Constructing porosity database for Al−Si alloy castings through 3D cellular automata model and machine learning

• • Increasing cooling rate from 1 to 20 °C/s at 1.5×10−3 mL/g initial hydrogen reduces average equivalent porosity diameter from 51.40 to 19.16 μm, directly enhancing fatigue life in automotive structural components by minimizing stress concentrators. • • The KNN model predicts porosity percentage with R2 = 0.94, RMSE = 0.035, and MAE = 0.022 on the test set, outperforming SVM, RF, and GBM, enabling reliable virtual quality assurance and reducing costly physical trial casts. • • Experimental validation via optical microscopy shows prediction errors within 20% for average equivalent porosity diameter and porosity percentage, meeting industrial tolerance thresholds for safety-critical castings. • • The coupled CA-ML framework generalizes across cooling rates from 0.25 to 50 °C/s, providing a scalable database that can be integrated into casting simulation workflows to cut computational time by orders of magnitude compared to pure physics-based models.