The 12th International Conference on Scour and Erosion (ICSE 2025)

In recent years, Computational Fluid Dynamics (CFD), used to model the fluid phase, coupled with the Discrete Element Method (DEM), employed to simulate individual particles, has become increasingly popular for the numerical investigation of sediment transport and local scour around hydraulic structures such as abutments. A key factor dominating the simulation accuracy in this coupled approach is the drag model, which governs the momentum exchange between the fluid and particle phases. Conventional drag models typically rely on ensemble-averaged flow intensity (characterized by Reynolds number) and solid volume fraction, while neglecting variations in drag forces among individual particles due to local differences in particle distribution. To enable more accurate modeling of the drag force on individual particles, an XGBoost-based drag model has been developed using training data from Particle Resolved Simulations (PRS) of flow through particle suspensions. The optimized XGBoost model was integrated into unresolved CFD-DEM simulations of two Geldart-D particle-fluid systems. Results show improved agreement with experimental measurements, particularly at high Reynolds numbers, highlighting the model's superior capability to capture complex particle-fluid interactions compared to traditional correlations.

XGBoost-based drag model,mean-normalized drag force,CFD-DEM,sediment transport,particle-fluid interaction