What is the best element for CFD analysis? In Icem there is a HEXA-DOMINAT mesh, is it better than TETRA? This question because I have a very large number of node simulation and the HEXA DOMINANT reduce this number, but the accuracy of symulation is the same?

Thank's in advance for the answer

m.

If you use higher-order differencing and a proper grid independence study you will get equivalent results.

no !!! Tetra is very very bad !!! It is enough to test standard cases and you will see.

So you say that is better to use Hex mesh when possible, so is better the hex-core mesh in icem...

yeah , even it is well know fact that HEX solution(grid independent)requires less nodes than tri solution(grid independent)

Ok, it is easier to mesh of the complex forms with tetra element.And in this case, it is advised to activate the HR Scheme to avoid a strong numerical diffusion on this type of grid. But, one observes a behavior sometimes strange of the physical models on this type of grid (tetra). It is enough to compare solutions obtained with hexa and tetra mesh on the traditional cases which are forward step and backward step. dd.

Hi DD,

Can u suggest, when we have to go for HEX based on physics( like simple flow, flow with heat transfer, reacting flows and combustion and etc). When we shouldnt use TET. Why TET mesh is very bad? Where we to stop using TET? Do we have any literature to support these?

Regards Devolt

Do a search in the forum. I recalled there're quite some nice posts about this topic.

Hi TB,

Thanks for suggestion. I have gone through many posting on the same. I found that many people are claiming that, they r getting different solutions with Hex and TET. Some people are saying with finer mesh of TET (say of 8 times more elements than of HEX) they are able match with that of HEX solution. Important thing is that, it is not clear at what refinement; solution of TET will be matching with HEX one. If we are getting two different solutions with two meshes means at least one must be wrong solution. Let us assume that at point (location) with HEX mesh (10K elements) the temperature is some 1000 K and with TET(100K elements) is 1400 K ( All imbalances are almost zero in both cases). That means one of the solution is wrong. As many people are saying (Supported by literature) HEX is bad. So, how one can solve some solution with TET and can say that is the result. Regards

Devolt

Hi TB,

Thanks for suggestion. I have gone through many posting on the same. I found that many people are claiming that, they r getting different solutions with Hex and TET. Some people are saying with finer mesh of TET (say of 8 times more elements than of HEX) they are able match with that of HEX solution. Important thing is that, it is not clear at what refinement; solution of TET will be matching with HEX one. If we are getting two different solutions with two meshes means at least one must be wrong solution. Let us assume that at point (location) with HEX mesh (10K elements) the temperature is some 1000 K and with TET(100K elements) is 1400 K ( All imbalances are almost zero in both cases). That means one of the solution is wrong. As many people are saying (Supported by literature) TET is bad. So, how one can solve some solution with TET and can say that is the result. Regards

Devolt

In order to get grid independant results with a tet mesh you need to use many more elements and the solver typically takes longer to converge.

Hence, the general rule of thumb is to use hex meshes. However in many real world cases this is simply impractical, requiring the use of tet meshes. This is not a problem if you do a proper grid independance study and understand you have to pay for the meshing conveniance of a tet mesh with a longer solve time (due to more elements and the nature of the tet element discretisation).

Compare a hex vs tet solution for a simple 2D sliding plate example and you will quickly get a feel for the required differences in mesh size and solver time.

What is worrying with the tetra mesh. It is the behavior of the physical models, in particular the behavior of the Turbulence models. It is enough to test the models on the simple cases (backward, forward step). One notes things surprising (reattachment length ). The grid skew (very, very skew) hexa can lead to the same thing.What becomes worrying when one wants to make design. dd

I think many models (especially empirical models) developed based on 2D structured (HEXA) mesh. People who proposed might have tested them on 2D mesh against experimental data. That might be also one reason for not matching with 3D simulations with TET of real geometry that of hexa. But I believe one of the meshing methodologies (hexa ,tet) should be discarded or at least there should be is predefined parameterization giving guide lines where(based on physics) we can use what, and should be the mesh density and etc

I think seriously that it is necessary to be careful with the physical models on tetra grid. In particular if there is a strong deformation of the flow. And even if the grid is refined, the solution is very far from that obtained on hexa grid. It is enough to test a simple case where a strong vorticity or shear strain rate exists. dd.

