Project field note
Materials Informatics
Predicting the Vickers microhardness of metal-ceramic nanocomposites with XGBoost and physics-informed features (Hall-Petch, Orowan, CTE mismatch), plus group-aware validation, ensemble uncertainty, and SHAP interpretability.
Research with the Li Group at the University of Pennsylvania on predicting the Vickers microhardness of metal-ceramic nanocomposites from composition and processing descriptors.
The model is an XGBoost regressor built on physics-informed features grounded in materials theory, including Hall-Petch grain-size strengthening, Orowan dispersion strengthening, and coefficient-of-thermal-expansion (CTE) mismatch.
To keep the results trustworthy on small, clustered materials data, it uses group-aware validation to avoid leakage across related samples, ensemble-based uncertainty estimates, and SHAP analysis to interpret which physical features drive each prediction.