Meshing in AcuConsole vs HW CFD X

Junta_20361

Hello experts,
I have tried using both AcuConsole and HW CFD X.

Both of them have their own pros and cons.

However, i found that surface meshing in HW CFD X seemed better than AcuConsole at some cases.

While volume meshing with AcuConsole seemed better (with the hybrid meshing - octree based mesh).

I have a questions related to surface meshing in AcuConsole as below:

2 above pictures are the meshing made by AcuConsole (left) and HW CFD X (right).
you can see that the mesh around the circle by AcuConsole look chaotic and not uniform/smooth as in HW CFD X.

May i ask is there any settings to be made in AcuConsole to generate the mesh looks like HW CFD X.

Thanks in advance!
 

 

QuyNguyenDai

Numerically, both meshes can give the 'same' results.

The 2nd is 'nice' to see, but nothing is guaranteed his numerical result is better than 1st one.

ydigit

Correctly mentioned above, that both will very likely give the 'same' results. 

 

Also, AcuSolve is a node based solver. Hence the way these nodes are connected is secondary for the solver, important is to have enough nodes in the region of interest. I am not aware of any setting in AcuConsole to get the 'nice' looking mesh. 

Junta_20361

Altair Forum User said:

Numerically, both meshes can give the 'same' results.

The 2nd is 'nice' to see, but nothing is guaranteed his numerical result is better than 1st one.

 

Altair Forum User said:

Correctly mentioned above, that both will very likely give the 'same' results. 

 

Also, AcuSolve is a node based solver. Hence the way these nodes are connected is secondary for the solver, important is to have enough nodes in the region of interest. I am not aware of any setting in AcuConsole to get the 'nice' looking mesh. 

thank you both for your answers.

i understand that AcuSolve just cares about the number of nodes, even with very bad quality mesh.
my questions is just for studying purpose.

QuyNguyenDai

In the Numerical World, ther most important thing is the Quality of your results, not the 'Beauty' of your mesh /emoticons/default_biggrin.png' srcset='/emoticons/biggrin@2x.png 2x' title=':D' width='20' /> 

acupro

As stated before, AcuSolve still requires enough mesh (nodes and/or elements) to resolve the relevant flow features and gradients.  It's just that AcuSolve is more forgiving of individual element quality than some other codes.  So the mesh still needs to be 'good', but the individual element requirements are much less stringent.

cfdguru

Here's a bit more detail on why the two meshes are different.  HW-CFD uses an advancing front meshing technique which will cause the mesh patterns to radially grow away from the edges.  AcuConsole is using a Delaunay triangulation approach.  So, you won't see the radial patterns in the mesh.

Also, your original comment about volume meshing caught my attention.  What is it about the AcuConsole volume mesh that you feel is superior to the HW-CFD volume mesh?

Junta_20361

Altair Forum User said:

Here's a bit more detail on why the two meshes are different.  HW-CFD uses an advancing front meshing technique which will cause the mesh patterns to radially grow away from the edges.  AcuConsole is using a Delaunay triangulation approach.  So, you won't see the radial patterns in the mesh.

Also, your original comment about volume meshing caught my attention.  What is it about the AcuConsole volume mesh that you feel is superior to the HW-CFD volume mesh?

Dear cfdguru,
Thanks for your detail explanation.

About the volume meshing, AcuConsole use hybrid technique which is combination of advancing front, delaunay and octree mesh (from AcuConsole Mesh Manual). This create an structured mesh in the tetra core.

Please refer to below images to see the difference:
AcuConsole

HW CFD / HyperMesh:

cfdguru

Right.  AcuConsole's mesher will create octree patterns in large open regions.  That mesher works as follows:

1.)  Create surface mesh using the delaunay triangulation approach
2.)  Extrude boundary layers from the surface into the volume regions
3.)  Build octree mesh in volume regions
4.)  Connect the surface/boundary layer mesh to the octree regions using delaunay or advancing front technique.

However, the question is what you feel is superior about the octree approach.  The structured nature of an octree mesh is not beneficial to AcuSolve.  The randomness of the nodal positions in the HW-CFD mesh tends to make the linear algebra less stiff and can actually lead to faster run times on meshes of similar node count.  Additionally, the 2:1 size jumps that are present in the octree mesh can cause a stiff linear system in AcuSolve as well.  Overall, you should see that the HW-CFD mesh performs better in the solver, and is also faster to generate due to the more effective use of mult-threading.  I'd be interested to hear if your experience is not consistent with these expectations.  

Nazrin

Hi @cfdguru, sorry for interrupting.
 

Altair Forum User said:

The structured nature of an octree mesh is not beneficial to AcuSolve.  The randomness of the nodal positions in the HW-CFD mesh tends to make the linear algebra less stiff and can actually lead to faster run times on meshes of similar node count.

Is this mean that advancing front meshing technique (HW-CFD) is numerically better than Delaunay triangulation (AcuConsole)? So, back to meshing case above, 'beautiful or radial growth'  mesh does effect the simulation runtime?

cfdguru

Hi @Nadzrin,

    The advancing front vs. Delaunay triangulation discussion actually refers to the surface meshing.  AcuSolve doesn't have much of a preference either way when it comes to surface meshing.  The runtime argument I made is in reference to the volume meshing (AcuConsole's octree vs HW-CFD's approach).  In this case, the linear solver *should* perform better with the mesh from HW-CFD.  Note that the performance won't be radically different between the two approaches, but according to our linear solver experts on the AcuSolve team, the more random mesh is beneficial.  Would be an interesting test for someone to run. and report the results!