Evolution of Aerodynamic Testing in F1

[Tommy Cookers]

overall the array of flexures is NOT compliant - it is quasi-rigid
(each component of load is sensed via compliance from non-sensing flexures and quasi-rigidity from sensing flexures)

strain gauges are NOT long wires - they are short grids of foil that act as long wires

and 2 component load cells are available for use in car WT

[Muniix]

Computational Fluid Dynamics would have better results if the vendors hired or re-hired core developers. They've only got GUI developers at the moment, the core developers having retired or teaching now.

This is evident in that no commercial package uses GPU acceleration capability in modern PC archictures - no you have to purchase Intel Xeon 22 core processors at $14k or AMD Epyc 32 core processors at $4k rather than use a GPU accelerator card with 4000 cores each more efficient at the equations as they're optimised for pde's that costs under $2k per card and you can put up to 6 in modern two socket Epyc servers.

A run would take minutes instead of weeks.

[trinidefender]

Computational Fluid Dynamics would have better results if the vendors hired or re-hired core developers. They've only got GUI developers at the moment, the core developers having retired or teaching now.

This is evident in that no commercial package uses GPU acceleration capability in modern PC archictures - no you have to purchase Intel Xeon 22 core processors at $14k or AMD Epyc 32 core processors at $4k rather than use a GPU accelerator card with 4000 cores each more efficient at the equations as they're optimised for pde's that costs under $2k per card and you can put up to 6 in modern two socket Epyc servers.

A run would take minutes instead of weeks.

As much sense as this makes isn't there a computing power cap for CFD placed on the teams that has been in place for a number of years. I was under the impression that this limit has been reached long ago and as such makes no sense to get new computing equipment.

[Muniix]

Computational Fluid Dynamics would have better results if the vendors hired or re-hired core developers. They've only got GUI developers at the moment, the core developers having retired or teaching now.

This is evident in that no commercial package uses GPU acceleration capability in modern PC archictures - no you have to purchase Intel Xeon 22 core processors at $14k or AMD Epyc 32 core processors at $4k rather than use a GPU accelerator card with 4000 cores each more efficient at the equations as they're optimised for pde's that costs under $2k per card and you can put up to 6 in modern two socket Epyc servers.

A run would take minutes instead of weeks.

As much sense as this makes isn't there a computing power cap for CFD placed on the teams that has been in place for a number of years. I was under the impression that this limit has been reached long ago and as such makes no sense to get new computing equipment.

Wind tunnel testing is limited, teams are always doing CFD on new parts development.

Lowering the cost of CFD bringing it within range of anyone with a decent graphics card would be good for everyone. Not that you will get two CFD packages even from the same vendor giving you the same numbers to more than 2 significant digits.

[trinidefender]

Computational Fluid Dynamics would have better results if the vendors hired or re-hired core developers. They've only got GUI developers at the moment, the core developers having retired or teaching now.

This is evident in that no commercial package uses GPU acceleration capability in modern PC archictures - no you have to purchase Intel Xeon 22 core processors at $14k or AMD Epyc 32 core processors at $4k rather than use a GPU accelerator card with 4000 cores each more efficient at the equations as they're optimised for pde's that costs under $2k per card and you can put up to 6 in modern two socket Epyc servers.

A run would take minutes instead of weeks.

As much sense as this makes isn't there a computing power cap for CFD placed on the teams that has been in place for a number of years. I was under the impression that this limit has been reached long ago and as such makes no sense to get new computing equipment.

Wind tunnel testing is limited, teams are always doing CFD on new parts development.

Lowering the cost of CFD bringing it within range of anyone with a decent graphics card would be good for everyone. Not that you will get two CFD packages even from the same vendor giving you the same numbers.

I know teams always do CFD testing on parts and that wind tunnel time is limited. However I am fairly certain that teams also have caps on computing power for their CFD work. The intention of which to stop the richer teams to just have consistently better CFD than the other teams.

[Muniix]

A single two socket AMD Epyc server with 4-6 GPU acceleration cards giving up to 24,000 cores costing under $50k would solve the cost issue for all teams.

