Optimized mesh refinement

PierreLCV · January 4, 2024, 14:19

Dear CFD users,

I've been working in CFD for several months now, and I'm wondering about a question to which I don't have a precise answer. Indeed, one of the essential steps in a CFD study is, of course, mesh refinement, with the aim of arriving at a solution that is as independent of the mesh as possible. I understand the idea: refine again and again until convergence of well-chosen quantities is achieved. However, in practice, I've come to realize that there are many ways to refine a mesh. Taking, for example, the case of a 3D mesh created with Ansys Fluent (but this is true for any mesher), how do you refine efficiently? In other words, is it better to refine the surface mesh, the thickness of the first layer of the boundary layer, the volume mesh?

I understand that the idea is to refine where necessary (i.e. where velocity and temperature gradients are such that refinement is required). However, I was wondering if, beyond the much-vaunted feeling and experience of CFD engineers, there isn't an optimized method for efficiently refining a mesh? In other words, how to obtain (not necessarily for sure, but perhaps to increase the chances) a mesh that offers precision without being too heavy? How can I avoid over-refining one aspect and under-refining another? I'm sure I'm not the first person to ask myself this question, and perhaps there are mathematical or other tools available to optimize this quest for the "best" mesh?

Thank you very much for your help!
Pierre

LuckyTran · January 4, 2024, 19:20

Call it what it is. There is no problem of mesh refinement, it is simply meshing. If you generate a good mesh the first time, you never need to refine it.

This question is a great example of engineers wanting to optimize things forever. Avoid this trap. Ask not how to waste your time most efficiently. Avoid wasting time altogether.

How do I refine efficiently? I don't.

PierreLCV · January 5, 2024, 04:02

Dear LuckyTran,
Thank you very much for your quick reply !
I have to admit I'm a bit taken aback by your answer. Until now, I've more or less tried to follow a method to achieve a good mesh:
1) Make an initial coarse mesh to identify the general flow pattern and pinpoint the high-gradient areas to be refined as a priority.
2) Refine using the information gathered from the results of the first mesh AND using your instincts (you could call them your "physical sense").
3) Repeat until the results I'm interested in converge.

However, now you're telling me that the refinement process isn't a necessary step and that it's enough to find the right mesh right away... I must admit I'm a bit puzzled... How can we find the right mesh and not need to refine it if we have no experimental results to compare?

Thanks for your help!
Pierre

LuckyTran · January 5, 2024, 18:59

Consider this. Take a bite out of a sandwich. Now take another bite. Do you call the second bite a sandwich eating "refinement?" No. Both are just steps towards eating the sandwich. You are creating a new issue about the same problem. Mesh refinement is taking that second bite and pretending that it is somehow different than the first bite towards eating the entire sandwich. You are not done eating the sandwich until you stop taking bites and you are not done meshing until you have the mesh that you want. There is no need to create a field of study of how to take the second bite out of a sandwich. Your first bite simply wasn't big enough to eat the entire sandwich.

Not everything in life needs to be a science. Many things can be left as art. No one will receive a nobel prize on how to optimally eat a sandwich. Just eat the darn thing. Just mesh it.

PierreLCV · January 8, 2024, 03:54

Dear LuckyTran,
Thank you very much for your interesting and original answer ! Indeed, when you look at it that way, it's understandable.
But given that mesh refinement is an integral part of what we call meshing, what is often the best strategy to follow in general to obtain the most accurate mesh at the lowest cost? Of course, I'm aware that there's a certain amount of "art" involved in this search for the best mesh, but I think that some general rules are still relevant...

