[marcanyada]

Hi!

I am trying to simulate the flow around several 2D airfoils used for wind turbines (S809, S812 and S813). The settings for the analysis are:

• Analysis type: steady state
• Fluid: Air at 25 C, Fluid model: isothermal, Turbulence SST with Gamma Theta for the transional turbulence model
• Solver Control: Advection scheme: high resolution, turbulence numerics: high resolution. Convergence criteria: MAX = 1e-4

I monitor the values of cl and cd. I have found two problems:

1) Most of the times I can not get the solution to converge. When using Auto Timescale or Physical Timescale the results oscillate, and when turning to Local Timescale the solution is steady, but the residuals do not achieve the target. I have tried almost everything of what is here suggested: http://www.cfd-online.com/Wiki/Ansys...gence_criteria (meshing finer, changing timescale,

How can I change from local timescale to physical timescale for last iterations?

2) The results I get differ quite a lot of the experimental data I got (even for low AOA the difference is around 20%). I have tried changing the transitional model, but the results do not improve.

I would be very grateful if someone could give me some clue of what I am doing wrong and how I could try fixing it. Thank you!

[marcanyada]

Hi!

I write again... Please, if anyone has any clue how to improve the convergence or to get more accurate results, it would be really appreciated.

I have tried running several transient simulations too, increasing and decreasing the timestep, but the solutions never converge... and I am absolutely run out of new ideas to experiment with...

Thank you!

[mvoss]

Did you run it transient?

[ghorrocks]

Please post an image of your mesh, your results and your output file.

[marcanyada]

Originally Posted by ghorrocks

Please post an image of your mesh, your results and your output file.

I know the scale for the values of cl is quite small, so I guess that the oscillations should not be such a big problem. It can also be noticed the change from "local timescale" to "auto timescale" after around 1500 iter.

But when changing the angle of attack it gets worse or when the stream velocity is increased. I could get some acceptable values for flow velocity of 11m/s at some angles of attack (around 10% more of the expected value), but for other angles of attack, using the exactly same configuration the difference is really big (for example for alpha=0º).

[marcanyada]

These results are using a transient model, with ideal gas, with higher velocity than in the former post.

In most of the cases when running steady, the solution with local timestep made some oscillations at the end, but when changing to auto timescale or to physical timescale it diverged.

I am quite new with this sort of simulation, so I am not sure when I can take a result for valid (e.g. when it is oscillating) or if it is complete wrong.

Thank you so much for the help!

[ghorrocks]

I would suggest your mesh quality is not as good as it should be. I see two problems: The orthogonality of the mesh near the foil is a bit wonky, and there are big jumps in mesh size especially near the trailing edge.

Things will be a lot easier if you improve the mesh quality.

[JuPa]

I remember ghorrocks saying 2D in Fluent is much faster than 2D in CFX (since you effectively aren't solving for any equations in the w vector). Saying this, I would just jump ship to Fluent. You don't seem to be using any complex multiphase models (where CFX excels in my opinion) therefore you may benefit from speedup using Fluent.

However if you perform a sensitivity study using both CFX and Fluent you should achieve the same results.

Edit: your mesh looks poor. Spend time on it.

[ghorrocks]

I agree with Mr CFD's comments - if you are doing 2D studies the genuine 2D model in Fluent is far superior than CFX's pretend one.

You need to do sensitivity studies regardless of the solver.

And time spent improving mesh quality is never wasted. And often it is critical in obtaining a good answer.

[pimpa]

Originally Posted by ghorrocks

... if you are doing 2D studies the genuine 2D model in Fluent is far superior than CFX's pretend one

Superior in speed, in accuracy, worthwhile papers documenting this, please ?

[ghorrocks]

It is superior because Fluent has a true 2D model (ie only models U and V velocities), whereas to do a 2D simulation in CFX you do a 3D model which is 1 element thick, so you still model the U, V and W equations. This will make CFX much slower to converge and use more memory.

This means for the same amount of effort you can do a finer mesh and/or smaller timesteps - so it results in accuracy improvements.

My reference is the theory documentation of the two softwares and any numerical modelling textbook which will tell you that a 2D model will run heaps better than a 3D model with the same number of elements.

[Ivanrips]

Hi;

Here a tutorial in CFX with convergence

https://www.youtube.com/watch?v=ngNZdyWTUIo

Regards

[marcanyada]

Thank you all for your replies. Moving to Fluent was not a possibility, as the 2D case was only as a preparation for the 3D. Improving the mesh quality solved the problem, however I had to use a really fine mesh, so I am now worrying what the calculation time for the 3D case will be... but that is another issue ;-)

[zahid hussain]

i want to simulate the airfoil in different angle of attack tail me the
procedure plz help me this is my final year project

[Ivanrips]

Originally Posted by zahid hussain

i want to simulate the airfoil in different angle of attack tail me the
procedure plz help me this is my final year project

Hi;

Here thre are tutorials, only change angle of attack.

Regards

Part 1
https://www.youtube.com/watch?v=ngNZdyWTUIo

Part 2
https://www.youtube.com/watch?v=QBcJubC6LEI

Part 3
https://www.youtube.com/watch?v=6RvLtWr07uE

Part 4
https://www.youtube.com/watch?v=2lhkyt9eV4g


Regards