GPU-accelerated CFD-DEM framework for large-scale gas-solid flows with superquadric particles

Gas–solid flows in industrial applications often involve a large number of non-spherical particles with complex interactions, but their high computational cost hinders the extensive use of particle-scale simulations. This study presents, for the first time, a Graphics Processing Unit (GPU)-accelerated Computational Fluid Dynamics - Discrete Element Method model (CFD–DEM) framework for high-speed modelling gas–solid flows involving non-spherical particles represented by superquadric (termed GPU-SQ-CFD-DEM). The framework fully leverages GPU parallelisation for DEM and coupling calculations, while CFD calculations are parallelised on the Central Processing Unit (CPU). Advanced strategies are incorporated to enhance computational efficiency, as well as coupling stability and accuracy. The developed model is validated against the measurement of a small-scale fluidized bed with particles of different shapes, in terms of gas pressure drop, distribution of particle height and orientation. The results show good agreement with measurement data and demonstrate a speedup of more than threefold compared to 200 CPU cores. The model is further applied to a large-scale fluidized bed to investigate mixing phenomena across different particle shapes, demonstrating high computational performance in handling superquadric particles. The simulation results reveal that spherical and ellipsoidal particles exhibit higher pressure drops, while cylindrical and cuboidal particles show lower values, with the maximum gap around 140 Pa. In terms of mixing efficiency, spheres perform best, reaching full mixing in 9.6 s, followed by cylinders at 12.3 s, cuboids at 13.0 s, and ellipsoids at 20.6 s. Besides, simulating superquadric particles with high sharpness (cuboid and cylinder) results in higher computational cost, whereas ellipsoidal shapes are relatively more efficient, achieving speeds about four times faster. This work provides a robust generic solution for simulating large-scale gas–solid flows involving non-spherical particle systems. • A high speed superquadric CFD-DEM model was developed to simulate the non-spherical large-scale gas-solid flow. • The advanced coupling strategies designed for GPU are applied to enhance the efficiency and numerical stability. • The model presents significant improvement of computation efficiency compared CPU based model. • Mixing phenomenon for different shapes particle are comprehensively studied.