[racerter]

Hi fellow fluid dynamics peeps,

I was making mesh of a 3d wing using the advancing layer method.
However when looking at certain sections of my mesh I noticed an irregular behaviour of the near wall layers as shown in the attached photos.

What could have caused this issue and is there a way to fix this?



Pic1.jpg

pic2.jpg

[cramr5]

Are you sure that's real or just a consequence of the location of the cutting plane?
Also, hoooly molly, how many layers do you have?

[racerter]

I am not sure, but when I look at different cutting planes I do not have this visual and irregularities.

hahah I got 40 layers. I wanted a prism layer total thickness of 0.05m so that I can have a very accurate boundary layer? I guess youre hinting it might be excessive? >.<

[cramr5]

Originally Posted by racerter

I am not sure, but when I look at different cutting planes I do not have this visual and irregularities.

mm ok, I would try to run a mesh quality check to see if it returns some errors and even try running it and check the results if they make sense

Originally Posted by racerter

hahah I got 40 layers. I wanted a prism layer total thickness of 0.05m so that I can have a very accurate boundary layer? I guess youre hinting it might be excessive? >.<

haha I don't know, I think it depends on what you do, if it's academic I think it might be OK, maybe too much (If my memory is not bad I would say 20-25 is fine). My CFD is mostly professional and there you need a compromise between simulation time (and size) and accuracy and you usually use less elements to speed up simulations and get "good enough" results.

End of the day, if you have a machine powerfull enough to run it ok go for it. You could potentially have problems if your mesh is too small but hard to say beforehand if that will happen. Just try and check

[racerter]

Originally Posted by cramr5

Are you sure that's real or just a consequence of the location of the cutting plane?
Also, hoooly molly, how many layers do you have?

Originally Posted by cramr5

mm ok, I would try to run a mesh quality check to see if it returns some errors and even try running it and check the results if they make sense



haha I don't know, I think it depends on what you do, if it's academic I think it might be OK, maybe too much (If my memory is not bad I would say 20-25 is fine). My CFD is mostly professional and there you need a compromise between simulation time (and size) and accuracy and you usually use less elements to speed up simulations and get "good enough" results.

End of the day, if you have a machine powerfull enough to run it ok go for it. You could potentially have problems if your mesh is too small but hard to say beforehand if that will happen. Just try and check

Sorry for late reply, I am very new to the forums and I hadn't even noticed I got a response

When you mean mesh quality check do you just mean to run it and check the simulations? I am new to star CCM software so I am not sure what features I have avaible.

Your advice was very helpfull regarding number of layers! I decreased it and it gave me much more cells to play with around my mesh! I had initially thought it was vital to massively discretize the near layer prism cells (to capture the boundary layer) so if anything I was initially worried 40 wasn't even enough hahah >.<
My work is also for professional applications so definitely speed was an issue I had, I only have like 3e6 cells to play with

[cramr5]

Originally Posted by racerter

Sorry for late reply, I am very new to the forums and I hadn't even noticed I got a response

When you mean mesh quality check do you just mean to run it and check the simulations? I am new to star CCM software so I am not sure what features I have availabe.

No worries, no need to apologize. Yes, I would do:
- Run the mesh diagnostics in "MESH-DIAGNOSTICS" and see if you spot any problematic region.
- Run the simulation and take a look at the residuals and see if everything converges as expected, if values like tke or tdr stay high then open a scalar and locate the problematic areas and improve mesh on that area.

Originally Posted by racerter

Your advice was very helpful regarding number of layers! I decreased it and it gave me much more cells to play with around my mesh! I had initially thought it was vital to massively discretize the near layer prism cells (to capture the boundary layer) so if anything I was initially worried 40 wasn't even enough hahah >.<
My work is also for professional applications so definitely speed was an issue I had, I only have like 3e6 cells to play with

Oh ok, if it's professional then go low. Keep in mind that even F1 teams use something like 4-10 layers depending on the area in order to speed up simulations and get "good enough" results. You know your application and how accurate the near wall simulation have to be. You can also monitor the Wall y+ value and play with the layers to keep it under control.

I would also recommend use of "Custom surface mesh control" and reduce the number of layers in areas of little or no interest. That will allow you to reduce overall mesh count

[racerter]

Originally Posted by cramr5

No worries, no need to apologize. Yes, I would do:
- Run the mesh diagnostics in "MESH-DIAGNOSTICS" and see if you spot any problematic region.
- Run the simulation and take a look at the residuals and see if everything converges as expected, if values like tke or tdr stay high then open a scalar and locate the problematic areas and improve mesh on that area.


Thanks alot for the headsup, I was not aware of the mesh diagnostics tool . Until now, I was only looking at the residuals and comparing with wind tunnel data, this definitely sped up my process alot.

Oh ok, if it's professional then go low. Keep in mind that even F1 teams use something like 4-10 layers depending on the area in order to speed up simulations and get "good enough" results. You know your application and how accurate the near wall simulation have to be. You can also monitor the Wall y+ value and play with the layers to keep it under control.

