Modeling Particle-Fluid Systems with a DEM-Lbm Solver on GPUs

Solid particles are frequently present in mixing and processing equipment. For systems where small particles occupy small volume fractions, a simple model including drag, gravity, buoyancy, and added mass can usually provide accurate results. However, systems with larger particles occupying significant volume fractions have complex particle-particle, particle-wall, and particle-fluid interactions that have important effects on overall physics. Such applications include slurry pipes, fluidized beds, agitated tanks, and granulators.

The discrete element method (DEM) can accurately describe particle-particle interactions and can be combined with more complex drag models to model particle-fluid coupling. One of the main hurdles in applying these models is simulation runtime as numerical integration of large numbers of particles with small timesteps can be required.

Here we implement DEM and couple it to an existing lattice Boltzmann LES CFD code. It targets the latest generation of GPUs which provide significant performance improvements over CPU based algorithms, achieving particle counts in the millions. This speedup widens the applicability of the method to more real industrial systems. We analyze the scaling performance of the algorithms and validate against experimental results.