[Zynerji]

They have an Epyc system loaded with 80 MI25 cards since last year....

One PETAFLOP of double precision in a single rack.
https://www.tomshardware.com/news/amd-i ... 35118.html

Processing power caps are idiotic in todays fast moving computer world. My desktop does 6Tflops, isnt the cap on teams at like 20?!?!

PS: OpenFOAM uses GPU acceleration. I find it slightly dumb that none of the commercial software packages offer the seemingly simple OpenCL plugin that this would require....

[Greg Locock]

Here's the 2017 CFD processing allowance (well part of it anyway)

2.6 The calculation used for the declaration of the eight week Aerodynamic Testing Period (ATP)
shall be carried out as below.
TotFLOPs = (MFPPC * CCF * NCU * NSS ) / (604,800 * 8 * 1000)
2017 F1 Sporting Regulations 57/68 9 March 2017
©2017 Fédération Internationale de l’Automobile
Where :
TotFLOPs = The total number of TeraFLOPs used per CFD solve run.
MFPPC = Peak double precision floating point operations per cycle per core of the
processing unit (excluding AVX if declared under Paragraph 2.5(d) or using
natural precision operations under Paragraph 2.5(e) if the core is not double
precision capable).
CCF = Peak processing unit clock frequency in GigaHertz achieved during the CFD
solver run. This will be the peak frequency theoretically achievable during the
run based on one of the following :
a) The standard clock frequency value from the processing unit
Manufacturer’s specification sheet (if overclocking or enhanced modes
are not used in the run).
b) The maximum “turbo”, “HPC” or other enhanced mode frequency
value.
c) The maximum overclocked frequency value.
NCU = Number of processing unit cores used for the run.
NSS = Number of solver wall clock seconds elapsed during the run. The message
passing time during calculation must be included.
All information required for auditing should be present in the output from the run including
the CCF value.
For the avoidance of doubt, any offload processing for example FPU, FPGA, GPU/GPGPU, VFP,
softfp etc. should be included and calculated using the same method as above.

...

6 The Limit Line is defined as follows :
WT <= WT_limit (1 – CFD/CFD_limit)

Where :
WT = Wind On Time
WT_limit = 25 hours
CFD = CFD TeraFLOP usage
CFD_limit = 25 TeraFLOPs

[Greg Locock]

So you can play computer games, or mess about in wind tunnels, or do a bit of both.

[bill shoe]

McLaren is having pace problems, and now they talk about aero causes. According to McLaren's Eric Boullier, the car's on-track aero behavior has a significant problem that does not show up in the wind tunnel because the wind tunnel does not simulate that aspect of car behavior--

“I think the Toyota wind tunnel is fine; I completely agree with you on that, it’s not a wind tunnel issue. The characteristic of the racecar — because you can’t simulate everything in the wind tunnel — and what we’ve identified as the areas of weakness simply doesn’t show up in the wind tunnel, whether it’s Toyota’s or someone else’s, so it’s something we need to develop at the racetrack.”

What are some possible car behaviors that affects areo, but can't be simulated in an extremely advanced wind tunnel?

• Rapid bouncing/pitching up and down on tires and suspension? Leads to aero degredation in reality, but not possible to make car bounce accurately at several hz in the wind tunnel?

• Running over various types of FIA curbs (excuse me, kerbs)?

• Aero flutter where the suspension/chassis is basically stable but the wing vibrates due to the airflow itself?

[Tommy Cookers]

this reminds me of something some F1 aero person wanted me to do (but iirc he then lost his job)

how are any of these not doable ?

we might want to know which is important
flutter work might be done elsewhere

[Zynerji]

Here's the 2017 CFD processing allowance (well part of it anyway)

