Data-Driven Optimization of Direct-Chill Casting Parameters for Defect Control in Al7xxx Alloy Billets.

High-strength Al7xxx alloys are widely used in aerospace structural applications due to their superior strength-to-weight ratio. However, the direct-chill (DC) casting of large-diameter billets for aerospace extrusion often results in solidification defects such as hot tearing, porosity, and ghost lines. These defects originate from complex thermo-mechanical interactions during solidification and are strongly influenced by casting parameters including casting speed, cooling intensity, and heat transfer conditions. Effective control of these parameters is therefore essential for achieving defect-free billets and stable downstream processing.

This study proposes a data-driven optimization framework for defect control in Al7xxx alloy billets during direct-chill casting. The target system is an 11-inch DC-cast Al7xxx billet, where defect formation is governed by local solidification behavior and segregation characteristics across the billet cross-section. Solidification simulations were performed using the THERCAST software to predict temperature evolution, solid fraction distribution, and defect-related indicators under various casting conditions. A systematic design-of-experiments (DOE) strategy was implemented to generate a large simulation database by varying key process parameters including casting speed, cooling rate, and mold heat transfer conditions.

The resulting database was used to train machine-learning surrogate models based on XGBoost to predict defect susceptibility across the billet. Bayesian optimization was subsequently applied to identify optimal combinations of casting parameters that minimize hot tearing, porosity, and ghost line formation while maintaining stable solidification behavior. To improve model interpretability, SHAP (SHapley Additive exPlanations) analysis was employed to quantify the relative importance of process variables and reveal the influence of casting parameters on defect formation mechanisms.

The proposed framework demonstrates an effective digital manufacturing approach for the optimization of DC casting processes in large aerospace aluminum billets. The integration of process simulation, machine learning, and explainable AI provides new insights into defect formation and enables data-driven optimization of casting parameters for high-strength aluminum alloys.