(求助）如何解决在fluent 里出现reverse flow

lalula2

大家好，本人第一次进入次论坛，由于本人中文有限，请大家谅解。
我在模拟一个outdoor thermal comfort, 我是用tetrahedral网格mesh的。因为在下尝试用hexahedral mesh, 结果还是有很多问题出先在fluent 里。 我的cell size 有 2百万个。
但是在iterate中，一直出现reverser flow 的情况。
请问有那为高手是environmental modelling 的专家？
请大家帮帮忙，因为在下急需要results.
谢谢

xrs333

Usually, reverse flow is not because of mesh or element type, but improper cumputational domain or/and ill-posed boundary conditions. Would you paste the mesh in the thread?

lalula2

Hi, thanks for reply.
How to paste the mesh in the thread?
I tried to simplified my model by recreate my mesh again with cooper mesh and now my mesh size reduced to 200,000.
I set my inlet as velocity inlet while the outlet is pressure outlet. While the laterel boundary condition i set as symmetry.

lalula2

This is the image of the iteration process after I recreated my mesh.
Somehow the k and epsilon residual are unable to converge while other show converge symptom.

xrs333

Hi, just paste some pictures of the mesh so as we can have a grasp of your model.

lalula2

I have uploaded my picture of grid. Basically I tried to simplified my model as simple as possible. I set symmetry boundary condition on 2 side while pressure outlet the top surface.

lalula2

The model as you can seen are an outdoor deck. The deck got a 4 meter glass fence wall infront which block most of the incoming wind. The reason I simulate this model is to study the opening gap at the bottom of glass wall to ensure thermal comfort around deck.

xrs333

IMHO, the computational domain is not adequate, its too small as to the boundary conditions cannot be imposed properly. Secondly, inlet and outlet boundary should not be placed adjacently.

通流

agree with xrs333's opinion. each dimension of your computational domain has to be at least twice of the side of the deck and the glass wall.  outlet condition is a bit tricky.  it should be far field.  but for incompressible flow, you cannot use it.  I have use a much larger domain in vertical direction, and set the top surface as a mirror.  it is kind of approximation.

lalula2

Thanks for the replies.
I agree with xrs33 and 通流. The computational domain is the root of the problem. But for my case I have limitation on my building dimension as I wish to save calculation time. That is why my lateral side on computational domain are too close.
I changed the lateral boundary condition to pressure outlet while I extend my vertical computational domain up to 20m and set the top surface as symmetry. But yet the reverse flow problem are still haunting me and I found some unrealistic turbulent flow in the domain. Any method to tackle this tricky problem?

浪子轻狂

according to your residual monitors window, both K and ε are shaking(especially for K). in my opinion,  at first you can use first order upwind scheme for k and ε. and then ,change into the second order upwind scheme supposing the solving is convergent.

with respect to the turbulent flow problem,  reasonable  turbulent model and parameter setting are both important. sometimes the reverse flow will appear at the beginning and disappear after some iterations. so a lot more iterations may be useful for the solving.

lalula2

Will it be helpful if I specified my target mass flowrate to a certain value?

turbulencest

FLUENT calculations are sometimes very tricky. But first of all, your setup for the problem is really important and needs to be checked.  For example,  BC's are correctly setup?
The convergence speed in most case depends on the grid quality. In most case, when the grid is fine enough, there is no significant difference between tetra and hexa.

Reverse flow is common during calculation. Finally it will vanish when the solution is converged in most cases.

"specified my target mass flowrate to a certain value?" may be not help you out. But you can try.

I suggest you to try the standard keps model first.

What's the BC values you used for k and eps? what's the relaxation parameter you choose?

lalula2

I tried to used the target mass flowrate, but is useless. I am using Reliazable K-E turbulent model without any changing on the default value.
I am using the default value for relaxation parameter.

I did some modification on the model as I added some fan below the canopy to increase thermal comfort around the deck. But yet the same problem are coming again that the solution are unable to converge.
I am mixing the meshing of tetrahedral and hexahedral as I was unable to mesh wif hexahedral around the deck area. I checked the skewness inside gambit and the worst element are around 0.8.
Anyone have tried to mesh with mixed meshing before?

浪子轻狂

"default value" is not a good way!