[jaiswal]

i am trying to implement spalart almaras turbulence model in 2D Navier-Stokes solver.

Even after spending lot of time im not successful.

My first doubt is whether im using d (distance from the closest surface) correctly or not.

How is d calculated?

I am doing calculation on NACA 0012 airfoil. And to find the distance from closest surface for a grid point, im calculating distance from the grid point to all the points on the body (in this case NACA 0012 airfoil) and using the minimum of the distance among them.

Is this the correct way to find distance from the closest surface?

[som87]

The way you are calculating the wall distance is highly dependent to your mesh.
If your have unstructured (say triangular) mesh, then it'll not give you correct answer.
If that is the case, I suggest to maintain the index and try to use the index (say j=1 to j=2) to get the grid distance

[jaiswal]

Hi Somdeb!... thank you for ur reply

I am not very clear what you are telling Somdeb. I am using structured mesh. I am using C-grid of 257*60 size. I have 197 points on the airfoil. Suppose i want to find closest distance from surface for a grid point say i = 30 & j = 20 then i find distance for this grid point from all the points on the airfoil i.e. j = 1 & i = 31 to 227 ( 197 points) and take minimum of the distance among them.

Is this the right way to find distance from the closest surface, for use in turbulence model calculations? Or is there a better way to do it?

[SergeAS]

Originally Posted by jaiswal

I am using C-grid of 257*60 size. I have 197 points on the airfoil. Suppose i want to find closest distance from surface for a grid point say i = 30 & j = 20 then i find distance for this grid point from all the points on the airfoil i.e. j = 1 & i = 31 to 227 ( 197 points) and take minimum of the distance among them.

Your algorithm is correct, but you keep in mind that for huge meshes and complicated wall shape this algorithm quite slow. For above specified mesh it will be insignificant, but if you want to propagate it to a larger 3D mesh, you will see this. Moreover, if you need to deform the mesh, this procedure will be repeated after each mesh modification. I use a similar algorithm in my 3D code but I have special parallel version for this case.

As example on 600k nodes mesh for ~1200 wall nodes this procedure takes 15 sec on 4-core CPU and less 5 sec on GPU

[som87]

Hi again, sorry for late reply.
Yes, your algorithm is correct.
About my previous comment, I didnt know about your grid's details

While we are in topic, SA turb model is a static type model right? So it will not give good result most cases

Cheers

[jaiswal]

Thank you Serge ... Now it is very clear to me... your information was very helpful... since i am calculating d correctly there is some other implementation issue... I will try to sort out...

[jaiswal]

Hi Somdeb!... thank u once again for replying...

Now it is clear to me that my algorithm to find distance is correct.

what do u mean by static type model?... I am still not successful in my implementation, so maybe it is important to me... can u pls explain?

[som87]

Hi again, I just commented on the fact that we marely calculate the eddy viscosity in SA model and there is no further coupling between the original NS PDE (like it is in k omega or k epsilon).
I am sure it will not affect your work in implementing SA model in your code. I didnt want to cause any misunderstanding

[jaiswal]

K... Thank you Somdeb