Computational-thermodynamics-based martensite-start temperature models

Abstract

Thermodynamic models for the austenite-to-martensite phase transformation in steels were developed using the CALculation of PHase Diagrams (CALPHAD) modeling method. Previous modeling efforts from early empirical to more modern machine-learning (ML) models are reviewed and compared with the CALPHAD approach. An open-source, multicomponent thermodynamic database for steels was developed and used for the martensite model is made available for public use and collaboration. CALPHAD-based models for lath, plate, and epsilon martensite, including the effects of prior-austenite grain size, were developed using experimental data for binary and ternary iron alloys. A Gaussian process classification ML model was developed to predict the type of martensite that will form given a steel composition and martensite-start temperature ($M_{\text{s}}$) because this information is not always reported with the experimental measurements. The lath and plate models extend previous work using updated thermodynamic assessments and the open-source database while the epsilon model is made available for the first time. The accuracy of each model was also assessed and found to be reasonable compared to the expected experimental error associated with $M_{\text{s}}$ measurements.