Locally mesh-refined lattice Boltzmann method for fuel debris air cooling analysis on GPU supercomputer

Abstract

A dry method is one of practical methods for decommissioning the TEPCO's Fukushima Daiichi nuclear power station. Japan Atomic Energy Agency (JAEA) has been evaluating the air cooling performance of the fuel debris by using the JUPITER code based on an incompressible fluid model and the CityLBM code based on the lattice Boltzmann method (LBM). However, these codes were based on a uniform Cartesian grid system, and required large computational time and cost to capture complicated debris structures and multi-scale flows at the actual reactor scale. The adaptive mesh refinement (AMR) method is one of the key techniques to accelerate multi-scale simulations. We develop an AMR version of the CityLBM code on GPU based supercomputers and apply it to thermal-hydrodynamics problems. The proposed method is validated against free convective heat transfer experiments at JAEA. Thanks to the AMR method, grid resolution is optimized near the walls where velocity and temperature gradients are large, and the temperature distribution agrees with the experimental data using half the number of grid points. It is also shown that the AMR based CityLBM code on 4 NVIDIA TESLA V100 GPUs gives 6.7x speedup of the time to solution compared with the JUPITER code on 36 Intel Xeon E5-2680v3 CPUs. The results show that the AMR based LBM is promising for accelerating extreme scale thermal convective simulations.

Keywords

• natural convection
• air cooling
• lattice boltzmann method
• adaptive mesh refinement
• high performance computing
• gpu