Hi chaps,

I'm simulating large (VERY LARGE) tank flows, which consists of a high velocity jet (~40-50m/s) blowing into a tank of around 25,000m^3 volume. To sort out turbulence, the bog-standard K-Epsilon model is used.

Sometimes after solution, the calculation of EP shows an extremely small minimum or extremely large maximum somewhere in the grid; for some reason, a single cell in the flowfield has a completely out-of-bounds EP value, yet this cell doesn't seem to affect anything. Other times, the length scale (and thus ENUT) seems to be very strangely computed, leading to large fields of large length scale and ENUT. This behaviour seems to be more predominant using 2nd order discretisation schemes, but using the hybrid/UDS still results in some strange single-cell EP results (unsurprisingly, EP doesn't converge well at all, and I get the feeling that if this behaviour isn't halted after more than the first 5 iterations, it gets increasingly worse with no recourse).

Can anyone shed any light or possible ways around this?

Do you have a pressure relief in the domain? What's the domain fluid setting?

The domain has a single fixed inlet and 3 other fixed pressure outlets; fluid is air (although I've done a bit of multi-phase work involving slightly heavier gases).

I'm thinking that setting a higher initial condition might stop the formation of extremely low EP (and KE, of order 1e-5 to -10) but even if one out of bounds cell does appear, it seems to propagate, even using a scheme like Van Leer.

I've had problems with higher-order discreetization scheems in PHOENICS. It's amazing how unstable things can become! More relaxation and gridsize (bigger) seem to help often.

I was afraid you would mention those two things. I can under-relax the flow (0.01 on EP and KE) and usually it would work okay, but as soon as I go above some threshold in the relaxation it'll start to go funny, and I don't know why. It's actually quite irritating, since I'm running a 540-timestep simulation that will take around 22-26 days, so I don't really want to increase the mesh density or do anything that will take even more time.

Can anyone shed any more light on this phenomenon?

My current understanding is that the solution matrix becomes unstable and I have seen that happen in other types of simulations - the existence of a relaxation threshold. I believe it's sometimes very difficult to get around this problem. I believe PHOENICS should have some built-in code to pre-detect when relaxation settings will cause numerical instabilities.

This is just a thought, but do you think that you can limit the "damage" by setting much tighter VARMAX and VARMIN? I know it won't reduce the propagation of the anomalies (well, it might, I don't know) but it should reduce the damage caused.

discretisation k-epsilon

pouvez vous s'il vous plait m'envoyer 3 à 4 thèses et quelques articles qui ont pour sujet les écoulements turbulents leur modélisation en modèle k-epsilon ; discrétisation spatiale simple et simpler ; méthode des projections ; discrétisation des équations de dissipation et celle de l'énergie cinétique turbulente. voici mes coordonnées: tèl: 0021698374840 adresse: 2 rue de tozeur cité elward oued ellil 2021 tunis tunisie merci pour votre collaboration

Ты бы еще на ру��ком напи�ал

Hi I am MS student of Chemical engineering in " IRAN UNIVERSITY OF SCIENCE AND TECHNOLOGY" I need to know mor about "k-epsilon model in turbulent flow".Please help me .

Sincerely yours Ali Safavi MS Student Chem. Eng. Dept. Iran University of Science and Technology, Tehran, Iran

details

I would like to know how about so-called shear stresses. How do they look like in that model.