I would also recommend use of "Custom surface mesh control" and reduce the number of layers in areas of little or no interest. That will allow you to reduce overall mesh count

I am suprised about the 10 layers. Because I was always in the impression the near wall thickness should be circa the size of a boundary layer (2-3cm for laminar). But with 10 layers, how will I be able to sufficiently measure boundary layer effects? Or am I completely misinterpreting what the near wall cells are meant to do?

[cramr5]

Originally Posted by racerter

I am suprised about the 10 layers. Because I was always in the impression the near wall thickness should be circa the size of a boundary layer (2-3cm for laminar). But with 10 layers, how will I be able to sufficiently measure boundary layer effects? Or am I completely misinterpreting what the near wall cells are meant to do?

10 or even less, most of the car was 3 or 4 I think. Well, keep in mind that the models are already 250-350M cells for a HALF F1 car and can take something around 8-10h from mesh to solution.
I am no CFD dev expert, I was more of an end user but from what I know the results where good enough. I think in some areas they used some surface roughness settings to adjust CFD to reality for example.

Sure you want to push to the limit but you also want a stable aero car at multiple yaws and car attitudes so you cannot design to the limit either. Additionally, parts are Wind Tunnel tested and validated before they hit the track and even with that some problems may happen.

[racerter]

Originally Posted by cramr5

10 or even less, most of the car was 3 or 4 I think. Well, keep in mind that the models are already 250-350M cells for a HALF F1 car and can take something around 8-10h from mesh to solution.
I am no CFD dev expert, I was more of an end user but from what I know the results where good enough. I think in some areas they used some surface roughness settings to adjust CFD to reality for example.

Sure you want to push to the limit but you also want a stable aero car at multiple yaws and car attitudes so you cannot design to the limit either. Additionally, parts are Wind Tunnel tested and validated before they hit the track and even with that some problems may happen.

I understand. This is very insightfull, however (and please correct me if I am wrong) I would imagine aspects such as boundary layer performance are not important for automotive applications (whereas for aerospace I would imagine they play a bigger role in performance), I doubt there are even much of a boundary layer with the exception of spoiler.

However, I am going to take your advice, I am going to try and see how a lower prism layer count effects run times and accuracy.

I am not sure I understand what you mean with Parts being wind tunnel tested and some problems may happen. Do you mean that Wind-Tunnel verification is inadequet?

[cramr5]

Originally Posted by racerter

I am not sure I understand what you mean with Parts being wind tunnel tested and some problems may happen. Do you mean that Wind-Tunnel verification is inadequet?

Well for open wheel race cars tested at 60% Scale there is so much you know and you can do but differences like tyre shape, tyre deformation, transient effect (see pourpoising), etc are still there and can make misscorrelation happen.

Not saying it's not good but it's not perfect and real world testing is still needed for some things.

[Aeronautics El. K.]

Originally Posted by racerter

I am suprised about the 10 layers. Because I was always in the impression the near wall thickness should be circa the size of a boundary layer (2-3cm for laminar). But with 10 layers, how will I be able to sufficiently measure boundary layer effects? Or am I completely misinterpreting what the near wall cells are meant to do?

One does not exclude the other. You can specify the 1st cell height, the total thickness and the number of layers or the stretch factor.

With regards to quantifying the effect of the prism cells on the solution, you could do a systematic study of the y+ and the number of layers.

[racerter]

Originally Posted by cramr5

Well for open wheel race cars tested at 60% Scale there is so much you know and you can do but differences like tyre shape, tyre deformation, transient effect (see pourpoising), etc are still there and can make misscorrelation happen.

Not saying it's not good but it's not perfect and real world testing is still needed for some things.

Oh I get what you meant now, yeah definitely, I was aware of the issues with wind tunnel testing, but atm my wing is still the design process. Thus I am using the windtunnel data of another reference model to get a rough estimate of the validity and errors I could expect of using Star-CCM for high fidelity design.

[racerter]

Originally Posted by Aeronautics El. K.

One does not exclude the other. You can specify the 1st cell height, the total thickness and the number of layers or the stretch factor.

With regards to quantifying the effect of the prism cells on the solution, you could do a systematic study of the y+ and the number of layers.

Understood, my process was to estimate the boundary layer with some mathematical relations. And then use the tools in cfdonline to measure the near wall thickness, after that I just used my judgement and some test runs to guestimate a sufficient layer number. Sadly, my computer is quiet weak so I am unable to do a design parameter study given my time restraints.

[Aeronautics El. K.]

Originally Posted by racerter

Understood, my process was to estimate the boundary layer with some mathematical relations. And then use the tools in cfdonline to measure the near wall thickness, after that I just used my judgement and some test runs to guestimate a sufficient layer number. Sadly, my computer is quiet weak so I am unable to do a design parameter study given my time restraints.

That's pretty much the approach.

