I am a beginner in using Fluent and Gambit and must admit I feel lost about assessing the mesh quality.

I am modeling a penstock with a very complicated geometry (narowing, elbows, change of cross-sections:rectangular, elliptical and finally circular). I have generated an automatic TGrid mesh with Gambit and obtained highly skewed elements in a proportion of 1%.

Then I created my own mesh with a Cooper meshing scheme and after many efforts obtained 0.06% of highly skewed elements (about 50 elements). Then viewing the velocity contours into Fluent, I have seen my results are very different.

What can I do to improve the mesh? How many highly skewed elements are acceptable?

I hope you can help!

Hi Julie,

difficult to answer your question. What do you call highly skewed? In principle one very bad cell can ruin your calculation...

When you ran in Fluent, did it converge well? The solution doen't depend on the quality of the mesh only, but on resolution as well, of course. Was your tet-mesh fine enough? Have you used boundary layer meshes? Is your mesh suitable for the turbulence model you've chosen?

Lot's of things to consider when you mesh a geometry.

If it's feasible, send me your gambit journal file. I'll have a look.

Did you contact Fluent support on this?

cheers, Laika, still orbiting

First of all, thank you. I feel less lonely!

So I had a look at my journal file, but as I built my geometry on several days; lots of steps are missing. Furthermore I think it is really not easy to use because I did a lot of do/undo operations,tried several interval sizes, .... Sorry about that.

I will try to answer your questions:

- when I ran into Fluent, I used unsteady state calculations and it took something like one hour to converge. The convergence criteria was set to 0.0001.

- Concerning the mesh, what I call highly skewed elements are elements with an equiangle skew value of more than 0.9. The maximum I have is 0.94.

- My penstock is approximately 1.9m in diameter, 43m long and the mesh size I used is 0.1m. What do you think of that?

- Finally the turbulence model is k-epsilon.

If you need other information, do not hesitate.

Hi Julie,

sounds ok. A skewness of 0.94 is not yet a disaster. The grid resolution should be ok to give an idea if the flow, if there are no recirculation regions or so. What about your Reynolds number? Have you looked at your Y+ values?

Why unsteady state? Is your flow unsteady? If you are lookinf for a steady state solution, you should use the steady solver. If your flow is unsteady, then how did you initialise your solution?

So you used a Tet-mesh all over? Can't you split the volume into smaller parts an mesh most of the geometry using a hex-mesh?

kind regards, Laika, still orbiting

Hi Laika,

I used your advice and managed to generate a Cooper meshing scheme. However I still have four elements with an equiangle skew over 0.90(worst:0.92).

I have a reynolds number of approximately 23x10^6.

Ok now I must admit that I do not understand why I should look the y+ values. I do not feel at ease with turbulent flows.

Furthermore, I used unsteady state because I wondered how we can say a flow is steady or not before calculations. I thought this would not change the results. It seems I am wrong, why?

Just before finishing, no there are no recirculation zone.

Thanks for your help.

Julie

Hi Julie,

you definetely sjould try to obtain a converged solution on the steady state solver!! If the solution is unsteady, you may have convergence problems, but at least you will have a good initialisation for the unsteady solution (if all goes well).

Have you managed to reduce averall cell count?

kind regards,

Laika, still orbiting