CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Why k-epislon model perform a bad result at corner (

Lwrence March 22, 2006 09:39

Why k-epislon model perform a bad result at corner
i used the CFX to simulate a compartment fire with k-epislon model, the fire located at corner.

why the k-epislon model perform a worse result when the fire at corner than teh fire at center.


Jonas Larsson March 22, 2006 17:59

Re: Why k-epislon model perform a bad result at co
I would guess that when the fire is close to a wall the near-wall treatment in the tubulence model behaves badly.

When you have fire close to the wall you'll probably have significant bouancy effects etc. in the near wall region and heat-transfer to the wall might also be important for how this works in the near-wall region. This is a very challenging task for any turbulence model.

This is one possible explanation to your observation. Without having seen anything from your simulations it is diffucult to say something certain. Just my $0.02.

bouria March 22, 2006 22:35

Re: Why k-epislon model perform a bad result at co
k epsilon is known not to perform very well for flows with high streamline curvature. use the SST model or RNG k epsilon model instead.

TB March 23, 2006 04:37

Re: Why k-epislon model perform a bad result at co
Adding more equations doesn't always give better solution. You might have to check the grid error around the corner before you jump into conclusion that turbulence model is the main error source.

If simple model predicts flow very badly, I'll have doubt that more complicated model will give more accurate solution.

Lawrence March 23, 2006 10:55

Re: Why k-epislon model perform a bad result at co
thank you for Ladies and gentlemen

i have check the grid , there are no problem in the domain and corner.

Recently, i found that when i set the static temperature to be 298K. the result was very good, but if i use other static temperaature even 297k or 300k , the result was very strength.

why have this strange result. anybody can give me some suggestion.


Robin March 23, 2006 11:40

Re: Why k-epislon model perform a bad result at co
Switching turbulence models does not necessarily mean adding more equations. k-e and SST are both two equation models, only they are formulated differently. I would follow the SST recomendation, which would allow more refinement near the wall and better capture the physics.

k-e is anything but simple near the boundary, but the way.


TB March 23, 2006 21:22

Re: Why k-epislon model perform a bad result at co
Hi Robin,

Someone might later suggest the use of RSM if SST doesn't work out better. THEORETICALLY, RSM SHOULD predict flow even better because it has more equations to describe the physics.

What do we need to take advantage of SST model? Yplus<2? I can afford such a fine mesh on airfoil or cyclinder. However, for a room or a tunnel with much bigger scale, you're going to get thousands more nodes everytime you add a layer of elements close to the wall. Besides, if you stretched the elements too much, solver might perform poorly.

I would imagine that this problem might go to transient in some stage. That's even worse if you need such a fine mesh. What's the time do you expect to wait for a solution and how many cpus do you need to reduce the computing time? I just can't convince myself to use SST model in this case and would be very happy if you can tell me how to do that effeciently.

Robin March 27, 2006 11:48

Re: Why k-epislon model perform a bad result at co
Hi TB,

You don't NEED to have Y+<2 for SST. The model allows for this flexibility, but will use a wall function just as k-e does if the Y+ is larger.

If you need to model something close to a wall, then you need the appropriate mesh resolution and the SST model allows for this. Anyway, my main point was that SST does not solve additional equations, only two. There are also advantages to SST away from the wall.

Regards, Robin

sam March 31, 2006 06:44

Re: Why k-epislon model perform a bad result at co
do u have idea wat does sst offers ????????????????????? it might be intersting that it does have any thing new............................................... .. surprized............... it is just a mixture of K-Epsilon away from wall and K-omega for wall. it is essentailly a low reynolds number flow means u need Y+ of order 1. It has function f as switiching. 0 = k-Epsilon and 1 = K-omega.

now comes to your question..... by solving ur flow by sst in interior is same as solving by K-Epsilon and along wall it would advantage of k-omega

understood ????????

Thien April 2, 2006 08:44

Re: Why k-epislon model perform a bad result at co
What I understand is that RANS or URANS simulates poorly strong unsteady flows. I suggest u define a region at the corner and then use LES for this region (u can perform Smagorinsky or Mixed-Scale model in CFX by user Fortran subroutine USRVIS)

sam April 2, 2006 15:45

Re: Why k-epislon model perform a bad result at co
for that purpose use DES deattached eddy simulation. it is mixture of LES and RANS. i.e near wall LES and away from wall RANs

Thien April 3, 2006 02:19

Re: Why k-epislon model perform a bad result at co
Dear Sam, Yes it is a kind of DES. But I mean we use LES only for the corner region not for the whole wall region of the computational domain

Robin April 3, 2006 10:02

Re: Why k-epislon model perform a bad result at co
Hi Sam,

Ease up on the coffee a little and you might regain some control of your keyboard.

SST is not simply a mix of k-omega and k-epsilon and it is NOT a low Reynolds number formulation. SST is a k-omega model, meaning that it solves the k and omega equations only. The terms of the equations have been expanded such that by eliminating one of these terms, it resembles the epsilon equation. What makes SST different from a straight k-omega model is the gradual elimination of this term as you move away from the wall and how it accounts for the transport of shear stress.

While what they solve for are similar, the numerics are quite different between SST and k-e. This is an important detail, as it means that additional physics can be accounted for much more easily. A prime example is how boundary layers are handled; while it is very difficult to make k-e behave well enough to integrate to the wall (this is what a so-called low Reynolds number k-e wall function does), it can be done very easily with k-omega. Again, this is not to say that you MUST integrate to the wall with SST, a common misconception. SST also includes limiters that are difficult to make numerically stable in k-e, this is what makes the difference in the free stream.

If you really want to know more, take the turbulence course given by Dr. Menter at the users conference.

Regards, Robin

sam April 8, 2006 07:46

Re: Why k-epislon model perform a bad result at co
thanks for your disucssion i have gained alot. can u tell little more abt sst . i will be thanksful

All times are GMT -4. The time now is 08:14.