Abstract Summary (Max 250 words)
This work presents a GPU-accelerated Lattice Boltzmann Method (LBM) coupled with Large Eddy Simulation (LES) to enable high-fidelity analysis of turbulent mixing and liquid-liquid dispersion in stirred tanks. Integrating the Immersed Boundary Method (IBM) for complex geometry modeling, the solver achieves a 1500-fold speedup over 16 CPU cores, supporting efficient simulation of ~10⁷ droplets and up to 1.38×10⁷ grids. A probabilistic droplet breakage model is proposed to eliminate Lagrangian time step dependence, ensuring accurate predictions of Sauter mean diameter (d₃₂) and droplet size distribution (DSD)—critical for optimizing mixing efficiency. The framework resolves >95% of turbulent kinetic energy in high-turbulence regions, capturing key mixing characteristics: increasing impeller speed enhances droplet trajectory tortuosity, strengthens radial/axial circulations, and shifts peak droplet velocity toward the tank wall. This GPU-accelerated LBM approach provides reliable insights into mixing hydrodynamics, offering a practical tool for optimizing stirred tank design in chemical, pharmaceutical, and process engineering applications.
