CFD Online Discussion Forums

CFD Online Discussion Forums (
-   ANSYS Meshing & Geometry (
-   -   [ICEM] Spatial oscillation in results of a model with ICEM mesh (

highhopes June 16, 2013 18:38

Spatial oscillation in results of a model with ICEM mesh
2 Attachment(s)
Dear Friends,

I have been trying to simulate flow in a 2D curved duct with a rectangular section using Fluent. For the runs, I prepared a block structured mesh using ICEM CFD. Even though mesh quality seemed fine, when I plotted the pressure coefficient distribution along a curve on the surface, I saw the value oscillating at some zones of the curve. However when I did a simulation using a very old Gambit mesh file with the same case file, the curve was smooth without any kind of irregularities. I have to move on with ICEM CFD but I cannot find what is wrong, I even tried ICEM meshes much finer than the Gambit mesh, still the same problem. Has anyone faced a problem of this sort? What do you think is the real issue causing this?
Attached you can see the curved part of the geometry, and the oscillating pressure coefficient curve along the inner surface of the curved part.

Kind regards.

kad June 16, 2013 21:25

Oscillations in your solution can be a sign for insufficient mesh quality in regions with large gradients when using 2nd order discretization schemes. As far as I know gradients are always 2nd order in fluent and 2nd order upwind is some kind of standard scheme for the convective term. Theoretically it should always be possible to get rid of oscillations by refining the mesh.

In your attached pictures it looks like the expansion factor normal to the wall boundary is a little large. I mostly choose ratio between 1.05 and 1.1 with geometric node distribution for example.

Maybe you could also try a transient run, because oscillations can also be a sign for the unsteady behaviour of the flowfield.

highhopes June 17, 2013 08:06

1 Attachment(s)
Thank you very much for responding. I changed the mesh upon your advice and did a transient run. However, I got similar results. Attached you can see the picture of pressure coefficient distribution in the proximity of the same boundary. By the way, yplus is around 1.2.

kad June 17, 2013 09:01

Further thing you can try is reducing under relaxation factors. This should make your solution more stable. You can also do a test run with 1st order approximations or even coarsen the mesh.

Did you set up the physics of the flow correctly? Is it really turbulent or could it be laminar. Are the boundary conditions defined appropriate?

The value of yplus depends on which wall function you are using. For standard wall functions 1.2 is too low, as the first node from the wall should be in the logarithmic layer. This means ystar between 30 and 300.

highhopes June 17, 2013 09:56

Under relaxation factors are 0.1 for pressure, 0.2 for momentum, 0.4 for turbulent kinetic energy and dissipation rate. Reynolds number is around 390000.
I am using RNG k-epsilon and enhanced wall treatment.
I will do a 1st order run as well. Thank you again.

highhopes June 18, 2013 10:44

I have done a 1st order simulation but same surface irregularities were formed even though I changed the absolute convergence criteria to 10e-10. After convergence, I continued the simulation with second order approximations and the irregularities disappeared considerably until convergence. However, In this way the simulations take too much time and I still I cannot figure the essence of this problem.

highhopes June 23, 2013 16:21

2 Attachment(s)
Dear Friends,

The reason I am posting a reply to my own thread is that I have not fully solved my problem despite the things I have done so far,
- Tried different boundary layer mesh thicknesses.
- Tried different yplus values around 1. Because I employed enhanced wall treatment.
- Generated grids with homogeneous distributions of cell volume.

The final distribution of pressure coefficient values along the same line as I mentioned in the previous messages is as you can find attached in the first picture below. The other one shows the distribution in the middle of the curve.

Could you please help me solve this problem, or at least tell me where to look for an answer. Might there be other causes related to the usage of ICEM CFD?

Note: I am posting this topic in the meshing forum because I believe it is mesh related. As I stated in the first message, the mesh generated by using Gambit gives flawless results. (However, the mesh is all I have and I have to go on with ICEM CFD)

Far June 23, 2013 16:54

please post some pics of Gambit mesh. please post some shots of your geometry (full scale)

highhopes June 23, 2013 18:17

1 Attachment(s)
Attached is the geometry of the duct. I will attach and post the gambit mesh as soon as possible. The vertical line on the left is the inlet. The line at the bottom is the outlet.

Far June 23, 2013 18:33

This isn't a difficult geometry to mesh. are you sure that the every thing is same expect change in mesh ?

highhopes June 23, 2013 19:01

Yes it is a very simple mesh and I have worked with much more complex meshes before. However, I cannot find the source of tihs simple surface irregularity. I just replaced the mesh during my runs, I did not change anything else. However, the surface results from the gambit mesh were simply smooth compared to icem cfd mesh.

highhopes June 24, 2013 04:13

2 Attachment(s)
Here are two snapshots of the gambit mesh that I imported into icem cfd.

Far June 24, 2013 19:44

hmmm. I cannot understand why they have different results ... may be due to different versions of solver?

highhopes June 25, 2013 05:32

Does Gambit have a different solver? I chose Fluent V6. Is there an alternative solver for Fluent V6 that I can use in ICEM CFD?

Far June 25, 2013 06:20

Ok. I confused things. I thought gambit mesh was used by someone in past.

Did you check the units of both meshes? other parameters are similar ?

highhopes June 25, 2013 07:15

I still have the gambit mesh, and I did a simulation using this mesh to compare the results. I need ICEM CFD to make some further changes on the geometry and the mesh.
The units and the sizes are the same. I even tried to make a carbon copy of the gambit mesh, I mean, as identical as possible. Funny thing is that the oscillating results were there again.

Far June 25, 2013 08:44

so you have three meshes? two made in ICEM and one in Gambit by someone else.

can you share your three cases?

All times are GMT -4. The time now is 12:55.