GPU-accelerated CFD–DEM coupling strategy for large-scale solid–liquid–gas flows in moving mesh systems, with application to wet grinding in an autogenous mill

Large-scale CFD–DEM simulations are often constrained by high computational costs, particularly for multiphase flows with moving mesh systems, which restricts their application to full-scale industrial systems. In this work, a fully GPU-based coupling strategy was developed within a CPU–GPU hybrid CFD–VOF–DEM framework. This strategy incorporated a bounding-box method for efficient dynamic cell mapping, achieving an 800 × speedup in particle-to-cell identification, enabling rapid retrieval of flow-field information and exchange momentum and volume fractions within moving mesh systems, supporting large-scale multiphase simulation. To demonstrate its practical utility, the model was validated and applied to simulate the wet grinding in a full-scale autogenous grinding (AG) mill and examine particle size effects. For the wet grinding case involving 10 million particles, a 50-second double-precision simulation was completed in 28 h, with coupling interface accounting for <2% of total runtime. The slurry developed a distinct kidney-shaped flow structure, mainly governed by particle motion, with dynamics intensifying as particle size increased. Larger particles experienced less slurry resistance, resulting in larger cascading zones, higher shoulder heights, increased power draw, and higher breakage probability.