CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Influence of mesh refinement on the convergence (

saisanthoshm88 November 24, 2010 08:14

Influence of mesh refinement on the convergence
3 Attachment(s)
I was making a quad mesh by blocking approach, to a 2D planar cylinder geometry (Cylinder Diameter = 0.1 m). I wanted to have a denser mesh region over the cylinder so as to have the Boundary layer captured well. Iíve created such denser region around the cylinder in two ways. In the first case Iíve used the curve mesh set up for the purpose wherein Iíve given the first layer height (= 0.003 mm) and height ratio (= 1.1). In the second case, the density region was created just by adjusting the number of nodes on the relevant edges and without indeed resorting to any curve mesh settings. The mesh law adopted in both the cases was Bi Geometric and the boundary conditions imposed for both the cases were exactly identical. The convergence in the second case was very good and in the first case the solver struggled a lot and even the convergence was so poor. I see that in the first case, the density region was more uniformly distributed than that in the second case but in spite of that, the first case has a problem with the convergence. Someone please provide a clarification to this. The respective attached images can throw a better illustration of the problem described.

ghorrocks November 24, 2010 17:52

Mesh quality affects convergence. Obviously the mesh quality in your case which did not converge was not very good. Can you describe what the images show?

saisanthoshm88 November 25, 2010 00:54

The images show the density region around the cylinder in the mesh. The first image shows the block topology adopted. The images were named as: Blocking.jpg, Firstcase.jpg and Secondcase.jpg to refer to the respective instances.

The mesh quality in both the cases was good and it was around 0.7. I actually found that there is a problem for the first case to converge in spite of having a uniform density distribution over the cylinder.
But in the second case though there is a relatively non uniform density distribution over the cylinder ( as shown by the images ) the solution converged of course in the second case I've also checked for the results in the post processing like the Friction drag, Pressure drag and the total drag over the cylinder and found them to be matching well with some theoretical calculations.

ghorrocks November 25, 2010 01:29

There are many measures of mesh quality. Aspect ratio, cell volume ratio, edge length ratio, orthogonality and many others. A single number does not fully describe mesh quality.

What do you mean by a "non uniform density distribution"? Non-uniform in what fashion?

saisanthoshm88 November 25, 2010 02:43

OK now I realize that the problem is something different. I ran the simulation just now trying out something else and that helped me gain some focus on the very problem.

Its been found that the convergence of the solution highly depends upon the mesh spacing (spacing 1 and spacing 2 in the premesh params) over the cylinder . I mean the first layer height over the cylinder.

It some how converged well when spacing1=spacing2=0.05 mm and there is a problem when this value was altered.

I've tried out the problem by using different values for this first layer height like : 0.1, 0.01,0.003 mm but wasn't able to achieve proper convergence. Some how the convergence was possible only for the value of 0.05 mm

What can be the reason for this?

The height ratio for all the cases is 1.1

Can you please guide me on setting an appropriate first layer height over the cylinder so as to have a better convergence.

ghorrocks November 25, 2010 07:53

For good resolution of boundary layers like this, try to make the edge length of adjacent elements within 5% or 10% of each other. For accurate models you need a more even mesh.

saisanthoshm88 November 26, 2010 07:58

Thanks for the suggestion Glenn , it works.

All times are GMT -4. The time now is 12:21.