(320c) Large Eddy Simulation of a Turbulent Incompressible Wake Flow - Comparison with Particle Image Velocimetry Data

Large Eddy Simulation (LES) is becoming a more and more popular tool for prediction of turbulent flows both in industrial applications and for research purposes. In many situations the averaged flow fields from Reynolds Averaged Navier-Stokes (RANS) simulations do not provide sufficient information; in cases where unsteadiness and turbulent structures are important, RANS simulations fall short. While Direct Numerical Simulation (DNS) can give very detailed information about the flow, it is computationally expensive and thereby limited to low Reynolds number flows and simple geometries. LES, which can provide detailed information down to the smallest resolved case, is at the same time applicable to a much wider range of flows. However, LES involves modeling and validation of the method is therefore required.

In this work Large Eddy Simulation (LES) is employed to study turbulence characteristics in an incompressible wake flow. Experimental data from Particle Image Velocimetry (PIV) of the same flow is used to validate the simulation results.

The flow domain is a rectangular channel of cross section 60 mm by 100 mm. The channel features three inlet streams separated by a pair of splitter plates. Using equal flow rate for all three inlet streams, a wake is produced downstream of each splitter plate. The Reynolds number in this study is 37 500 based on the channels hydraulic diameter. This flow setup has the advantage that is resembles flow configurations found in industrial applications. A further advantage is that this configuration produces a complex flow well suited for evaluation of LES beyond the frame of simplified reference flows traditionally used for validation.

For the LES, the effect of time step, grid resolution, turbulent inlet conditions and subgrid model is investigated through a series of preliminary simulations. The inlet condition proved to be very influential to the results over most of the domain while the subgrid models that were tried gave only small differences in the simulation results.

For both simulation and experiment large ensembles (10 000) of instantaneous velocity fields from the channels center plane is collected. A thorough comparison between experiment and simulation data including mean and root-mean-square values of velocity, shear stress and energy dissipation rate as well as spatial velocity correlations were performed. Good agreement is found between simulation and experiment, showing that LES with the employed method to generate turbulent inflow conditions is successful in reproducing the wake flow.