2.6 The calculation used for the declaration of the eight week Aerodynamic Testing Period (ATP)
shall be carried out as below.
TotFLOPs = (MFPPC * CCF * NCU * NSS ) / (604,800 * 8 * 1000)
2017 F1 Sporting Regulations 57/68 9 March 2017
©2017 Fédération Internationale de l’Automobile
Where :
TotFLOPs = The total number of TeraFLOPs used per CFD solve run.
MFPPC = Peak double precision floating point operations per cycle per core of the
processing unit (excluding AVX if declared under Paragraph 2.5(d) or using
natural precision operations under Paragraph 2.5(e) if the core is not double
precision capable).
CCF = Peak processing unit clock frequency in GigaHertz achieved during the CFD
solver run. This will be the peak frequency theoretically achievable during the
run based on one of the following :
a) The standard clock frequency value from the processing unit
Manufacturer’s specification sheet (if overclocking or enhanced modes
are not used in the run).
b) The maximum “turbo”, “HPC” or other enhanced mode frequency
value.
c) The maximum overclocked frequency value.
NCU = Number of processing unit cores used for the run.
NSS = Number of solver wall clock seconds elapsed during the run. The message
passing time during calculation must be included.
All information required for auditing should be present in the output from the run including
the CCF value.
For the avoidance of doubt, any offload processing for example FPU, FPGA, GPU/GPGPU, VFP,
softfp etc. should be included and calculated using the same method as above.

...

6 The Limit Line is defined as follows :
WT <= WT_limit (1 – CFD/CFD_limit)

Where :
WT = Wind On Time
WT_limit = 25 hours
CFD = CFD TeraFLOP usage
CFD_limit = 25 TeraFLOPs

My 4 year old desktop with 4 MI25 cars from 2017 would be over 25TFLOPS.

I'm surprised they are still stuck in 2009 computing hardware limitations... Unbelievable.

[jjn9128]

McLaren is having pace problems, and now they talk about aero causes. According to McLaren's Eric Boullier, the car's on-track aero behavior has a significant problem that does not show up in the wind tunnel because the wind tunnel does not simulate that aspect of car behavior--

“I think the Toyota wind tunnel is fine; I completely agree with you on that, it’s not a wind tunnel issue. The characteristic of the racecar — because you can’t simulate everything in the wind tunnel — and what we’ve identified as the areas of weakness simply doesn’t show up in the wind tunnel, whether it’s Toyota’s or someone else’s, so it’s something we need to develop at the racetrack.”

What are some possible car behaviors that affects areo, but can't be simulated in an extremely advanced wind tunnel?

• Rapid bouncing/pitching up and down on tires and suspension? Leads to aero degredation in reality, but not possible to make car bounce accurately at several hz in the wind tunnel?

• Running over various types of FIA curbs (excuse me, kerbs)?

• Aero flutter where the suspension/chassis is basically stable but the wing vibrates due to the airflow itself?

Yeah you can use hydraulic shakers to create high frequency vibrations in the wind tunnel. TMG use an external balance so they can put the shaker where the balance would normally go inside. You can do a tare to remove the vibration forces from the balance measurements.

I think it's more likely a scale and compressibility effect - like the Williams issue. We have to remember always that Wind Tunnels and CFD are just simulations - they approximate reality and there are compromises.

It could also be a cross-wind/natural turbulence thing - to my knowledge gusting winds isn't something teams investigate - more steady state yaws. Doesn't have to gust much to affect aero when its' so finely tuned.

[bill shoe]

I think it's more likely a scale and compressibility effect - like the Williams issue. We have to remember always that Wind Tunnels and CFD are just simulations - they approximate reality and there are compromises.

I understand on a basic, literal level what scale and compressibility are in a wind tunnel, but I don't know what you really mean by scale and compressibility as the problem. What is it that wind tunnels can't simulate (accurately?) due to scale and compressibility effects? Or, how do some teams handle that limitation better?

[Zynerji]

I think it's more likely a scale and compressibility effect - like the Williams issue. We have to remember always that Wind Tunnels and CFD are just simulations - they approximate reality and there are compromises.

I understand on a basic, literal level what scale and compressibility are in a wind tunnel, but I don't know what you really mean by scale and compressibility as the problem. What is it that wind tunnels can't simulate (accurately?) due to scale and compressibility effects? Or, how do some teams handle that limitation better?

I would expect 60% models to behave much differently than 100% scale, so they're probably having issues in their conversion math.