CALPHAD-informed Machine Learning Models for Materials Property Prediction

Machine learning (ML) has emerged as an effective tool for predicting materials microstructure and behavior. In this work, we present a physics-informed approach that integrates CALPHAD (CALculation of PHAse Diagrams) simulations with ML models to enhance predictive accuracy for critical materials properties. CALPHAD-derived thermodynamic and microstructural features serve as meaningful inputs, capturing complex phase and composition-dependent phenomena. Two case studies demonstrate the effectiveness of this framework: (1) creep rupture life prediction for Ni-based superalloys and (2) hot cracking susceptibility prediction for alloys printed by additive manufacturing (AM). In the first case, the integration of CALPHAD features with alloy composition and loading conditions enables reliable lifetime predictions. In the second, a Hot Cracking Susceptibility Index (HCSI) is developed using Scheil-based simulations combined with ML. Both cases demonstrated the value of CALPHAD-informed ML models in improving predictive performance and establishing a generalizable strategy for ICME model development.