[Anna Tian]

Hi,

The structured mesh is strongly suggested to be used in CFD to have better accuracy in my school. So I never used unstructured mesh. But when I come to this forum and search for the papers and mesh pictures online, I found that a lot of people actually use unstructured mesh and also cartesian mesh.

I'm wondering when(for what flow) can I use the unstructured mesh and cartesian mesh? How to check how reliable are they besides comparing with the structured mesh simulation results?

[majidmng]

Hi
structure mesh generation is hard and for a complex geometry it is very difficult. boundary layer mesh and structure is professional gird and anyone dont work it. unstructure grid is low accuracy and result of analys not validate but it is useful !
what is your geometry?

[lovecraft22]

I guess it depends also on the type of software. For instance PowerFLOW from Exa uses structure grids…

[Anna Tian]

Originally Posted by majidmng

Hi
structure mesh generation is hard and for a complex geometry it is very difficult. boundary layer mesh and structure is professional gird and anyone dont work it. unstructure grid is low accuracy and result of analys not validate but it is useful !
what is your geometry?

But you know sometimes the unstructured grids can give very unreasonable prediction results. It is very hard to control the unstructured grids prediction accuracy and convergence might be very difficult.

I can understand that some persons might still choose unstructured grids for simplicity. The question is how do they know how inaccurate is it before they use it? Any experience or theory to tell when tetra grids will give good prediction results? When I see pictures online that tetra grids are even generated for turbo 3D modeling and cartesian grids are used for car aerodynamics modeling, it's kind of hard for me to trust the result to a certain level.

I don't have a certain geometry to model. This is a general question.

[Anna Tian]

Originally Posted by lovecraft22

I guess it depends also on the type of software. For instance PowerFLOW from Exa uses structure grids…

Some softwares have both options. The tetra grids generation is always much simpler. I never tried unstructured grids. The question is: how do you get a clue of the unstructured grids simulation accuracy before you run the simulations?

[majidmng]

for exmple I generate unstructured grid on wing,results of it was aproximate %30 diffrence with experimental data, with structured gird results was below %10 diffrence with experimental data. i work in private company and for commerical geomtry use unstructured mesh beacuse it is easy and results not bad...but for acadmic research and high accuracy result structured gird is need.

[cfdnewbie]

Originally Posted by majidmng

but for acadmic research and high accuracy result structured gird is need.

Hello,
this statement is not generally true. In fact, there are a number of high accuracy research codes that are fully unstructured. Check out Nektran or Nek5000. Both have spectral accuracy on unstructured grids. Almost all FE schemes are unstructured, and their dirty cousins, the high order discontinuous Galerkins, as well. All high of those order or spectrally accurate. So especially in research, many many of the currently interesting codes are unstructured.

[Anna Tian]

Originally Posted by majidmng

for exmple I generate unstructured grid on wing,results of it was aproximate %30 diffrence with experimental data, with structured gird results was below %10 diffrence with experimental data. i work in private company and for commerical geomtry use unstructured mesh beacuse it is easy and results not bad...but for acadmic research and high accuracy result structured gird is need.

I'm wondering, generally to say, for what kind of flow will tetra grids give not too unreasonable results? Is there any way to know or judge that before running the simulation and comparing with the experimental data? Any theory or experience on that?

By the way, you work for aircraft wing? If your inaccuracy is 30%, how do you use this prediction for design? The difference between 2 different designs is usually smaller than 30%, right? How do you compare their performances?

[lovecraft22]

Originally Posted by Anna Tian

If your inaccuracy is 30%, how do you use this prediction for design? The difference between 2 different designs is usually smaller than 30%, right? How do you compare their performances?

I agree. 30% is pretty far in terms of accuracy. Probably you'll can catch the trends anyway but still…

[Anna Tian]

Originally Posted by lovecraft22

I agree. 30% is pretty far in terms of accuracy. Probably you'll can catch the trends anyway but still…

For what flow will tetra grids give good prediction?

[lovecraft22]

Unfortunately I don't know the answer.

