Integrating Design of Simulation Experiments and Machine Learning to Predict the Local Secondary Dendrite Arm Spacing for Recycled AlSi7 Cast Parts

Casting simulations are essential for designing cast parts together with their foundry equipment, as they allow prediction of nonuniform mechanical properties resulting from solidification dynamics. However, when using secondary alloys, accounting for all potential composition variations becomes impractical due to excessive simulation times required. This study investigates the relationship between alloying composition variations and mechanical properties in secondary AlSi7 casting alloys using a database of simulations and a Machine Learning (ML) approach. Secondary Dendrite Arm Spacing (SDAS) is adopted as representative of mechanical properties due to its strong correlation with strength and ductility in cast aluminum parts. The Design of Simulation Experiments (DOSE) uses a spherical casting geometry with varying diameters to capture local phenomena across different solidification times. The alloy compositions in the DOSE reflect the natural variability characteristic of secondary AlSi7 feedstock. The resulting dataset captures the influence of alloying elements on SDAS across diverse thermal conditions. A ML model was trained on this data to predict SDAS based solely on alloy composition, thereby avoiding the need for extensive additional simulations. This approach enables robust design of foundry equipment involving recycled aluminum, where the ability to optimize performance despite composition fluctuations is essential for maintaining product quality.