Strength prediction and microstructure quantification of Ti-6Al-4V alloy by selective laser melting based on machine learning

• First integrated framework connects manufacturing parameters mechanical properties and microstructure for 3D printed titanium alloy • Advanced machine learning model predicts material strength with exceptional accuracy using process data. • Combined deep learning approach achieves high precision analysis of material microstructures automatically. The process parameters, microstructure, and mechanical properties of Ti-6Al-4V alloy fabricated by selective laser melting (SLM) are strongly coupled, limiting the effectiveness of conventional trial-and-error optimization. This study proposes a cross-scale framework integrating machine learning and deep learning to establish quantitative process–structure–property relationships. Based on 750 process parameter sets and 1,225 metallographic images, a multi-scale prediction, analysis, and optimization model is developed. A Bayesian-optimized XGBoost model achieves high-accuracy tensile strength prediction (R 2 = 0.98, MAE = 8.50 MPa), with Shapley Additive Explanations(SHAP) analysis identifying annealing temperature and laser power as the dominant factors. The Broyden-Fletcher-Goldfarb-Shanno(BFGS) algorithm identifies parameters yielding a theoretical maximum tensile strength of 1273.17 MPa. In parallel, deep learning models enable automated microstructure analysis: an enhanced ResNet50 achieves 98% accuracy in classifying lamellar, bimodal, and acicular microstructures, while U-Net and DeepLab V3+ models provide high-precision segmentation of lamellar and bimodal structures, enabling quantitative extraction of α-lamellar and primary equiaxed α phase fractions and characteristic sizes. This framework provides a data-driven pathway for linking SLM process parameters, microstructure evolution, and mechanical performance, highlighting the potential of artificial intelligence in metal additive manufacturing.