[nero235]

Hello,

I've created a modified version of the kEpsilon model for the openFOAM version 2.2.0. It can be found here: https://github.com/11101011/modKEpsilon

The models and my modification is based on this paper here: http://www.icmc.usp.br/~gustavo.busc...ew01_ijcfd.pdf

The initial reason for this modification were many tests I've run with the standard OF model and I had always problems with floating point exceptions based on divisions by zero and also negative turbulence values. OpenFOAM uses a scheme in its standard kEpsilon model which bounds negative or values smaller than 1E-15. The problem is that this number generates also floating point exceptions, when for example initializing a case with k and epsilon values of this magnitude. Someone can overwrite this bounding minimum by putting the following definitions in the RASProperties dictionary:

kMin         1e-5;
epsilonMin   1e-5;

This will mean that k, epsilon < kMin, epsilonMin will get bounded to this value.

const scalar minVsf = min(vsf).value();
    
if (minVsf < lowerBound.value())
{
vsf.internalField() = max
(
   max
   (
         vsf.internalField(),
         fvc::average(max(vsf, lowerBound))().internalField()
       * pos(-vsf.internalField())
    ),
    lowerBound.value()
);

vsf.boundaryField() = max(vsf.boundaryField(), lowerBound.value());
}
...

return vsf;

In this code, the lowerBound can be the unchanged (standard) value of 1e-15 or any other value, which can be defined by kMin, epsilonMin. The bounding itself of k and epsilon was not the solution, proposed in the paper above. It was more the bounding of nut > 0, the reaction coefficients and the source fluxes to positive values.

Regards, Sebastian

[Tobi]

Dear Sebastian,

you also checked out the

limitedLimitedLinear 1 1e-8 1e4

scheme? This scheme should bound your variables between 1e-8 and 1e4.

The problem is that this number generates also floating point exceptions, when for example initializing a case with k and epsilon values of this magnitude.

In a programming point of few, you should not get a floating point exception if you divide by 1e-15. Question: If you know that "maybe" your small values of k and epsilon will start the floating-point exception, why didn 't you use other values in your files k and epsilon?


Finally thanks for sharing.

[nero235]

Hello Tobi,

you also checked out the

limitedLimitedLinear 1 1e-8 1e4

scheme? This scheme should bound your variables between 1e-8 and 1e4.

I haven't tested this scheme but pretty much every other, including cellMDLimited grad schemes. At some point during my tests didn't believe anymore that I could bound the epsilon & k values in this way.

In a programming point of few, you should not get a floating point exception if you divide by 1e-15. Question: If you know that "maybe" your small values of k and epsilon will start the floating-point exception, why didn 't you use other values in your files k and epsilon?

I have found the lower bounding limit by modifying the bound.C. Then I used this value in my k & epsilon initial conditions to find out if value is the problem or something else. In my other test cases I always used 1E-1 as initial conditions for k, epsilon.

If you have some time, I would be glad if you could run some tests with the model.

Regards, Sebastian

[Tobi]

Hey Sebastian,

I really would Check it but i have to much stuff to Do.