Where did my TKE go??!
 

Hello everyone,

I am having some issues with my code, and I am looking for a savior! :D

To summarize:

1. I wrote a RANS solver and verified the implementation with several NASA cases online. I obtained the exact same predictions than CFL3D and FUN3D for both the flat plate and the 2D bump cases. I am using Roe FDS with 2nd order MUSCL (and 1st order for turbulent quantities), and 2nd order central differences for the viscous fluxes. Basically, I have more or less EXACTLY what they have. I am using the kw-SST model (2003 version based on strain rate).

2. Now, I am running a 2D NACA0012 case (also provided on the NASA website). I was expecting it to go very smoothly, since it is very close to the 2D bump case, but for some reason it does not work! (that would have been too easy...)

My simulation converges to the correct solution (checked CD, CL, etc), with very low residuals (1.0e-7 or so) and then everything suddenly diverges... (see attached figure). The simulation does not blow up in one iteration. It happens quickly, but progressively, which makes it harder to track down...
Right after the 'explosion', all the TKE is gone from the airfoil.

Would anyone have any idea? Where could something like that come from, given the fact that my 2D bump solution was perfectly fine? I was thinking that maybe my limiters were causing this (if k or w go below a small value epsilon, then I impose k or w = epsilon, with epsilon = 1e-12). what value should I use for epsilon? larger than that? smaller? Since I use the same non-dimensionalization than FUN3D, k is divided by the speed of sound squared -> it is very small (on the order of 1.0e-9 in the freestream, and then it is decays further). If I choose a smaller value, then some production terms become ridiculously large...(k or omega at the denominator)

Thank you for your help!

Joachim

I am no RANS expert, but from looking at your residuals, omega seems to have some instabilities, that causes this blowup. Can you test an other RANS model to ensure that all your problem is set up in a consistent way? Can you look at the solution before the blow up, may be that will give you an idea what goes wrong?

Thank you CFDnewbie!

I tried to run the same case with SA but got the same issue...

Once again, the residuals drop (solution looks good) and then suddenly diverge.
It seems that the instability begins in the wake. Interestingly, the y-momentum residual starts to diverge before the others.

It is initially very small, but then the magnitude increases until it diverges (see table below)

iteration y-mom residual (in diverging cell)
5034 -8.11329028786722528E-009
5035 8.47770253303847809E-009
5036 -8.92618973180603348E-009
5037 9.33175742408309805E-009
5038 -9.82312127473239372E-009
5039 1.02741911558318355E-008
5040 -1.08130061068217875E-008
5041 1.13143975733615010E-008
5042 -1.19057363880820111E-008
......

Is there a way to avoid this positive/negative growth??
Thank you so much!

Joachim

I just checked, and the exact same thing happens with the kw SST model. I also tried a finer grid and obtained the same results...

I cut the grid in two, so that I am only solving the upper airfoil (aoa = 0 deg).
With both grids, and both turbulence models, the divergence appears a few cells after the trailing edge, at the inviscid wall / symmetry.

I attached a pretty figure to show you my issue...

Any idea?
Thanks!

Joachim