[hitzhwan]

Can anyone tell me how to increase the mesh elements for the mesh independency , need I to increase all the direction mesh size or just study one direction that is important?

[vinerm]

There is no fixed method for this. The only objective of grid sensitivity analysis is to ensure that the effect of the mesh is not significant on the solution. And its a misconception that the mesh size should only be increased to do a grid sensitivity analysis. You can try by reducing from its current state as well. And you can do that in a particular direction or in two or three directions. It doesn't really matter.

[hitzhwan]

Originally Posted by vinerm

There is no fixed method for this. The only objective of grid sensitivity analysis is to ensure that the effect of the mesh is not significant on the solution. And its a misconception that the mesh size should only be increased to do a grid sensitivity analysis. You can try by reducing from its current state as well. And you can do that in a particular direction or in two or three directions. It doesn't really matter.

But I find the variable that I monitor does not reach a steady state after the mesh increasing， and it has matter for different mesh direction increasing.

[mCiFlDk]

Originally Posted by hitzhwan

Can anyone tell me how to increase the mesh elements for the mesh independency , need I to increase all the direction mesh size or just study one direction that is important?

Regarding this I agree with Vinerm. There is not a particular method for it. You should read some lierature about the topic and plan it manually to see what values should you increase/decrease.

Originally Posted by hitzhwan

But I find the variable that I monitor does not reach a steady state after the mesh increasing， and it has matter for different mesh direction increasing.

If the problem you're facing is the lack of convergence, probably the problem is very different than the mesh size. It means that it may have problems (e.g. free faces), a lack of quality or, if you're using inflation, a bad usage of it. Check the mesh out first.

[hitzhwan]

Originally Posted by mCiFlDk

Regarding this I agree with Vinerm. There is not a particular method for it. You should read some lierature about the topic and plan it manually to see what values should you increase/decrease.



If the problem you're facing is the lack of convergence, probably the problem is very different than the mesh size. It means that it may have problems (e.g. free faces), a lack of quality or, if you're using inflation, a bad usage of it. Check the mesh out first.

Yes, it may be, I can find some bad mesh element in my mesh model, but I do
not know how to improve it in Mesh software, do you know how to improve it ?

[duri]

Originally Posted by hitzhwan

Can anyone tell me how to increase the mesh elements for the mesh independency , need I to increase all the direction mesh size or just study one direction that is important?

Mesh resolution is critical near the flow gradients. Mesh independence can be achieved by targeting the gradients. In case of direction it is better to maintain the mesh metrics like aspect ratio which would again influence the gradient estimation. So, if your are refining 4 times in x direction it is better to refine at least twice in other directions. Overall mesh count can be optimized by coarsening mesh in non-gradient region. This can be done based on the observed flow feature from initial runs.

[duri]

Originally Posted by hitzhwan

But I find the variable that I monitor does not reach a steady state after the mesh increasing， and it has matter for different mesh direction increasing.

There are two possibilities,


1. Mesh near the monitor point is bad and error doesn't settling down. In case of flux balance or force monitors few bad cells doesn't prevent reaching steady state solution. In this case check the max residuals it should be always on the bad cells for all iterations.



2. If the monitor oscillation is periodic or bounded then probably mesh is picking up some unsteady flow feature which has been dampened by coarse mesh. Switching to unsteady solver could give more insight.

[hitzhwan]

Originally Posted by duri

There are two possibilities,


1. Mesh near the monitor point is bad and error doesn't settling down. In case of flux balance or force monitors few bad cells doesn't prevent reaching steady state solution. In this case check the max residuals it should be always on the bad cells for all iterations.



2. If the monitor oscillation is periodic or bounded then probably mesh is picking up some unsteady flow feature which has been dampened by coarse mesh. Switching to unsteady solver could give more insight.

1.How to check the max residuals it should be always on the bad cells for all iterations.

2.I just used the unsteady solver, what is the max aspect ratio for the mesh? Should it be less than 5? You mean we should change all the direction when we refinement the mesh?

