Multi-GPU Based Lattice Boltzmann Flow Simulations in Porous Media

Date

2018-12

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Characterization of rock and fluid properties is vital in producing oil and gas from reservoirs in an economically viable fashion. Digital rock physics for the estimation of petrophysical properties has become a powerful tool that greatly compliments lab experiments by combining advanced imaging techniques with numerical simulations. The lattice Boltzmann (LB) method is a well-applied numerical method to simulate fluid flows in porous media at multiple length scales. The LB simulation is typically resource intensive due to its computational complexity and hence faces great numerical challenges in extremely large-scale computation. In this dissertation, I propose a multi-GPU solution to the LB method on a hybrid high performance computing cluster to perform large-scale single-phase and two-phase fluid flow simulations in reconstructed digital rocks. The program provides solutions at both pore scale and representative elementary volume (REV) scale based on the resolution of digital rock images. Petrophysical properties such as absolute permeability and relative permeability at multiscale are evaluated from the numerical results. The proposed parallel implementation of the high performing LB method on multi-GPU features both strong and weak scalability in both single-phase and multiphase multicomponent (MCMP) LB models. When running on a cluster of 32 GPUs, the multi-GPU pore-scale LB method implementation achieves a speedup of 1074x comparing to the in-house sequential program when simulating single-phase fluid flows and the scalability is approximately 0.7. For the multi-GPU MCMP LB method, the scalability is close to 0.9, which shows a good scalability of the parallel program. With the work proposed, we can obtain the absolute permeability of digital rock samples of sandstone, carbonate and shale formations by using single-phase LB simulations at both pore scale and REV scale. In addition, the MCMP LB method implementation can simulate transport phenomenon of the binary system, wetting phase and nonwetting phase, in porous media. The relative permeability curves then are characterized from simulation results.

Description

Keywords

GPU, LBM, Digital rock physics, CPU-GPU hybrid cluster, MCMP LBM

Citation