[mostanad]

Hello
I am very confused about wall distance definition in turbulence models for a grid in a domain. I attached a picture. Can anybody draw the wall distance from upper wall on it?
thank you

[FMDenaro]

You should consider the distance in normal direction to the wall. However, if the botton side is also a wall, it makes no sense to measure the distance from the upper wall.

[mostanad]

thank you Prof. Denaro
absolutely, it is true that for a node near the lower wall, I do not need to upper wall distance but I want to know how I can wall distance for an arbitrary node in a CFD code. As I found from your reply, Is the sketched distance in attached file true?
thanks a lot.

[arjun]

Originally Posted by mostanad

thank you Prof. Denaro
absolutely, it is true that for a node near the lower wall, I do not need to upper wall distance but I want to know how I can wall distance for an arbitrary node in a CFD code. As I found from your reply, Is the sketched distance in attached file true?
thanks a lot.

https://www.cfd-online.com/Wiki/Tran...ce_calculation

[FMDenaro]

My opinion is that if the upper wall ends as shown in the figure, the node is quite far from wall to define the distance along the normal direction to the wall. If conversely the wall continues you can trace the normal that intercepts the node.

[mostanad]

This is exactly true Prof. Denaro. But I want to calculate the distance of each node from all walls and compare them for finding the minimum value. As you mentioned, for shown node in the figure, the lower wall has the minimum value. So I want to find all of the distances.

[FMDenaro]

This is a classical geometrical issue: https://en.wikipedia.org/wiki/Distan...oint_to_a_line

[sbaffini]

Dear mohammad, for a given point in a fluid, the wall distance used in CFD codes is usually computed as the minimum distance between the given point and any of the points on the wall boundaries of the domain, irrespective of the angle formed between the wall and the line connecting the two points.

The method suggested by arjun, even if not geometrically exact, is among the most used as it is relatively simple to implement using elements already present in the code.

In contrast, exact geometrical methods become quite complex in parallel and I don't know a single classical code using it.

However, the main point to highlight is that such wall distance is basically only used in turbulence models and all the functions based on it saturate within few viscous distances from the wall, so exact calculation away from the wall (e.g., in the middle of the domain) is not even required. Moreover, where the distance is important it actually tends to be also along the normal to the wall.

[sbaffini]

Just to give a complete picture:

Differential equation based method
https://www.researchgate.net/publica...tial_Equations
http://enu.kz/repository/2010/AIAA-2010-1082.pdf

A possible geometric method
http://www.iccfd.org/iccfd7/assets/p...1204_paper.pdf

Note the complexity of the second one and consider that the parallel distribution of the surface mesh is not even considered as would make things even worst.

I programmed a simpler variant of the latter and I can confirm that there is no free lunch in geometric point-surface distance calculation.

[arjun]

Originally Posted by sbaffini

Just to give a complete picture:

Differential equation based method
https://www.researchgate.net/publica...tial_Equations
http://enu.kz/repository/2010/AIAA-2010-1082.pdf

A possible geometric method
http://www.iccfd.org/iccfd7/assets/p...1204_paper.pdf

Note the complexity of the second one and consider that the parallel distribution of the surface mesh is not even considered as would make things even worst.

I programmed a simpler variant of the latter and I can confirm that there is no free lunch in geometric point-surface distance calculation.

If he really clicked that link he would know that it presented to something that he really wanted to do ie to find the closest distance to wall.

That transport method also shows the direction of that wall. (wall distance vector). Even if he wants exact wall distance, he could do predictor corrector method, where transport eqn tells the approximate distance and the vector where that wall exists. He could then go in that direction and find the exact distance in much less efforts.

All it takes is to click the link. :-)

[sbaffini]

Originally Posted by arjun

All it takes is to click the link. :-)

Which is what I actually did (busy in boring stuff, in need of some distraction here) and then got into the geometric paper. I liked the idea of directly showing (vs just referencing) the difference in complexity among the two approaches... also, I want to reach 750 posts, so I'm also quite verbose these days

[mostanad]

Thank you, friends, for your replies
I have one method for calculation of wall distance and also direction. but I can not understand it. I found that this method marches along all nodes in the wall and find the maximum value of "inverse x (1/dsx) and y (1/dsy) distance from the wall". Then it obtains inverse(radical((1/dsx)^2+(1/dsy)^2)) and determines this value as the distance from the wall. My problem is in the inverse procedure an I believe this method has some problem. Is this method familiar for anybody?

[sbaffini]

Uhmmm... not really. From what you describe it looks like a brute force search with some fancy (and far from cheap) distance function. I actually use just the squared distance as it only involves two squares.

Might it be related to some hardcoding of your underlying grid? Maybe some algebraic parameter related to its definition?

EDIT: If you define a right triangle whose legs are respectively dsx and dsy, and the hypotenuse is the actual distance between the point and the wall, that formula actually defines the height (to the hypotenuse) of the triangle. Does this make sense to you?