Cyclone separator RSM convergence
 

Hey foamers,

first of all i'm quite new to OpenFoam and CFD in general, so please forgive my lack of knowledge.

So i have this case where i like to simulate the turbulent swirling flow field of a cyclone separator in regard to the pressure drop and the velocities. Since it is a swirling flow inside the cyclone i found out that reynold stress models work best for such cases. I used the blockMesh utility to generate my mesh (1.5 million cells) and then started to get my initial conditions and values for the Reynold stress model using standard k-epsilon successfully. I was using pimpleFoam solver and standard schemes which i mostly got out of some tutorial cases.

Up to this point everything ran smooth and nicely, even LES models (dynamicKEqn/Smagorinsky) did work, although taking much longer to run through but delivering better results.

Using the R utility i calculated the R file necessary to run RSM models. I chose the SSG model and started running the case. At first everything is doing well but after a couple thousand of iterations the simulation just blows up, mostly within just 1 or 2 iteration steps. I get the floating point exception etc. error and i just don't know how to get rid of this problem. I changed my schemes to first order to produce better numerical stability and also tried to first run the case half through with k-epsilon and then change to SSG, both with no success :(

So my question here is if anybody does have experience with RSM models and knows what really is important especially when working with these models. The fact that every other turbulence model works is just confusing:confused:

My boundaries are inlet, outlet and a couple of wall patches. My inlet velocity is 25 m/s and i calculated values for epsilon and k (assuming that the turblulence intensity is ~ 3.5%) using the turbulence properties tool provided by cfd-online.com.

I hope you guys can help me with my problem, any help is highly appreciated!

thanks in advice

As you probably figured out by now, RSM models are very hard to
converge. I did some RSM simulations in channels and pipes and the procedure that you used (to do k-epsilon simulation, then to use R utility and then to change to RSM simulations) really worked, although I had other problems. Also, I started from denser mesh, and after convergence was achieved, I mapped that solution to fine mesh and I did again RSM simulation. You have to be patient and to play with under relaxation factors. This page was helpful to me

https://openfoamwiki.net/index.php/O...hm_in_OpenFOAM

As far as I know, there are LRR and SSG models implemented in OpenFOAM, you could try both.


Best regards,
Darko

Hi Darko,

thanks for answering. My relaxation factors right now are:

relaxationFactors
{
fields
{
p 0.3;
}
equations
{
U 0.7;
R 0.7;
k 0.7;
epsilon 0.7;
}
}

I will give it a try and reduce the values, espacially for R and epsilon. And i'm going to try LRR as well and post my results later on :)

Best regards,
Robin

Hi,

LRR diverged even faster than SSG. It started bounding epsilon almost at the beginning and after a while it finally blew up.

Changing the relaxation factors did improve the simulation, but way too little so still no satisfying convergence :(

Do you have any other ideas left? I'm desperate :confused:

Best regards,
Robin

Robin,

Could you upload case?

I could have a look, but I can't promise anything.


Regards,
Darko