[vangelis]

Hello everyone,

I am in the process of creating high aspect ratio layer elements for OpenFOAM and have been surprised by the sensitivity of the calculation of non-orthogonality in cases where the aspect ratio of the elements is high.
Consider the following example:
I have two perfect hexa layer elements of width 1000 and height 1, so an aspect ratio of 1000. All angles are 90deg.

In this case I have 0 non-orthogonality.

However if I begin to move one of the four common nodes along the height direction, then non-orthogonality increases at an amazing rate.

So for a movement of 1.5% of the height, non-orthogonality reaches 60
and for a movement of about 10% orthogonality reaches 85

This is an idealized case where all edge lengths are the same, angles are orthgonal etc.
In a real geometry where layers are extruded it is practically impossible to satisfy non-orthogonality criteria due to this extreme sensitivity.
Should it be this way?

Thanks
Vangelis

[deepsterblue]

It might be worthwhile to plot the centroids of cells on either side of the face in question (i.e., the values computed by OpenFOAM in primitiveMesh), to see if they actually make sense.

[vangelis]

Hi Sandeep
Thanks for your reply.
Indeed the calculated centroids move and the vector that connects them forms an angle with the normal of the common facets of the two hexas.
I was wondering however if the calculation of centroids is good for such cases.
Why should a movement in the normal direction, affect sideways the centroids of the two elements?

I mean these elements are near perfect to the eye after the nodel movement.
In the end it is impossible to create mesh with high aspect ratio that can satisfy orthogonality, in real cases.

[deepsterblue]

Are you using the OpenCFD version or OpenFOAM-extend? Might be useful to compare the two as well.

[vangelis]

I am using OpenCFD version but I believe it is the same for all.

[deepsterblue]

I had changed the centroid calculation a bit on the extend side, so I'm curious to see if there's a difference, which is why I mentioned it. If you look at primitiveMeshCellCentresAndVols.C, you'll notice there's a difference between versions.

[vangelis]

Thank you Sandeep we will look it up.
So in your custom version do you see the same sensitivity and tendency for the orthogonality to shoot up when making small such nodal movements?

[deepsterblue]

I've never tried a test like this to be honest, which is why I'm curious about your results. It looks like your face is essentially unchanged, so its normal / centroid would be unchanged as well. Which leaves the adjacent cell centroid as the only variable.

[be_inspired]

Any news about this topic? I am facing a similar problem where orthogonality reaches 89.5 but due to high aspect ratio cells with y+ less than 1.