F1technical.net

So, here's my car for Suzuka (with a dude for scale!):




I'm pretty sure i'm not going to win this one... The car has improved since the last race, but not enough.
Downforce and efficiency are a bit higher, and balance should be a bit closer to ideal. Unfortunately, despite my efforts, i could not reduce drag, which is my main problem (in case you're wondering, a less aggressive rear wing messes up balance or efficiency). I'll probably have to re-think the whole car concept...

My podium prediction is:
1st - CAEdevice
2nd - Koldskaal
3rd - Variante

Car submitted. Not the greatest Cl or Cd, but at least this time it has 100% cooling and COP should be on point.

I did find what I think it is an issue with the mesher:



Apparently the prism layer collapse settings don't allow to capture the leading edge of the profiles. The profiles themselves are within the rules with more than 10mm thickness. Is there anything we can do about this?

AratzH wrote: ↑

30 Sep 2019, 03:45

Apparently the prism layer collapse settings don't allow to capture the leading edge of the profiles. The profiles themselves are within the rules with more than 10mm thickness. Is there anything we can do about this?

Known issue with snappyHexMesh. It collapses the layers in the most interesting areas. This is one of the reasons performance will differ in the race if you used a another CFD mesh & code to design your car.
In your case (and not allowed as part of the challenge), it would help to refine the mesh in this areas, which of course will increase the overall mesh size, which will impact CFD run times. F1 teams face exactly the same issue. This is one of the reasons, their meshes are 250M cells.

This is actually one of the interesting parts of this challenge. That you don't design a F1 car for the real world but for a universe where everything is meshed rather coarse. Copying real world high detail features like complex front wing is unlikely to work, as the mesh fidelity is just not up for the job. First to resolve the tiny gaps sufficiently and then to keep vortices "alive" away from the surface refinements.


Thomas

Thomas2019 wrote: ↑

30 Sep 2019, 06:23

AratzH wrote: ↑

30 Sep 2019, 03:45

Apparently the prism layer collapse settings don't allow to capture the leading edge of the profiles. The profiles themselves are within the rules with more than 10mm thickness. Is there anything we can do about this?

Known issue with snappyHexMesh. It collapses the layers in the most interesting areas. This is one of the reasons performance will differ in the race if you used a another CFD mesh & code to design your car.
In your case (and not allowed as part of the challenge), it would help to refine the mesh in this areas, which of course will increase the overall mesh size, which will impact CFD run times. F1 teams face exactly the same issue. This is one of the reasons, their meshes are 250M cells.

This is actually one of the interesting parts of this challenge. That you don't design a F1 car for the real world but for a universe where everything is meshed rather coarse. Copying real world high detail features like complex front wing is unlikely to work, as the mesh fidelity is just not up for the job. First to resolve the tiny gaps sufficiently and then to keep vortices "alive" away from the surface refinements.


Thomas

I would not call it a "iusse" but a misuse: simply the thickness of the airfoil is less than the geometric accuracy of the software (graphic engine): not only SnappyHxMesh but any CAD would have problems to distinguish upper and lower surface near the trailing edge (you can test this just trying to draw a line connecting upper and lower surface in a section, less than 2mm from the trailing edge). This happens because the difference of the tangent angles of the upper and lower surfaces is to small.

An heavier mesh is not alwyas the most smart solution.



In my opinion the "10 millimeters" rule (see the rullebook) is not so good to avoid this problem, and it is a non realistic restriction far from the traling edge. I would simply replace it with this rule: general minimum thickness (including the trailing edge) 2mm, except for the parts that are placed in the "High Resolution Folder" where 0,50mm would be ok. It would be also more realistic than cuspides (carbon layers can't be less thick than 0,20-0,15mm: in real cars perfect cuspides don't exist).

By the way: to avoid that problem, you can design your airfoild with a round (0,25mm) running on the tralining edge.

PS: I use Solidworks, but once I tried to model that kind of cuspdies with an high level CAD (Creo Parametric Advanced) and I had to manually set special values for the accuracy parameters.

He probably created a nicely defined profile in whatever CAD software he choose to use. That doesn't matter. Can be FreeCAD, Blender or any of the more costly packages. It would look great in the CAD package. In fact when you use parametric CAD packages, you could zoom in really close and it would still look perfectly rounded.
I was referring to the fact that in the Challenge universe everything gets discretized by 4mm/8mm. So all the ultrafine feature one designs get lost in the mesh resolution. And that's also how you lose your well rounded leading edge of the profile.
The above is true for every Mesh/CFD approach and purely a function of chosen mesh resolution.

The collapsing of the layers near problematic areas, is purely a snappyHexMesh "issue" and inherent in the method snappy uses to create meshes and produce high mesh quality. Snappy first creates the refined, snapped mesh and then tries to squeeze in the layers. If that squeezing creates cells (layer cells or squeezed "above" cells) that are of bad quality, it will undo the layer generation locally to preserve overall mesh quality metrics.
Other mesh packages will be able to grow perfectly looking layer stacks around these profiles. But they have other downsides.

Thomas

Car has been submitted. I was able to run simulations with this one so hopefully no DNS for me. Numbers are not great, expecting a time in the 100sec region which will probably be mid to tail end of the field.
Good luck to everyone

The layer issue with snappy is something we have to live with. I use optimization algorithms to determine the best setup over your cars before each season. What we have is truly the best compromise a machine can find. It is interesting that on some cars it works a bit better than on others. The quality of the stl files has a very small impact on the layer coverage. The best thing is that in my experience, not having a perfect coverage does not affect results very much. I know that every CFD expert now will spit out their coffee.
One day, a generous HPC sponsor will come along and sponsor our championship. Then you will all be able to submit your runs to a ´Cloud´ and run realistic meshes...
If we grow like this year it might not be far away.

Thomas2019 wrote: ↑

30 Sep 2019, 10:56

I was referring to the fact that in the Challenge universe everything gets discretized by 4mm/8mm.

Sounds strange to me: I have problems only when thickness is less then 1,00mm (maybe 0,5mm) and I am not sure that in real cars (made of carbon fiber layers) can exists something less thick than 0,5mm.

LVDH wrote: ↑

30 Sep 2019, 11:06

The layer issue with snappy is something we have to live with. I use optimization algorithms to determine the best setup over your cars before each season. What we have is truly the best compromise a machine can find. It is interesting that on some cars it works a bit better than on others. The quality of the stl files has a very small impact on the layer coverage. The best thing is that in my experience, not having a perfect coverage does not affect results very much. I know that every CFD expert now will spit out their coffee.
One day, a generous HPC sponsor will come along and sponsor our championship. Then you will all be able to submit your runs to a ´Cloud´ and run realistic meshes...
If we grow like this year it might not be far away.

I am out office, but... how thick are the layers? Thinner layers would help phisycs and they would be welcome (the HPC sponsor would be welcome too), but I am not so sure that they would improve the geometry discretization.

No, the layer thickness or number has nothing to do with discretization. The layer collapse will have an impact on the local boundary layer and change the way the flow "leaves" the trailing edge.

The restriction on the mesh size (and this is where the HPC sponsor comes in) allows only 4mm cells for the fine regions. Any little features or gaps you design in CAD that are smaller than 4mm will be fused over by the mesher.
Any gaps slightly larger, like in Aratz`picture, will be able to have like 2-3 4mm cells in there spanning the gap. But since the layers will lower the local mesh quality, snappyHexMesh will not put any layers there.
If you make the mesh finer, as the F1 teams do, you can fit easily 10-20 cells across the same gap and then the layers can be fitted as well, as the squeeze is not as bad.

Thomas

variante wrote: ↑

30 Sep 2019, 02:03

So, here's my car for Suzuka (with a dude for scale!):

The addition of the dude is a nice touch.

Thomas2019 wrote: ↑

30 Sep 2019, 11:24

If you make the mesh finer, as the F1 teams do, you can fit easily 10-20 cells across the same gap and then the layers can be fitted as well, as the squeeze is not as bad.

Yes, this makes a great difference. In order to avoid a strange behaviour of the flow (not exaclty a blockage, but with similar effects) I use greater gaps between the airfoils than I would use in a real car (10mm is the safer option, but 5mm can work too).

CAEdevice wrote: ↑

30 Sep 2019, 11:46

Yes, this makes a great difference. In order to avoid a strange behaviour of the flow (not exaclty a blockage, but with similar effects) I use greater gaps between the airfoils than I would use in a real car (10mm is the safer option, but 5mm can work too).

Exactly!

The car for Japanese GP chasing the rising sun.




Lot of changes on the rear, to get better balance. Obviously only by getting more downforce there. I think finally my diffuser works as it should (but still lot of potential there), so tunnels didn't work out as wanted, but that gave some options for the future. I'm only afraid that I could have too much cooling and maybe that if one bit risky move with car setting may not pay off, but let's see - it's also part of the game

General plan is to be around CAEdevice performance level

PS
I could spend definitely too much time on details and cleaning up flow in many places, and mount wings and louvres everywhere

Hello, I also submitted a car, but with only a small balance improvement. I might use the joker if I get better balance with a new car. New front wing, sidepods, diffuser.
I had my last exam just after second race submission, then I wasn't at home for a week. I have a better looking design (still very simple) and I would love to use it, but my first car is too fast

Continues wrote: ↑

30 Sep 2019, 21:51

I might use the joker if I get better balance with a new car.

Good news about the car, but you need to declare the Joker (by E-mail to Andre) before the normal deadline... this is to avoid Andre starting your CFD analysis...