Machine-learning interatomic potentials enabled multi-scale modeling of rapid solidification in Al-Si alloys: From atomistic simulations to process-level stress regulation

YH Liu and Z Li and ZG Song, CHEMICAL ENGINEERING JOURNAL, 525, 170640 (2025).

DOI: 10.1016/j.cej.2025.170640

Laser additive manufacturing enables the fast fabrication of lightweight, geometrically complex components, while always suffers from residual stress accumulation and microstructural instability. To deal with it, a multiscale analysis framework, empowered by machine-learning interatomic potentials (MLIPs), is developed by integrating molecular dynamics (MD) and finite element analysis (FEA). Taking the near- eutectic Al-12Si alloy as a test bed, the MLIP is generated based on a high-fidelity density functional theory (DFT) dataset, which can accurately capture solidification kinetics and complex many-body interactions under rapid cooling conditions, revealing key mechanisms such as temperature-gradient-induced directional solidification and silicon segregation. The predicted solidification velocities are in close agreement with experimental measurements, demonstrating the accuracy of the generated MLIP. The results indicate that substrate preheating and reduced scanning speed act synergistically to suppress residual stress. This proposed coupled MD-FEA framework elucidates the interdependence between process parameters, solidification behavior, and stress evolution, offering theoretical guidance for process optimization. Finally, this study provides a systematic strategy for tailoring the process-structure-property relationship in the additive manufacturing of Al-Si alloys.

Return to Publications page