I believe you said you're doing this in an industrial setting. I don't know what kind of industrial setting that is but in my case, I get to review CFD reports from various providers and about 90% they haven't done a mesh sensitivity study. So they have no idea about the error of their simulations.
And because that stuff is for regulatory compliance it matters a lot.
Then people come up with all sorts of excuses why they didn't do it or cannot do. And my answer is pretty much the same all the time... Excuses are not getting the CFD study approved. Simple as that.

So, what I mean to say is, try to do a mesh sensitivity. And you can link the prism layer parameters to the base size of the mesh so your layers are refined along with the surface and the volume mesh.

Now if the CFD you do is for internal consumption it may not matter that much, but if your work is going outside your organisation, someone might be asking you questions later on.

[racerter]

Originally Posted by Aeronautics El. K.

That's pretty much the approach.

I believe you said you're doing this in an industrial setting. I don't know what kind of industrial setting that is but in my case, I get to review CFD reports from various providers and about 90% they haven't done a mesh sensitivity study. So they have no idea about the error of their simulations.
And because that stuff is for regulatory compliance it matters a lot.
Then people come up with all sorts of excuses why they didn't do it or cannot do. And my answer is pretty much the same all the time... Excuses are not getting the CFD study approved. Simple as that.

So, what I mean to say is, try to do a mesh sensitivity. And you can link the prism layer parameters to the base size of the mesh so your layers are refined along with the surface and the volume mesh.

Now if the CFD you do is for internal consumption it may not matter that much, but if your work is going outside your organisation, someone might be asking you questions later on.

Thanks so much for your response. Even though this is for internal consumption only, I do wish to do a mesh sensitivity study in the future at my work's HPC, my own work laptop struggles with refined meshes hence most analyses I do are constrained within a low cell count.
However before I can send it to the HPC, I wanted to get a mesh with a base size of 1 to be in somewhat an acceptable form such that, my supervisor, wont have to play with mesh parameters much.

I must say I am suprised to hear from you people do not do mesh studies for external work. I cannot imagine what sort of excuses they have for not doing so.

[Aeronautics El. K.]

Originally Posted by racerter

Thanks so much for your response. Even though this is for internal consumption only, I do wish to do a mesh sensitivity study in the future at my work's HPC, my own work laptop struggles with refined meshes hence most analyses I do are constrained within a low cell count.
However before I can send it to the HPC, I wanted to get a mesh with a base size of 1 to be in somewhat an acceptable form such that, my supervisor, wont have to play with mesh parameters much.

I must say I am suprised to hear from you people do not do mesh studies for external work. I cannot imagine what sort of excuses they have for not doing so.

I get you. It's good to be mindful of the resources.
To make your life easier, perhaps you already do that, you can link the base size to a characteristic dimension and have it as a multiple of that. Makes it easier to make changes later.

As for excuses, they usually say that they don't have resources, that they solve the flow on laptops or workstations.
And the problem is that often the mesh they present, the cell size, doesn't have much consideration for the scale of the phenomena they investigate..

[racerter]

Originally Posted by Aeronautics El. K.

I get you. It's good to be mindful of the resources.
To make your life easier, perhaps you already do that, you can link the base size to a characteristic dimension and have it as a multiple of that. Makes it easier to make changes later.

As for excuses, they usually say that they don't have resources, that they solve the flow on laptops or workstations.
And the problem is that often the mesh they present, the cell size, doesn't have much consideration for the scale of the phenomena they investigate..

Yup I made sure my mesh is all linked to the base size such that in the design study it wont need much interaction.

Damn, I guess not everyone in the profesionnal field has a grasp on y+ application?

[Aeronautics El. K.]

Originally Posted by racerter

Yup I made sure my mesh is all linked to the base size such that in the design study it wont need much interaction.

Damn, I guess not everyone in the profesionnal field has a grasp on y+ application?

I'm glad that you're so positive and you think that it's just y+ that's outside of their grasp...

[racerter]

Originally Posted by Aeronautics El. K.

I'm glad that you're so positive and you think that it's just y+ that's outside of their grasp...

Hahaha oh noo XDD. I cant lie I only began CFD as a professional career a few weeks back, so all my experience is just from my uni courses.
So I wont judge them too hard yet cos Ill definitely be making my stupid mistakes as well.

[Aeronautics El. K.]

Originally Posted by racerter

Hahaha oh noo XDD. I cant lie I only began CFD as a professional career a few weeks back, so all my experience is just from my uni courses.
So I wont judge them too hard yet cos Ill definitely be making my stupid mistakes as well.

We all started from the same starting point I guess. I'm not judging anyone, in fact I still think I'm not very knowledgeable myself, but I know my limitations and I read and I try to get out of my comfort zone.

However, when certification authorities are involved, I would expect CFD studies to be solid. (See what happened with the FAA and the 737MAX, which is not CFD related but some entities dropped the ball...)

I think what is important is to think about the physics of the problem under consideration. Large scale phenomena, small scale phenomena and the associated timescales. Then, in an industrial setting, you decide how much detail is needed and what you can afford (money-wise, and also time-wise... and because time is money, actually money-wise :P) and then you build your model accordingly. Just don't cut corners, there are a few fundamental aspects that need to be there.