URANS of lobed shear layer
 

Hello folks,

I am performing a transient simulation of a shear layer ejected from an axisymmetric lobed mixer. The shear layer is unstable and rapidly rolls-up into coherent vortical structures which then interact and break-down into smaller-scale turbulence. Initially, I performed this as a laminar simulation without a turbulence model and allowed my spatial grid resolve the growth and convection of the structures in time and space (ie a DNS approach). The results were qualitatively favorable compared with an experimentally-obtained time-averaged flowfield, both in terms of the velocity and turbulence kinetic energy.

However, when I turned on the SST turbulence model (ie URANS approach) in an attempt to improve the agreement with experiments, all unsteadiness in the flow disappeared, even though I was solving the transient equations. The same thing occurs when I use the SAS-SST model. My initial suspicion is the physical timescale of the vortical structures is of the same order as the turbulence timescale, so the unsteady structures are averaged out by the Reynolds-averaging process.

Does anyone have experience on URANS of shear layers in CFX? Can you share any insight on the sensitivity of SST on the physical timescales of the unsteady flow structures present in the flow?

Thanks!

Interesting. Have you worked out the turbulence time and length scales from the SST simulation and compared them to the laminar/DNS approach? It would be interesting to see if SST is getting the time and/or length scales close to what it should be.

You may also find the result dependant on time step size for the SST model. Coarse timesteps = steady solution, fine timesteps = transient solution.

Thanks for the reply.

I haven't. Honestly, I'm not sure how to do that.

I used the same timestep size in both the laminar and SST simulations. The timestep size was chosen conservatively based on the Courant number and the expected amplified instability frequencies in the unstable shear layer. From the structures observed in the laminar/DNS simulation, the temporal resolution was adequate. Actually, I also suspected the timestep size, but subsequent reduction by 1/3 had no effect on the SST simulation.

I assume you meant "I'm not sure how to do that"...
Any turbulence modelling textbook should have them, I use Turbulence modellign for CFD by Wilcox. It is also here: http://www.cfd-online.com/Wiki/Turbulent_length_scale

Tiemsteps - OK, that does not seem to be the issue.

What did you set as inlet condition for turbulence?

I use turbulence kinetic energy and dissipation measured in a companion experimental study.