[buffi]

Hi,

I am simulating the flow in a channel using a modified simpleFoam and the k-omega-SST model. Every now and then, the solver can't solve for k or omega and their values just explode. Sometimes the simulation recovers from that state within a few hundred timesteps, sometimes it doesn't. Here is a sample output for a normal timestep and a problematic one:

Time = 4180

DILUPBiCG:  Solving for Ux, Initial residual = 0.002454946797335455, Final residual = 2.118201461652769e-08, No Iterations 6
DILUPBiCG:  Solving for Uy, Initial residual = 0.001256415853039685, Final residual = 1.037692770489594e-08, No Iterations 6
DILUPBiCG:  Solving for Uz, Initial residual = 0.003283733002456741, Final residual = 2.981279918575045e-09, No Iterations 7
GAMG:  Solving for p, Initial residual = 0.02201556239585755, Final residual = 1.282423077370965e-07, No Iterations 16
GAMG:  Solving for p, Initial residual = 0.008240116175764454, Final residual = 8.139580885669834e-08, No Iterations 12
time step continuity errors : sum local = 1.531473863748542e-06, global = -7.156743507425181e-15, cumulative = -1.45393053629309e-12
DILUPBiCG:  Solving for omega, Initial residual = 8.626697921486998e-06, Final residual = 8.156673087094997e-11, No Iterations 5
bounding omega, min: -877.5914029924204 max: 9963507.932995282 average: 104575.2141635721
DILUPBiCG:  Solving for k, Initial residual = 0.003874747711704138, Final residual = 5.92104344143439e-09, No Iterations 7
bounding k, min: -0.0001332252168376633 max: 0.05713142005592581 average: 0.003205000635194682
S_PRO= 4.802711559863782e-06W/K
S_PRO,D= 2.028083115231596e-06W/K
S_PRO,D'= 2.774628444632186e-06W/K
S_PRO,C= 0W/K
S_PRO,C'= 0W/K
ExecutionTime = 1411.6 s  ClockTime = 1434 s

Time = 4181

DILUPBiCG:  Solving for Ux, Initial residual = 0.002460388678483852, Final residual = 9.17931475513598e-10, No Iterations 7
DILUPBiCG:  Solving for Uy, Initial residual = 0.001256478971499746, Final residual = 1.16998685088727e-08, No Iterations 6
DILUPBiCG:  Solving for Uz, Initial residual = 0.003284937635407826, Final residual = 2.670988173110144e-08, No Iterations 5
GAMG:  Solving for p, Initial residual = 0.02191503352549878, Final residual = 1.226826542535508e-07, No Iterations 16
GAMG:  Solving for p, Initial residual = 0.008195653865403671, Final residual = 7.521263032788438e-08, No Iterations 12
time step continuity errors : sum local = 1.415562053229773e-06, global = 1.775764236376485e-15, cumulative = -1.452154772056713e-12
DILUPBiCG:  Solving for omega, Initial residual = 8.719510075834142e-06, Final residual = 4808428634545.903, No Iterations 1001
bounding omega, min: -2.164755144508585e+21 max: 3.598949521217116e+21 average: 1.308210762796994e+17
DILUPBiCG:  Solving for k, Initial residual = 0.3147840750606549, Final residual = 1.995451694496316e-16, No Iterations 1
bounding k, min: -4.418770186074049e-05 max: 0.05208677946321078 average: 0.001778830031969948
S_PRO= 1262457298.893128W/K
S_PRO,D= 2.02929169556324e-06W/K
S_PRO,D'= 1262457298.893126W/K
S_PRO,C= 0W/K
S_PRO,C'= 0W/K
ExecutionTime = 1448.59 s  ClockTime = 1471 s

I changed some values in my fvSolution during this simulation, but not when this behavior occured.

My fvSchemes for k and omega:

ddtSchemes
{
  default steadyState;
}

gradSchemes
{
  default cellLimited Gauss linear 1;
}

divSchemes
{
  default Gauss linear;
  div(phi,U) Gauss limitedLinearV 1;
  div(phi,T) Gauss upwind; 
}

laplacianSchemes
{
  default Gauss linear limited 0.333;
}

interpolationSchemes
{
  default linear;
}

snGradSchemes
{
  default limited 0.333;
}

fluxRequired
{
  default no;
  p;
}

relTol for k and omega is 1e-5, relaxationFactors are 0.7 (I had them as low as 0.3, but glitches occured as well)

Has anyone some advice for me regarding what could be wrong and what I could do about this?

Regards

Christoph

[msuaeronautics]

Try dropping your relaxation factor for omega and/or k to something less than 0.1, i.e. 0.05 or so.

I have recently run a kOmegaSST sim on a submarine and would have similar issues with the bounding omega exploding. Dropping the relaxation to 0.05 fixed it for me.

[chegdan]

Also, you have your default divScheme set to Gauss linear...which is unbounded. You might want to explicitly set the divScheme for your turbulence parameters to something bounded.

[buffi]

Thanks to both of you, I will have to try a number of different things now.

A relaxation factor of 0.1 or even 0.05 sounds really low, but I'll give it a shot if all else fails.

I tried a limited scheme:

div(phi,omega) Gauss limitedLinear 0 1e9;

which gave me a floating point exception when I restarted the simulation with this scheme enabled, but enabling it afterwards worked (odd!?)

I am currently using a smooth solver, which also seems to run without glithces in k and omega but it's slower than before.

[chegdan]

I personally have an affinity to the upwind family. For 2.1.x, you could try

div(phi,omega) Gauss linearUpwind grad(omega);

with

grad(omega) cellLimited leastSquares 1.0;

and then you can choose from the limiters

cellMDLimited
cellLimited
faceMDLimited
faceLimited

where these are arranged in order or least diffuse to most diffuse (citation http://www.cfd-online.com/Forums/ope...tml#post281280). I like them because they are a little dissipative and help when finding a solution. Also, believe I read in some ANSYS documentation that limiters can lead to stalling of the solution (maybe only in FLUENT?), so be aware of that.

[buffi]

I have OF 2.0.1 here, but I can use your suggestion anyway (at least OF didn't complain). These are my fvSchemes now:

gradSchemes
{
    default     cellLimited Gauss linear 1;
    grad(omega) cellLimited leastSquares 1.0;
}

divSchemes
{
  default        Gauss linear;
  div(phi,U)     Gauss limitedLinearV 1;
  div(phi,omega) Gauss linearUpwind grad(omega);
}

Let's see if it blows up again!

[buffi]

Well, I'm not seeing any problems regarding the glitches I described, but as they only show up occasionally I can't be sure that the schemes you suggested actually did the trick.

One problem though: I have cyclic boundaries and it seems that the suggested schemes have amplified oscillations I already had in my solution.

[chegdan]

you're still using Gauss linear as a default scheme for some of your variables. This linear scheme may be the source of your errors. I would change your default scheme to none, run your solver and then set each one of the schemes manually in your fvSchemes dictionary.

[buffi]

Originally Posted by chegdan

you're still using Gauss linear as a default scheme for some of your variables. This linear scheme may be the source of your errors.

Are you now referring to the glitches I described in the first post or to the oscillations I mentioned in my latest post? However, I now have all schemes set manually, but I have no experience regarding what might be suitable or not.