior. JMatPro successfully reproduced the overall experimental trends in hardness and phase evolution, showing good agreement with the experimental measurements. Among the investigated models, CatBoost achieved the highest predictive performance (R<sup>2</sup> = 0.990, RMSE = 1.163 × 10<sup>−6</sup> mm<sup>3</sup>/N·m). SHAP analysis identified friction coefficient, hardness, and phase constitution as the most important contributors to the CatBoost model output. An optimal surface carbon concentration range of 0.79–0.85 wt.%C was identified for improved wear resistance. The proposed framework demonstrates the potential of integrating physics-based simulations with explainable machine learning for data-driven optimization of carburizing processes.},
DOI = {10.32604/cmc.2026.086653}
}