Large Eddy Simulation of Fully Developed Turbulent Pipe Flow

Abstract

Fully developed turbulent pipe flow is investigated by large eddy simulations (LES). The three-dimensional, unsteady, incompressible, filtered continuity and Navier-Stokes equations in cylindrical coordinates are discretized by a finite difference method. The spatial derivatives are approximated by second order conservative schemes. This scheme eliminates the numerical generation or dissipation of energy. The pressure Poisson equation is solved by FFT method and time is advanced through a third order Runge-Kutta method. The commonly used subgrid scale (SGS) models — the Smagorinsky model and the dynamic model are implemented and simulations are performed for fully developed turbulent pipe flow at two different Reynolds numbers. The flow features in terms of mean velocity as well as higher order turbulence intensities and correlations are presented and compared to experimental and DNS data available in literature. Extensive comparisons are made for cases using different grid resolution, different streamwise domain dimension, different sub-grid scale model, and, at two different Reynolds number. For two Reynolds numbers (5,000 and 30,000) tested in this study, the fine mesh (64 × 96 × 64, circumferential × radial × longitudinal) produces better results than the coarse mesh (32 × 48 × 32), indicating the significance of the grid resolution, especially near the pipe surface. On the fine mesh for the two Reynolds numbers, the results exhibit a slight Reynolds number effect, indicating the mesh needs to be further refined at higher Reynolds number. Simulations were performed for two domain sizes, namely 6D and 12D, where D is the pipe diameter. When the streamwise grid resolution remains unchanged, the two simulations show negligible difference. This ensures that a 6D domain is adequate to include the largest eddies in a fully developed turbulent pipe flow at the current Reynolds number. When the fine mesh is used, the subgrid scale models (Smagorinsky and Dynamic) provide limited contribution to the total turbulent kinetic energy. Although the current results agree quite well with other published LES simulations, when compared with the Law of the wall, benchmark experiments and DNS results, the simulated mean velocity in the log region is higher than the experimental and DNS data. Overall, it was observed that the numerical methods work satisfactorily well for turbulent pipe flows at low and high Reynolds numbers, and, the method has capability to be used in the simulation of flows with practical interest.