Thank you !
Best regards,
Pierre

January 4, 2024, 14:19	Optimized mesh refinement	#1
PierreLCV Member Pierre LECHEVALIER Join Date: Mar 2023 Posts: 31 Rep Power: 3	Dear CFD users, I've been working in CFD for several months now, and I'm wondering about a question to which I don't have a precise answer. Indeed, one of the essential steps in a CFD study is, of course, mesh refinement, with the aim of arriving at a solution that is as independent of the mesh as possible. I understand the idea: refine again and again until convergence of well-chosen quantities is achieved. However, in practice, I've come to realize that there are many ways to refine a mesh. Taking, for example, the case of a 3D mesh created with Ansys Fluent (but this is true for any mesher), how do you refine efficiently? In other words, is it better to refine the surface mesh, the thickness of the first layer of the boundary layer, the volume mesh? I understand that the idea is to refine where necessary (i.e. where velocity and temperature gradients are such that refinement is required). However, I was wondering if, beyond the much-vaunted feeling and experience of CFD engineers, there isn't an optimized method for efficiently refining a mesh? In other words, how to obtain (not necessarily for sure, but perhaps to increase the chances) a mesh that offers precision without being too heavy? How can I avoid over-refining one aspect and under-refining another? I'm sure I'm not the first person to ask myself this question, and perhaps there are mathematical or other tools available to optimize this quest for the "best" mesh? Thank you very much for your help! Pierre

January 5, 2024, 18:59		#4
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,751 Rep Power: 66	Consider this. Take a bite out of a sandwich. Now take another bite. Do you call the second bite a sandwich eating "refinement?" No. Both are just steps towards eating the sandwich. You are creating a new issue about the same problem. Mesh refinement is taking that second bite and pretending that it is somehow different than the first bite towards eating the entire sandwich. You are not done eating the sandwich until you stop taking bites and you are not done meshing until you have the mesh that you want. There is no need to create a field of study of how to take the second bite out of a sandwich. Your first bite simply wasn't big enough to eat the entire sandwich. Not everything in life needs to be a science. Many things can be left as art. No one will receive a nobel prize on how to optimally eat a sandwich. Just eat the darn thing. Just mesh it. SARALL61 likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[snappyHexMesh] Celllevel does not satisfy 2:1 constraint	zippy	OpenFOAM Meshing & Mesh Conversion	1	July 30, 2024 06:58
[snappyHexMesh] snappyHexMesh sticking point	natty_king	OpenFOAM Meshing & Mesh Conversion	11	February 20, 2024 10:12
[snappyHexMesh] Sphere tutorial - refined mesh fails	nikosb	OpenFOAM Meshing & Mesh Conversion	0	March 17, 2022 16:49
[snappyHexMesh] Edge Refinement	fracasce	OpenFOAM Meshing & Mesh Conversion	3	December 2, 2017 14:30
[snappyHexMesh] SnappyHexMesh for internal Flow	vishwa	OpenFOAM Meshing & Mesh Conversion	24	June 27, 2016 09:54

January 4, 2024, 19:20		#2
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,751 Rep Power: 66	Call it what it is. There is no problem of mesh refinement, it is simply meshing. If you generate a good mesh the first time, you never need to refine it. This question is a great example of engineers wanting to optimize things forever. Avoid this trap. Ask not how to waste your time most efficiently. Avoid wasting time altogether. How do I refine efficiently? I don't.

January 5, 2024, 04:02		#3
PierreLCV Member Pierre LECHEVALIER Join Date: Mar 2023 Posts: 31 Rep Power: 3	Dear LuckyTran, Thank you very much for your quick reply ! I have to admit I'm a bit taken aback by your answer. Until now, I've more or less tried to follow a method to achieve a good mesh: 1) Make an initial coarse mesh to identify the general flow pattern and pinpoint the high-gradient areas to be refined as a priority. 2) Refine using the information gathered from the results of the first mesh AND using your instincts (you could call them your "physical sense"). 3) Repeat until the results I'm interested in converge. However, now you're telling me that the refinement process isn't a necessary step and that it's enough to find the right mesh right away... I must admit I'm a bit puzzled... How can we find the right mesh and not need to refine it if we have no experimental results to compare? Thanks for your help! Pierre

January 8, 2024, 03:54		#5
PierreLCV Member Pierre LECHEVALIER Join Date: Mar 2023 Posts: 31 Rep Power: 3	Dear LuckyTran, Thank you very much for your interesting and original answer ! Indeed, when you look at it that way, it's understandable. But given that mesh refinement is an integral part of what we call meshing, what is often the best strategy to follow in general to obtain the most accurate mesh at the lowest cost? Of course, I'm aware that there's a certain amount of "art" involved in this search for the best mesh, but I think that some general rules are still relevant... Thank you ! Best regards, Pierre