Artificial intelligence enabled microstructure prediction in Al alloy castings

Gas porosity defects and secondary dendrite arm spacing (SDAS) are the key microstructure influencing the mechanical properties of Al alloys and their predictions are critical for the safety and reliability of automotive casting components. Existing works mainly utilize experimental methods or numerical simulations to characterize the microstructure, which cost highly and offer limited physical insights. In this study, we generated a comprehensive porosity dataset (472 samples) via 3D cellular automata (CA) simulations and curated an SDAS dataset (310 samples) derived from published literature. Seven artificial intelligence (AI) algorithms have been systematically evaluated, and the eXtreme Gradient Boosting (XGBoost) was identified as the most robust model for microstructure prediction. To validate the AI models, X-ray computed tomography (X-CT) and metallographic experiments were conducted, and the results indicated an accuracy exceeding 90%. Beyond prediction accuracy, we employed SHapley Additive exPlanations (SHAP) analysis to elucidate the impact of alloy elements and processing parameters on the microstructure features, bridging the gap between “black-box” AI and physical insights behind.