[hitzhwan]

Originally Posted by vinerm

There is no fixed method for this. The only objective of grid sensitivity analysis is to ensure that the effect of the mesh is not significant on the solution. And its a misconception that the mesh size should only be increased to do a grid sensitivity analysis. You can try by reducing from its current state as well. And you can do that in a particular direction or in two or three directions. It doesn't really matter.

Can't I use the inflation and face meshing in the same time？ Why does it show can not coexist？

[hitzhwan]

Originally Posted by vinerm

There is no fixed method for this. The only objective of grid sensitivity analysis is to ensure that the effect of the mesh is not significant on the solution. And its a misconception that the mesh size should only be increased to do a grid sensitivity analysis. You can try by reducing from its current state as well. And you can do that in a particular direction or in two or three directions. It doesn't really matter.

What’s wrong with the '-' beside the face meshing 2?

[mCiFlDk]

Originally Posted by hitzhwan

Yes, it may be, I can find some bad mesh element in my mesh model, but I do
not know how to improve it in Mesh software, do you know how to improve it ?

I think you're asking for tools to fix a mesh. If not, just correct me. But if you're looking for it, there are some PDFs if you look for them, with very useful information given by ANSYS inc. I let you one (it's for Fluent meshing, I don't know what mesher are you using).

Regards.

http://oss.jishulink.com/upload/2016...9%E7%A8%8B.pdf

[duri]

Originally Posted by hitzhwan

1.How to check the max residuals it should be always on the bad cells for all iterations.

Some solvers would give the location of max residual. I don't know how to get in fluent for each iterations. But you can stop randomly and check the post processing option.

2.I just used the unsteady solver, what is the max aspect ratio for the mesh? Should it be less than 5? You mean we should change all the direction when we refinement the mesh?

Max aspect ratio within 5 is very difficult to achieve. Near wall mesh would be in order of 1000s. But let say your mesh is 2D and it has 1e-5 as first cell height and along surface the node distance is 1e-2. This gives AR of 1000. When you make the first cell height to 1e-6 AR jumps to 10000. Now it is difficult to understand the mesh dependency of the solution. If possible it is better to maintain the same mesh metrics but this will lead to 8 times increase in mesh size which is again lead to computational cost in terms of memory and time step size. Also fine mesh has more high frequency error components which can reduced the numerical stability and take more time to die down. So it is up to the user to decide what need to be compromised.

[hitzhwan]

Originally Posted by mCiFlDk

I think you're asking for tools to fix a mesh. If not, just correct me. But if you're looking for it, there are some PDFs if you look for them, with very useful information given by ANSYS inc. I let you one (it's for Fluent meshing, I don't know what mesher are you using).

Regards.

http://oss.jishulink.com/upload/2016...9%E7%A8%8B.pdf

What is fluent meshing? Where can I open it ？ I use the Ansys mesh software, are they the same?

[hitzhwan]

Originally Posted by duri

Some solvers would give the location of max residual. I don't know how to get in fluent for each iterations. But you can stop randomly and check the post processing option.


How to check in the post processing option, can you describe it clearly?



Max aspect ratio within 5 is very difficult to achieve. Near wall mesh would be in order of 1000s. But let say your mesh is 2D and it has 1e-5 as first cell height and along surface the node distance is 1e-2. This gives AR of 1000. When you make the first cell height to 1e-6 AR jumps to 10000. Now it is difficult to understand the mesh dependency of the solution. If possible it is better to maintain the same mesh metrics but this will lead to 8 times increase in mesh size which is again lead to computational cost in terms of memory and time step size. Also fine mesh has more high frequency error components which can reduced the numerical stability and take more time to die down. So it is up to the user to decide what need to be compromised.

But if the AR is too high, I think the calculation may easily divergence, how to solve it ?

[mCiFlDk]

Originally Posted by hitzhwan

What is fluent meshing? Where can I open it ？ I use the Ansys mesh software, are they the same?

