[dtmith]

Hello all,

I'm wondering if there's some way within Ansys post to smooth out isosurfaces? It seems to like lining up the faces of the isosurface with the mesh, so if I use a tetra mesh the isosurface is rough, if I use a hexa mesh it's smoother but you can visibly see the lines used to make up the mesh.

When plotting a 2D slice surface I can specify "sample" to interpolate the results on to an evenly spaced grid which works well, but there doesn't seem to be anything similar for isosurfaces.

I understand a possible solution is "run a finer mesh", I have tried a finer mesh and the results are similar enough that I don't want to waste the computing power running the parameter sweeps with the finer mesh.

Thanks in advance for any assistance!

[ghorrocks]

What variable are you trying to plot a isosurface of? The isosurface just follows the variable contour so if the variable is lumpy then the isosurface will be lumpy. In a good quality simulation the variable fields are smooth and does not follow the mesh lines.

[dtmith]

So you wouldn't expect a tetrahedral mesh to produce rougher looking isosurfaces than a hexahedral mesh assuming the physics are being accurately captured?

I was plotting Q-criterion and Vorticity.

I didn't think the solution was lumpy because plotting planes of vorticity using flat shading and "sample" looked quite good. But looking at it again if I look closely on the planes I can see a bit of lumpiness there as well, but it seems the isosurface lighting/shading exaggerates the lumpiness more than the plane.

So if I want nicer isosurfaces I just need a finer mesh? I'm reasonably happy that the physics I'm interested in and on the scale I'm interested in are being well resolved at the moment.

I'm currently using LES, running sims with both tets and hexes on the same geometry (the hex ones are actually hex in the near field and tets in the far field), about ~5 million nodes. I've tried running up to about 7 mill nodes on the tet mesh and about 11 million nodes on the hex mesh but time averaged forces are within 1% for all cases and the overall strength/position of the vortex structures is extremely close, the only noticeable difference is the texture of the isosurface plots, so I was intending to run the lower resolution for my parameter sweeps.

Running the finer mesh I still see the mesh grid on the isosurfaces, but since the mesh elements are smaller I have to zoom in further before it's an annoyance.

[ghorrocks]

So you wouldn't expect a tetrahedral mesh to produce rougher looking isosurfaces than a hexahedral mesh assuming the physics are being accurately captured?

Correct. An accurate simulation (usually) means a smooth variable field regardless of what mesh is used.

So if I want nicer isosurfaces I just need a finer mesh?

Sometimes that can help. Certainly if your mesh is way too coarse you will get lots of lumpyness in the post processing. But it can also be cause by numerical instabilities. And sometimes it is real.

If you are plotting q-criterion and vorticity - note that these are a derivative of the simulated variables and are therefore noise will be amplified. This means it will be tricky to make these plots smooth. (integrated functions are, on the other hand, less noisy and tend to reduce noise).

[singer1812]

If this is for visual only, and you dont want to try creating a smoother mesh at this time, one trick you can do is create a user surface off your iso surface and then try plotting the variable on the user surface.

[dtmith]

Glenn, you are correct that the U,V,W components look much better, it's the derivative terms where the lumpiness is noticeable.

It is also worse when using central differencing vs specified blend, but specified blend is noticeably damping things.

I don't particularly want to create a finer mesh given the fluctuations in Q and Vorticity are a couple of orders magnitude smaller than that of the vortices themselves and everything else seems well converged (or at least insensitive to changes in mesh/timestep of the order I can solve in a realistic time on my current computer) and it also seems to match well with stereo PIV experiments I've done previously (though the SPIV stuff is much lower resolution).

I don't actually think it's too bad, I'm running Re1400 and the vortices seem will resolved, it's just I'm zooming in to them far enough that the much smaller oscillations due to the mesh are visible.

It's a bit like this GIF where you can see the underlying mesh on the isosurface (just took this link off the ansys site, it's not actually my sim, if anything my sim doesn't look quite so bad!)

http://www.ansys.com/staticassets/AN...ena-bg-ani.gif

Originally Posted by singer1812

If this is for visual only, and you dont want to try creating a smoother mesh at this time, one trick you can do is create a user surface off your iso surface and then try plotting the variable on the user surface.

That doesn't seem to work (unless I'm doing it wrong) because the lumpiness appears on the surface itself so the the user surface just has the same lumpiness.

Certain viewing angles it's not apparent, it's just when the lighting reflects off the isosurface in a certain way.

Because the lumpiness is much smaller than any of the actual vortex structures I would be tempted to just run a low pass filter over Q and vorticity (though it does sound a bit dodgy!), however I don't think there's any way to do that through CFX itself anyway?

[ghorrocks]

You might be able to construct a low pass filter using the variable and the variable gradients. I am not sure about it - I will it up to you to do the thinking to determine if that is going to work.