I guess what I said above is not completely true though.

Let's have a look at these grids:

This one, is what is generally denoted as an unstructured mesh



This one is instead a structured mesh:


Now, what would you call this one?


According to the definition, it should be an unstructured one but still it doesn't look to me as the same thing as the first picture.

Thanks!

[cfdnewbie]

by definition, it is an unstructured mesh, as it needs to be treated internally as an unstructured data structure and as there are no lines of constant I,J,K.

[JasonWu]

Generally speaking, unstructured tetrahedral mesh by commercial software is for people who knows nothing or only little about mesh generation, CFD methods, and numerical methods, while structured hexhedral mesh is for people who knows much more about mesh generation, cfd methods and numerical methods.

For the difference between structured hexhedral grid and the unstructured tetrahedral mesh, please see the below:

http://www.advanceddo.com/mesh_generation.html

[cfdnewbie]

Originally Posted by JasonWu

Generally speaking, unstructured tetrahedral mesh by commercial software is for people who knows nothing or only little about mesh generation, CFD methods, and numerical methods, while structured hexhedral mesh is for people who knows much more about mesh generation, cfd methods and numerical methods.

For the difference between structured hexhedral grid and the unstructured tetrahedral mesh, please see the below:

http://www.advanceddo.com/mesh_generation.html

I find this statement to be overly broad. How would you classify unstructured hex meshes then? You need to know a lot more about numerical properties, meshing and programming to design an unstructured code than a structured one...

[ghost82]

Hi,
As I know structured hexa mesh is very good when you have the flow aligned with your mesh.
As you can understand when you have tetra meshes the flow will never be aligned with the mesh.

Daniele

[Anna Tian]

Originally Posted by lovecraft22

Unfortunately I don't know the answer.

I guess what I said above is not completely true though.

Let's have a look at these grids:

This one, is what is generally denoted as an unstructured mesh



This one is instead a structured mesh:


Now, what would you call this one?


According to the definition, it should be an unstructured one but still it doesn't look to me as the same thing as the first picture.

Thanks!

The third one is cartesian grids. That shall not be classified to the structured grids.

[Anna Tian]

Originally Posted by ghost82

Hi,
As I know structured hexa mesh is very good when you have the flow aligned with your mesh.
As you can understand when you have tetra meshes the flow will never be aligned with the mesh.

Daniele

So in what case will tetra grids give good prediction results compared with structured grids? Very very complex flow?

[Anna Tian]

Originally Posted by JasonWu

Generally speaking, unstructured tetrahedral mesh by commercial software is for people who knows nothing or only little about mesh generation, CFD methods, and numerical methods, while structured hexhedral mesh is for people who knows much more about mesh generation, cfd methods and numerical methods.

For the difference between structured hexhedral grid and the unstructured tetrahedral mesh, please see the below:

http://www.advanceddo.com/mesh_generation.html

Okay. But when I look at the papers in the engineering field, like vehicle aerodynamics engineering, there are quite some papers researching on the numerical engineering design method by using tetra grids. Some of them even discusses the geometry optimization by using tetra or cartesian grids. And at the end of the paper, they even conclude that the numerical designing method can be well applied to that field because they have good comparison with experiment. This is kind of surprise to me because I was always told that only structured grids give good CFD prediction. So I'm interested in knowing more about unstructured grids. I'm wondering when could we trust the results from unstructured grids?

[cfdnewbie]

What is totally confused in this discussion is element type vs. mesh type. Element types make the quality of the mesh (and any interpolation / prolongation between nonconforminng ones), NOT the way the mesh is stored in a data structure. Think about an unstructured hex mesh!
Structured meshes are ok for simple geometries, or mildly complicated ones, but beyond that, meshing becomes a nightmare and you waste too many DOF in regions where you don't need them. Look at the airfoil example posted in this thread....

[cfdivan]

Greetings all,

By definition, it could be a structured cartesian mesh...Cartesians are just a particular case of structured meshes, among others, block structured, H- O- C- type, isn't it?

Regards,