I don't think you need to worry about the grid topology to the extent that you have been. First of all, the control volumes in CFX are neither hex nor tet (or prism or pyramids for that matter). Control volumes are the mesh dual of your grid, essentially giving arbitrary polyhedrals.

Most physical models, including turbulence, are developed at the continuum level, not at the grid level. As long as they are implemented consitently, it doesn't matter what grid type you have.

That said, factors such as grid resolution and quality will have a real effect on your solution. Which grid topoly will give you the best resolution will depend on what you are meshing. For instance, it makes good sense to use hexehedral or prismatic elements near the wall. Not because these are better elements, but because they allow you to maintain good element quality while packing several layers of mesh near the wall with high aspect ratio's. You could probably get as good a solution with tet elements, but in order to maintain the grid quality and still have 5 to 20 nodes within the boundary layer you would need millions of elements.

Hexehedral or swept topologies are useful where you have small gaps or elongated channels, since you can stretch the elements in one direction. However, if the flow in these regions is highly three dimensional, you'll need nearly isotropic mesh anyway, so this may be a moot point.

If you have a very complicated geometry, you may be better off with a good quality unstructured mesh (usually tet + prisms, but possibly hex and pyramid as well). While it may be possible to generate a purely hexehedral mesh on very complicated geometry (particularly with a tool like Hexa), it is neither efficient nor advisable. If you do manage to get a hex mesh, the odds are pretty good that you'll have poor element quality in many areas due to striping of the block structure. What's worse is that where the poor quality elements occur in a hex mesh, there are often several elements of poor quality in the same area, exacerbating the errors.

While the unstrucutred mesh may require more memory and take a little longer to run, it can often be generated much faster (and, let's face it, manual mesh generation time costs a lot more than batch solver time), be set up to give more resoltion in approriate areas, and if there are any poor quality elements, they are usually isolated and have little influence on the solution.

Arguments stating that models are not appropriate on a given mesh topoly are without merit. The only way to evaluate the accuracy of a solution is to do a proper evaluation of the available models and perform mesh independance studies on the same class of problem. All the while ensuring that you put the appropriate mesh where you need it. An expensive proposition, but a reality of the situation.

The question of which mesh topology to use should only relate to which one can allow you to work efficiently while acheiving an appropirate solution and this can only be found through validation.

Regards, Robin

Robin, yes, yes i agree with what you say. But, the k-ep (and other physcical model)gives strange solutions on tetra mesh with HR scheme. You can test it on a backward step... dd

what's your strategy to define the grid error for the domain with hex & tet elements? I find that mesh dependency test is difficult to implement consistently, and makes little sense for problem with different type of elements in a flow domain.

DD and TB,

"Strange" solutions can be obtained on any problem with any mesh type, it all depends on how well you have created your mesh. You can obtain reasonble solutions on a backward facing step with a hybrid mesh, although I would not condone doing it entirely with tetrahedrons. Tet's are not meant to be used in isolation, you should always use prisms near the wall. For a backward facing step, I have no doubt that a more efficient solution would be to use hexehedrals, but engineers are not interested in backward facing steps and in a more complicated geometry it may not be worth your time to make an efficient hex mesh, nor will it necessarily be the most efficient method.

As for grid error, you need to refine the mesh in a consistant manner and compare solutions. This goes the same for tet's or hexes.

Regards, Robin

Thank you. You confirm my suppose.

Regards,

Max

This comment is fairly crude as well....

Balancing the time & accuracy needed for a calculation is heavily dependent on user experience. You may have to spend a lot more time later to analyse what is going wrong, if the mesh you use can't give you the required accuracy.

I used to have the convergence problem with tet+prism mesh generated in Build (Not all cases). When I switch to hex mesh in ICEM, problem disappears. You can't argue meshing is not a problem there. That's why I can hardly be conviced when one tell me that tet+prism mesh should be used whenever possible. I have wasted lots of time there before. So I will say it's problem dependent. If tet+prism mesh doesn't give you what you need, you should give hex mesh a go. It could save you lots of effort later by investing some times in meshing...