GPU accelerated MFiX-DEM simulations of granular and multiphase flows by Dr. Liqiang Lu
MFiX-DEM is a powerful tool capable of directly tracking the collisions and movement of particles in chemical reactors. NETL is using this model for challenging applications like biomass pyrolysis, carbon dioxide capture, and coal gasification. The DEM approach provides accurate and detailed descriptions of multiphase flows, but has a high computational cost. This paper describes a recent NETL-sponsored effort to accelerate the DEM approach by porting the Fortran-based DEM solver to GPU architectures using CUDA/C++ programming. The speedup to serial CPU simulation is ranges from 130 to 243 fold in the modeling of granular flow applications like particle packing. To simulate fluidized beds, the GPU-CUDA-based DEM solver was then coupled with the CPU-Fortran-based CFD solver through inter-process pipes. An overall speedup of 15 fold was achieved in the simulation of a bubbling fluidized bed. Heat transfer and chemical reaction models were then added and the code was used to simulate fluidized bed biomass drying as a demonstration to a problem of interest to our research program. The code will be released as part of the MFiX code base in the near future after further development and validation.