Machine learning–driven prediction of mechanical properties in heat treated L-PBF Ti-6Al-4V from microstructural images

• A machine learning framework predicts stress–strain curve of heat-treated L-PBF Ti-6Al-4V. • Quantitative microstructural features extracted from as-built OM images using 2-point correlation and autoencoder. • Microstructure-based prediction accounts for effects beyond process parameters, including geometry and environment. • Machine learning framework enables optimization of manufacturing process without repeated heat treatment or tensile testing. In the L-PBF process of Ti-6Al-4V, rapid solidification leads to the formation of acicular martensitic microstructures, which result in high strength but significantly reduced elongation. To overcome this limitation, heat treatment is applied, and the resulting mechanical properties vary greatly depending on the conditions. In this study, we propose an artificial neural network (ANN)-based model that predicts the entire stress–strain curve, including tensile properties, based on the initial microstructure images of Ti-6Al-4V alloy. Microstructural features were quantitatively extracted using a two-point correlation function and an autoencoder. The developed model reconstructed the full stress–strain curve by predicting the fitting parameters of the Voce equation, thereby capturing plastic behavior. This approach demonstrates that mechanical properties can be predicted using initial microstructure information and heat treatment conditions, underscoring the potential of data-driven models to optimize processes without repeated tensile testing.