Session: K20-03 Computational Methods

Paper Number: 117092

117092 - Accelerating Dem Code for Heat Transfer Simulations Using Gpu: Challenges and Solutions 

Efficient and accurate simulations of heat transfer in granular materials have become increasingly important
in recent years. One popular approach is to couple the Discrete Element Method (DEM) with Computational Fluid
Dynamics (CFD) to accurately simulate the dynamics of granular materials. However, these simulations can be
computationally expensive due to a large number of particles and the use of Monte Carlo ray tracing for radiation
heat transfer.

This study presents a new method for accelerating DEM and Fluent coupled simulations of radiation and
convection heat transfer in granular materials using GPU acceleration. The study utilized the NVidia CUDA library
to develop a DEM code that is compatible with CUDA GPUs. The dynamic library was built from the DEM
function and linked to a Fluent User Defined Function (UDF). The DEM code was designed to simulate radiation
heat transfer using Monte Carlo ray tracing and conduction between particles. The Fluent UDF was used to
simulate convection heat transfer and the effect of fluid flow on particles. The DEM code was two-way coupled,
and data exchange between the DEM and Fluent was performed at CFD time step intervals. The use of GPU
acceleration allowed for a significant reduction in computational time.

This study discusses the challenges faced in developing a GPU-compatible DEM code using Object-oriented
programming and nested structures for DEM searching grids. The results showed that the GPU-accelerated DEM
code was about 4-6 times faster on NVidia A100 GPU compared to AMD 128 core/256 threads EPYC 2. The
radiation function was 120 times faster for 1000 rays for each particle. The study highlights the potential of GPU
acceleration in improving the computational efficiency of DEM and Fluent coupled simulations.

Presenting Author: Alireza Kianimoqadam University of Maine