[northfly]

When you are using LES method to calculate the overflow of an airfoil, is there any requirement on the span width of the calculation domain? Is this related with the Reynolds Number of the flow? If anyone is aware of this, please let me know, no matter it is a link to a paper, a criteria equation, or some explanation on why this should be concerned. Thank you very much!

[FMDenaro]

Originally Posted by northfly

When you are using LES method to calculate the overflow of an airfoil, is there any requirement on the span width of the calculation domain? Is this related with the Reynolds Number of the flow? If anyone is aware of this, please let me know, no matter it is a link to a paper, a criteria equation, or some explanation on why this should be concerned. Thank you very much!

Using periodic conditions in LES (but also in DNS) requires a lenght in spanwise direction enough large to allow a correct development of spanwise structures. In general this is controlled by checking the spatial correlation, allowing the separation lenght to be smallest of the computational lenght.
On the other hand, an airfoil is a 3D geometry so you cannot set a too large section using periodic conditions.
There are several simulations you can see in literature using the lenght

[LuckyTran]

I'm mostly repeating what FMDenaro already said using different words. I'm assuming you have an airfoil in say a large freestream, if it was an airfoil inside a small wind tunnel then you'd model the walls of the enclosure.


1. If you apply some freestream boundary conditions (or any other boundary condition intended to mimic a natural boundary condition at infinity) then the location of your freestream boundary needs to be appropriately placed. This requirement is in the same sense as when you do any model, not specific to LES. E.g. you run steady RANS or whatever.

2. You are doing LES, which is time-resolved & partially spatially resolved. You need to show that there is no statistical influence of the boundary condition onto the turbulent structuers near the airfoil. To prove this, you need to show that the two point correlations of velocity fluctuations decay to zero going from the airfoil to the boundary. Actually you need to do this also in the streamwise direction to show that the inlet and outlet are far enough away. If the two-point correlations don't decay to zero, then your turbulence statistics are contaminated by the boundary condition being too near (and you are not solving for the actual physical turbulence of the flow).

[northfly]

Originally Posted by FMDenaro

Using periodic conditions in LES (but also in DNS) requires a lenght in spanwise direction enough large to allow a correct development of spanwise structures. In general this is controlled by checking the spatial correlation, allowing the separation lenght to be smallest of the computational lenght.
On the other hand, an airfoil is a 3D geometry so you cannot set a too large section using periodic conditions.
There are several simulations you can see in literature using the lenght

Thanks FMDenaro!

About how wide should be the span, is there a equation criteria to determine it? or is it purely based on experience?

Because I heard from some people who are doing Cylinder Overflow Problems, they usually use 20~30 times of the cylinder diameter as the span width, and saying when the reynolds number is smaller, the span should be wider. It is not hard to understand that when Reynolds number smaller, the turbulent structure is bigger, so need more span. My airfoil overflow, the Reynolds Number is between 200,000-Dozens of Millions, Relative Thickness is around lower than 25%, how should I choose this span width? I still cannot find this out. Please help out. Thank you very much!

[northfly]

Originally Posted by LuckyTran

I'm mostly repeating what FMDenaro already said using different words. I'm assuming you have an airfoil in say a large freestream, if it was an airfoil inside a small wind tunnel then you'd model the walls of the enclosure.


1. If you apply some freestream boundary conditions (or any other boundary condition intended to mimic a natural boundary condition at infinity) then the location of your freestream boundary needs to be appropriately placed. This requirement is in the same sense as when you do any model, not specific to LES. E.g. you run steady RANS or whatever.

2. You are doing LES, which is time-resolved & partially spatially resolved. You need to show that there is no statistical influence of the boundary condition onto the turbulent structuers near the airfoil. To prove this, you need to show that the two point correlations of velocity fluctuations decay to zero going from the airfoil to the boundary. Actually you need to do this also in the streamwise direction to show that the inlet and outlet are far enough away. If the two-point correlations don't decay to zero, then your turbulence statistics are contaminated by the boundary condition being too near (and you are not solving for the actual physical turbulence of the flow).

Thank you very much LuckyTran!

You remind me another important problem, how far the the farfield boundary conditions should be when doing LES?

When I am doing RANS, usually I would like a far field as far as possible. When doing benchmark on NACA 0012 airfoil, the recommended farfield is 500 times of chord length far away. I did the calculation domain independence study, when choosing 50 times of chord length or larger, the difference could be negligible.

When it comes to LES, I am new on LES calculation, I am not sure I can state the question correctly or not. Because the grid volume size in far field is usually much larger, if I impose BC there, will the free stream turbulent structure information be dissipiated before they reach the airfoil and influence the final results? If far field is too close, then like RANS, you won't be able to impose a correct BC. Using a far far field and keep the grid volume size small all the way from airfoil to the far field? That would be too expensive. How would you handle this? Please help out, thank you very much!

[FMDenaro]

Originally Posted by northfly

Thanks FMDenaro!

About how wide should be the span, is there a equation criteria to determine it? or is it purely based on experience?

Because I heard from some people who are doing Cylinder Overflow Problems, they usually use 20~30 times of the cylinder diameter as the span width, and saying when the reynolds number is smaller, the span should be wider. It is not hard to understand that when Reynolds number smaller, the turbulent structure is bigger, so need more span. My airfoil overflow, the Reynolds Number is between 200,000-Dozens of Millions, Relative Thickness is around lower than 25%, how should I choose this span width? I still cannot find this out. Please help out. Thank you very much!

Your problem is that while a cylinder has a section that is constant in spanwise condition, so that you can enlarge the spanwise domain as you want, an airfoil is 3D in its geometry. So if you enlarge arbitrarily the spanwise domain, the solution is not physically relevant to a real airfoil. You have to find a compromise.
Furthermore, for increasing Re number, the energy cascade is enlarged so you have more turbulent structures of smaller lenght as the Kolomogorov lenght scale is smaller. For this reason one has to refine more the grid for higher Re number.

[northfly]

Originally Posted by FMDenaro

Your problem is that while a cylinder has a section that is constant in spanwise condition, so that you can enlarge the spanwise domain as you want, an airfoil is 3D in its geometry. So if you enlarge arbitrarily the spanwise domain, the solution is not physically relevant to a real airfoil. You have to find a compromise.
Furthermore, for increasing Re number, the energy cascade is enlarged so you have more turbulent structures of smaller lenght as the Kolomogorov lenght scale is smaller. For this reason one has to refine more the grid for higher Re number.

Thanks FMDenaro!

There is a midundertanding, the airfoil doesn't change with the span position either, which is the same case as cylinder.

[FMDenaro]

Originally Posted by northfly

Thanks FMDenaro!

There is a midundertanding, the airfoil doesn't change with the span position either, which is the same case as cylinder.

Well, this is a non physical 3d geometry but if you need a simplified model that’s ok.
You can compute the spanwise autocorrelation to assess that it goes to zero. At that value of the separation variable you can be sure that the spanwise extension is sufficient.