Direct Numerical Simulation of Turbulent Channel Flow

Several Direct Numerical Simulations (DNS) of a fully developed turbulent channel flow are performed. The aim of this thesis is to analyze how grid resolutions and domain sizes affect the turbulence statistics. The simulations have been divided into four cases with a variable number of simulations within each case: Coarse-grid, Short-domain, Narrow-domain and Large-domain. The simulations, in each case, are compared to both each other and to the results obtained by Kim et al. The latter is used as a reference and is considered very accurate. The friction Reynolds number of the flow, which is based on the channel height, is 360. The Coarse-grid case consists of three simulations with grid resolutions of 24x24x96, 48x24x96 and 48x48x192, in x, y and z respectively. The domain size is kept constant at lx=1.5, ly=0.75 and lz=1, where lx, ly and lz is the channel length, width and height, respectively. The results obtained in this case indicates that the grid resolution of 24x24x96 is too coarse and that the domain size is too small. The results are therefore unreliable. For the Short-domain case the channel length is doubled twice to lx=3 and lx=6 and the grid resolution in the streamwise direction is increased accordingly to keep the resolution constant. Domain size effect is the expected outcome for this case. The results show expected data for the large scale motion compared to Kim et al. The small scale motion show a grid effect between the simulations of length lx=1.5 and lx=3. Between the channels of length lx=3 and lx=6 the discrepancy is less. The two longest channels show satisfying results, even though the two-point correlation deviates from zero. In the Narrow-domain case the channel width is doubled two times, while keeping the length constant. The results are satisfying for the large scale motion. The small scale motion shows smaller discrepancies compared to each other and bigger discrepancies compared to Kim et al.[1] than for the Short-domain case. This suggests that there is a small or non-existent grid effect. The two-point correlation shows that only the widest channel is adequate. The Large-domain case consists of a simulation with grid resolution of 192x192x192 and domain size lx=6 and ly=3. The turbulence statistics correspond well with Kim et al.[1] and the two-point correlation shows that the domain size is adequate in both directions. The discrepancies are small compared to the Short-domain case, which indicates that the simulations in Short-domain case yields satisfying results. This suggests that increasing the channel length will yield better results and have greater impact than increasing the channel width