CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   Mesh Generation & Pre-Processing (https://www.cfd-online.com/Forums/mesh-generation/)
-   -   Parameters of a good airfoil mesh (https://www.cfd-online.com/Forums/mesh-generation/199546-parameters-good-airfoil-mesh.html)

 CharlieBra7o March 8, 2018 12:31

Parameters of a good airfoil mesh

Hi guys,

I need to run airfoil simulations with multiple angles of attack and high Reynolds numbers (> 1 million).

My questions for this thread (I'll just ask it at this point as a TLDR):

Which areas in an airfoil mesh are the most important ones when creating a mesh (for lift, drag and transition point prediction) and why? What do I need to pay attention to?

For the ones with a little more time, here is the long version:

Since I am new to CFD, I have first read through probably every single thread on this forum regarding airfoils and also various papers, lectures, websites and video tutorials about this topic and tried to reproduce the meshes they came up with. But after trying many many different meshing approaches, I just cannot seem to see a solid trend in the different meshing parameters that seem to be of greatest importance in generating a good mesh. The results I am getting were not very good (CL error of 5-50% and CD error of 10-50%) when compared to wind tunnel data. I now think that I have tried everything I can think of and found in literature and the internet without much success - I am at a loss... So as my last resort, I am hoping that the experts amongst you that can hopefully help me out a little! :)

So now some specific details about of my simulations:

- Different airfoils are being looked at with different relative thickness ratios
- Different AOAs (region of attached flow) and different Reynolds numbers (> 1 million) need to be run with the same mesh
- Mesh topology: unstructured O-mesh with R = 100c (tried both quad and tri elements - mesh is swept => 2.5D)
- Analysis Type: Steady State
- Turbulence Model: SST
- Transition Model: Gamma Theta model => y+ < 1
- Software: CFX

- Max iterations: between 500 and 1000 (CL and CD look well converged after 300 iterations)
- Timescale: Chord/Uinf
- Residual Target: very small, so this won't interrupt the simulation
- Interrupt Control: change in average and standard deviation of CL/CD below a certain value within N iterations
- Turbulence Intensity and Viscosity Ratio are correct at the airfoil

How I approached my mesh generation:

First, I looked at the different basic sizing parameters CFX provides separately, keeping everything else constant, to see how they influence the results:

(1) Domain Growth Rate
(2) Number of elements chordwise around the airfoil
(3) Bias Factor/Growth Rate of the elements chordwise around the airfoil (smaller at the leading/trailing edge)
(4) Boundary Layer Elements
(5) Boundary Layer Growth Rate

I also manually refined certain areas with high velocity (gradients):

(6) Refinement of the area around the airfoil
(7) Refinement of the area around the pressure stagnation point at the leading edge
(8) Refinement of the wake behind the trailing edge

Generally, I expected that the "finer" I set one of these parameters, the closer I should get to the wind tunnel data. However, with many of these parameters, only the lift prediction (usually over-shoots) improved and drag prediction (usually too low) is worse, or vice versa. With some parameters the results for both lift and drag with the finer sizing parameters get even worse when further refining a mesh (sometimes 50-100% off).

I also expected that I should be able to use my findings from this study so that I am able to use those values for each of the investigated sizing parameters that gave me better results to sort of "create the ultimate mesh". However, this absolutely did NOT work for any of these combinations - the errors were larger than the coarser mesh with only one of the refinements.

Even with meshes that - according to everything I have read - should be more than fine enough (> 100k elements), I cannot even seem to reliably get lift and drag errors of < 10% compared to the wind tunnel data. To me it seems absolutely counter-intuitive that when I refine a mesh a little, the results get closer to the experimental data and when then refining it even further, the errors get larger again.

Another thing I observed and do not understand at all is that the pressure coefficient at the stagnation point sometimes is way below 1 (around 0.7) with rather fine meshes. Not capturing this properly is probably the main cause of my problem, I think...

OK, I think I wrote down all the essential information and hope that the more experienced guys of this forum can give me some more input on how to generate a good mesh around an airfoil :)

Thank you in advance!

 colinda1 March 12, 2018 02:23

Hello,

Difficult to answer without seeing the solution you obtained so far.
Maybe you could provide a color contour plot of your result on a fine mesh together with:
- information on the boundary conditions used: which type is used where, and its justification looking at the information of the wind tunnel experiments.
- a Cartesian plot of the velocity magnitude in the boundary layer; how many cells are present in the boundary layer? Just having a y+ < 1 for the first cell at the wall is not sufficient.
- how many cells do you have along the airfoil in streamwise direction from leading to trailing edge on your fine mesh?
- which type of mesh are you using in the boundary layer? Hexahedral?

Please post a picture of the mesh including a zoom on the boundary layer.

Best regards,
Colinda

 CharlieBra7o March 12, 2018 07:30

My boundary conditions are:

- Opening (subsonic) around the domain with fixed velocity components and fixed turbulence (k and Omega)
- Airfoil = Smooth, No Slip Wall
- Symmetry

I also also added a source term for a sustaining turbulence intensity.

The justification for comparing my results to the wind tunnel data is that these experiments have been done under the same conditions as my CFD setup.

The BL (hex) consists of a total of 696 elements streamwise around the complete airfoil (most of them reside near the pressure stagnation point and also the trailing edge) and 56 elements in normal direction. Unfortunately, I haven't found out yet how to create a cartesian plot of the velocity in only the boundary layer/around the airfoil - but from the contour plot it looks like laminar boundary layer takes up about 40 of the 48 elements - I tried to pay attention that it's resolved well enough.

Here are some pictures of the mesh with overlayed velocity contour: https://imgur.com/a/cPyGJ
The boundary layer zoom is at about 25% chord of the suction side.

 colinda1 March 19, 2018 05:50

Hi,

That looks good to me at first sight indeed.
I am not sure about the source term you mention though.
Maybe someone else has an idea?

Best regards,
Colinda

 shivam.hankare November 20, 2018 14:42

Hello,

On seeing your computational domain, few points to be noted are:

1. How have you segregated the inlet, outlet and no slip wall conditions to the
boundaries ?

2. The mesh implementation such as Triangular and Quadrilateral elements
have quite influence on the results and mesh qualities. Try implementing
sphere of influence around the airfoil.

3. What value of the airfoil area have you used in the setup in Fluent ?

The mesh generated in your case needs lot of improvement.

Regards,
Shivam

 All times are GMT -4. The time now is 00:34.