Fluent meshing is also known as TGRID, and it's basically used to create large, unstructured meshes for complex geometries. In ANSYS meshing (what you're using) you mesh the volume, in Fluent meshing, you mesh the surface semi-automatically, having much more control on what you're doing. It's a bit less intuitive than your way, but in my opinion is the best thing to avoid headaches together with ICEM CFD (another mesher).

You can access both of them from the workbench interface, and there are (not many) tutorials, but everything is practice!

PD: How have you attached thumbnails from your PC? I haven't found anything in FAQs section

[hitzhwan]

Originally Posted by mCiFlDk

Fluent meshing is also known as TGRID, and it's basically used to create large, unstructured meshes for complex geometries. In ANSYS meshing (what you're using) you mesh the volume, in Fluent meshing, you mesh the surface semi-automatically, having much more control on what you're doing. It's a bit less intuitive than your way, but in my opinion is the best thing to avoid headaches together with ICEM CFD (another mesher).

You can access both of them from the workbench interface, and there are (not many) tutorials, but everything is practice!

PD: How have you attached thumbnails from your PC? I haven't found anything in FAQs section

Do you have some material to learn the fluent meshing? Does it easy to learn?


You can see the image I send for you, just click the smiling face, you can send the image.

[mCiFlDk]

Originally Posted by hitzhwan

Do you have some material to learn the fluent meshing? Does it easy to learn?


You can see the image I send for you, just click the smiling face, you can send the image.

It's quite weird at the beginning, but it has like 2 steps: 1) using it and looking at the beauty of the mesh you've just created and 2) start to realize that despite being beauty, it has errors that you've to fix manually. The key of this meshing tool is to learn how to fix the errors you have (PDF I sent you previously), because only doing that, the mesh will be very acceptable.

The only material I can let you was the PDF I sent you in the other quote and two attached docs here. There's a lack of info about it, but you can manage it quite well in a couple of projects.

Video URL: https://www.youtube.com/watch?v=_4C7vhIUpPo&t=1184s

PDF: http://dl.mr-cfd.com/tutorials/ansys..._Workflows.pdf

PD: cheers for the thumbnail, it works

[duri]

Originally Posted by hitzhwan

But if the AR is too high, I think the calculation may easily divergence, how to solve it ?

It depends on the implementation. It usually affects the gradient estimation and most of the commercial solvers are quite robust to handle high AR cells.

[hitzhwan]

Originally Posted by mCiFlDk

It's quite weird at the beginning, but it has like 2 steps: 1) using it and looking at the beauty of the mesh you've just created and 2) start to realize that despite being beauty, it has errors that you've to fix manually. The key of this meshing tool is to learn how to fix the errors you have (PDF I sent you previously), because only doing that, the mesh will be very acceptable.

The only material I can let you was the PDF I sent you in the other quote and two attached docs here. There's a lack of info about it, but you can manage it quite well in a couple of projects.

Video URL: https://www.youtube.com/watch?v=_4C7vhIUpPo&t=1184s

PDF: http://dl.mr-cfd.com/tutorials/ansys..._Workflows.pdf

PD: cheers for the thumbnail, it works

Thanks. Do you know how to use the Mesh software, I have some problems. For example, when the edge mesh and inflation both exist, the inflation's growdth does not work, why?

[mCiFlDk]

Originally Posted by hitzhwan

Thanks. Do you know how to use the Mesh software, I have some problems. For example, when the edge mesh and inflation both exist, the inflation's growdth does not work, why?

Honestly, it'd be better that I don't help you using the mesher you're using. It's been 2 years or so since I don't use it. I'd recommend you to look for some literature related to it, and then mesh, because the order you're following when doing some operations in this mesher sometimes influences and deactivates some features.

It seems that you're trying to run before starting to walk, and that's a very bad idea using Fluent and leads to frustration. Go step by step. My advice is to write a specific post about it, but meanwhile start looking for information. Many times it works much better than simply asking.

Best of luck hitzhwan.