6 Things to Consider Before Switching Turbulence Model

Turbulence modeling for the Reynolds Averaged Navier-Stokes (RANS) equations - the basis of most industrial Computational Fluid Dynamics (CFD) software - is a complex field. Trying to strike a balance between accuracy and computational efficiency has given rise to a relatively large number of different turbulence models. I am not aware of a definitive list to match turbulence models to applications - if you are, please share! With this in mind I'll share 6 suggestions before you consider switching the default turbulence model in your CFD software.

RANS CFD Simulation of Air Flow Around a ChimneyUsing the k-omega SST turbulence model

The k-omega SST turbulence model is considered to be a relatively robust, accurate, and efficient model for a wide range of applications and is the default turbulence model in Caedium. Unless you are seeing results that call this into question I would suggest you stick with the default model. Before switching to a different turbulence model first consider that poor CFD results could also be due to:

1. Poorly positioned boundaries, e.g., an outlet too close to strong velocity gradients

2. Poor mesh elements (skewed)

3. Insufficient mesh resolution in critical flow regions

4. Unconverged results

Remember that when using standard wall functions (to avoid extremely fine near wall meshes) they require y+ values to be in the range 30-300 for reliable results.

Independent of the turbulence model, you can also try the following to improve accuracy:

5. A second order difference scheme. In Caedium select your volume and set: Substance->Solver->Schemes->Divergence = Linear Upwind.

6. A universal wall function valid over a wider y+ range (i.e., 1-300) near the wall. In Caedium select your wall boundaries and set: Physics:Wall->Type:Default->nut = U Spalding Wall Function.

Note that these options can sometimes result in less robust simulation behavior.

If you know that the fluid flow is likely to be entirely laminar due to the Reynolds number (e.g., in a pipe flow with Re < 2300) then you should switch the  turbulence model for the laminar model.

If after trying these suggestions you'd still like a second opinion, then I suggest you try the Realizable k-epsilon model, which is another widely used and reliable turbulence model.