Large Eddy Simulation of Turbulent Incompressible Flow Using GPU with Multigrid Algorithm

Shaikh, M (2013) Large Eddy Simulation of Turbulent Incompressible Flow Using GPU with Multigrid Algorithm. Masters thesis, Indian Institute of Technology Hyderabad.

Preview

Text
ME11M1016.pdf - Submitted Version
Download (917kB)

Abstract

The research presented in this dissertation is divided into two parts,the first part focuses purely on computational science, and the second part focuses on fundamental fluid dynamics. The computational science aspect involves implementing an incompressible Navier-Stokes solver on Graphics Processing Units (GPUs), with optimized algorithm like multigrid algorithm to achieve the goal of enhancing performance of existing computational facilities and enabling them to solve more complex and computationally expensive fluid flow problems. The fluid dynamics aspect involves using the GPU-based Navier-Stokes solver to study turbulent flows. Large Eddy Simulation (LES) turbulence model has been implemented successfully to analyze 2d incompressible flow. Computational Fluid Dynamics (CFD) simulations can be very computationally expensive, especially for Large Eddy Simulations (LES). In LES the large, energy containing eddies are resolved by the computational mesh, but the smaller (sub-grid) scales are modeled. Clusters of CPUs have been the standard approach for such simulations, but an emerging approach is the use of Graphics Processing Units (GPUs), which deliver impressive computing performance compared to CPUs. Recently there has been great interest in the scientic computing community to use GPUs for general-purpose computation (such as the numerical solution of PDEs) rather than graphics rendering. To explore the use of GPUs for CFD simulations, an incompressible Navier-Stokes solver was developed for a GPU. The Navier-Stokes equation is solved via a SMAC algorithm and is spatially discretized using the finite volume method on a rectangular collocated grid. Validation of the GPU-based solver was performed for fundamental bench-mark problems, and a performance assessment indicated that the solver was over an order-of-magnitude faster compared to a CPU. This solver was later extended to perform an LES of turbulent flows. LES has been known to be sensitive to inlet boundary conditions, the effect of different inlet boundary conditions has been observed and summarized for a mixing layer problem.

[error in script]

IITH Creators: