Hi! During a simulation and after changing some parameters (I don't remember which exactly, probably some underrelaxation factors), in order to achieve convergence in my problem, the epsilon residual (of the simple k-e model) was gone flat and remained that way! And I do not mean in an exponentially way but steeply! And it was not only the plotted line that seemed to be but the printed value of the residual was exactly 8.9932e-01 in all the iterations after the specific on which the problem appeared. (the problem appeared after interaction with discrete phase).

I tried to change the underelaxation factor of epsilon but the residual in any case jumped to a new value where it remain again constant.

I also tried to change k-e to ke rng but nothing happened. Epsilon was still.

When I switched to RSM turbulence model, epsilon residual quickly decreased below 1e-3. After that and when I switched back to simple k-e, epsilon residual jumped (normally) until the DPM iteration where it became flat again!

Can anybody tell me what's going wrong? It's the first time this happens to me.

Thanks in advance,

Ilias

PS FLUENT VER 6.2

Do you mean that the value of the value of the eps residual actually freezes???

Yes when I use the k-e standard model. Any suggestions?

Ilias

You should decide on the particular turbulence model after running with the available ones. Usually, k-eps is pretty dependable. try using all models with default settings first for all convergence related inputs.

swarup

it only means that you are not able to get convergence with k-e model, try monitoring the residuals in absolute or normalised way, rather than scaled, this might give you more idea of the flow or convergence

Are you sure that it is not a bug of FLUENT? I mean that even I change the discretization scheme for epsilon, its constant value doesn't change (even when I switch from 1st to 3rd order). And when I say it doesn't change I mean that it remains absolutely the same, even its last digit. The same happens for every change of parameters I try. It is not possible to make the residual change. And it is not the usual exponential convergence to a value, that could be partly resolved with multigrid techniques. Because in that case the switch between ke std and ke rng would make a jump of the residual. Even after DPM calculation the value of the residual should have been changed.

Further more I have always been using ke in this kind of problems and I had never faced any difficulty. To be precise I have already converged the turbulence equations of this problem with the ke std model. That's why I am about to conclude that it is a bug of the software...

Ilias

there are very little chances of bug in that, because if you can see k-e working on other cases, (and since its been tested very much), fluent can not dicreminate that if you give me this case i will not converge otherwise every other case i will converge. i hope you got what i am saying. So the problem might be somewhere else than the solver, my suspision is that, since k-e model is difficult to converge, i might end up in a situation where solver 'stalls' or stops to remove the error. If it is a solver issue you can check this by switching to FAS, use 4 levels for it, and see if the solution now converges. if you get the converged value here, this means that it was really a solver issue, that is it is convergence problem rather than problem set up or bug in software. I asked to monitor other forms of residual since, scaled residual might stay at one place but one of the other two forms (normalised and absolute) may change, try doing little experimentation with it. it will help.

FAS is not available beacause I use the segregated solver. Is anything else that I can change in multigrid?

The other forms of residual, I will check...

Thanks,

Ilias

i know that FAS is not available with segregated solver, just switch to coupled explicit solver with small courant number. i am just saying to check the behaviour.

Anyway if you can not switch to FAS, another option is first use a very coarse mesh and get the converged solution, here you will get the solution, write the whole solution field as initial guess for finer mesh , read this initial guess in you main cas and do the calculations, if it were convregnce problem of k-e this will remove this problem.

When I turn off the two-way turbulence coupling for the discrete phase then the epsilon residual is going down at the correct value (when I say correct I mean comparing to the initial value before the residual jump). And this is the only way to un-froze the e residual as I have already described.

Can you explain it?

[pipolaki]

Hello,

I'm currently experiencing the same problem with my DPM simulation. Epsilon residual is stuck at 0.97 and, as soon as I uncheck the 2-way turbulence coupling option, the epsilon residual is decreasing.
Is there any explanation about this ?
Thanks you for your help.

Originally Posted by HP
;123393

When I turn off the two-way turbulence coupling for the discrete phase then the epsilon residual is going down at the correct value (when I say correct I mean comparing to the initial value before the residual jump). And this is the only way to un-froze the e residual as I have already described.

Can you explain it?