[jaxk]

Hi all,

I am trying to make a structured mesh for an oval-shaped body. So far it looks good (attachment 1), but only if the oval has a certain angle. If I change the angle a few degrees the structure is still there, but it results in highly skewed elements (attachment 2). After a certain angle the kind of structure I want is no longer there (attachment 3).

Does anyone have some tips on how to keep the structure I want? Can I add some guidelines like the red line in attachment 4 without splitting the green edge? I tried splitting the body into two using the red line, but the green edge should remain one edge since I use Edge Sizing with a certain number of divisions there.

Thanks!

[jaxk]

I added an attachment to give some insight in the meshing tools I used.

[jaxk]

In the end I'll go with something like this (attachment r2).

Another problem I have now: the curve is not really round (attachment r3). Does anyone know how that can be fixed? Setting Capture Curvature to yes, with a small angle doesn't work...

[ghorrocks]

What are you trying to do with this analysis? What physics are you modelling? Have you done a mesh sensitivity study to work out what size mesh you need for the accuracy you want?

You need answers to those 3 questions to determine what mesh is required for your model.

[jaxk]

Yes, I will do a mesh sensitivity analysis, but I do need a working mesh structure first I believe. I will measure the drag and torque coefficient depending on the angle of the blade, at high reynolds number (Re= 1e6). A 2D approximation at Steady state. And possibly transient and/or a longer z-distance if that gives other values.

[ghorrocks]

Yes, you do need a working mesh for a mesh sensitivity analysis. But you should only need to do this at one angle of attack, so you do not have to make it work over a range of AOAs.

Also - What are you trying to do with this analysis? What physics are you modelling?

[jaxk]

Thanks for the reply. I intended to do the mesh sensitivity analysis at one angle, I chose 90 degrees, so vertical. But then even a super coarse mesh resulted in the expected drag coefficient and torque coefficient (0). However, with that mesh at an AoA of 15 degrees, it resulted in very different torque coefficients compared to a finer mesh. So now I intend to do the mesh sensitivity analysis at 15 degrees only. (I chose 15 since I expect a peak in torque coefficient there, but I think another angle could work as well, apart from 0 (horizontal) or 90 degrees).

I'm analyzing water flowing around a blade at high Reynolds number, and want to see how the torque coefficient and drag coefficient change with the AoA of a blade.

I am using an SST turbulence model with 5% turbulence.

To find the influence of the Reynolds number, I will also so a run with 1e4, 1e5, 1e7 to see if that changes the results.
To find the influence of the turbulence on the results, I will also do a run with 2,5% and 10% turbulence.

I hope that answers your question and allows you to give me more feedback, which is very much appreciated!

[ghorrocks]

Thanks, yes that explains what you are doing.

I would:
* Put a large inflation from the body, starting with y+ = ~1 and with a gentle expansion ration, 1.1 or 1.05 even better. Go at least 8 elements out, but around 20 would be better.
* I would forget all the blocking you have done as I think the poor orthogonality is going to cause problems.
* Put a block in the area you expect the wake if you want to capture the wake accurately.
* And everything else just mesh with a hex or tri paver mesh.

If this is a steady state run you can use a big mesh as it will still converge quickly. You might need to be more careful with your mesh settings for transient simulations to keep it realistic. So do the mesh sensitivity on a steady state run and work out exactly what mesh you need before you start the transient simulations.

[jaxk]

Is this the kind of mesh structure you propose, using fine hexahedrons within the block?
I have been using the 4 diagonals to get a structured mesh with fine elements near the blade and a more coarse mesh away from the blade.

I have been using the kind of inflation layer you mention, with 20 layers, and a y+ of ~1.

In which region do you expect poor orthogonality that could cause problems?

[jaxk]

Or more something like this, which is less structured

[ghorrocks]

The less structured mesh looks much better to me. Almost much easier to generate, and should be able to handle big changes in AOA with no problems. I would do the mesh sensitivity study on that approach.

[jaxk]

Have you ever come across this behavior near the inflation layer with a 2D approximation? The sweep method doesn't project exactly along the z-axis. As a result, the first layer thickness isn't everywhere the same.

[jaxk]

I solved it by using 'Automatic Thin' within the Sweep Method, instead of 'Manual Source and Target'.
Somehow that also results in a more structured mesh overall.

However, the inflation layer is not possible with this approach...

[ghorrocks]

You will just have to try things to try to work it out. I find ANSYS Mesher a frustrating mesher as it does unexpected things (like this).

I would check your geometry, the faceting I can see means that there might be something happening there.

Also, you have a big jump in mesh element sizing from the end of your inflation to the bulk mesh. It is better to keep growing your inflation mesh until it reaches the same size as the bulk mesh and you then get a nice even transition. So I would put more layers in your inflation mesh.