Identification of Strengthening Factors in Hot-Rolled Dual Phase PM Ti-Fe Alloys by a Machine Learning Approach

Due to the complex and nonlinear correlations between microstructures and mechanical properties in dual phase alloys, it is difficult to estimate their strength by conventional methods. This study attempts to model the relationship between microstructure and mechanical properties in sintered and hot-rolled α+β titanium-iron (Ti-Fe) alloys using machine learning. In the preparation of the models, 4-9 microstructural factors were investigated to identify the most important predictors of mechanical properties. A Random Forest (RF) model was found to have best predictive power, producing a good match with experimental data in samples which were outside of the training dataset. Moreover, the average α grain diameters, β phase widths (by intercept method), and β phase area fractions were found to be the strongest predictors of mechanical behavior.