General aero discussions

[PlatinumZealot]

I wish to add that perhaps I’ve mischaracterised my posts on the side pods. It’s not to say that they are not important, but rather that they are relatively well understood. The wind tunnel and CFD tools available to the teams, and especially to the top teams, are well within their capacity to accurately model well understood aerodynamic aspects such as front wings, side pods, rear wings.

The error inducing parameter is the tyres because the deformation of the tyres with the road and interaction with the road and how combined that affects the actual aero and interaction between the tyres and other elements is of course more difficult to understand and model. However, the general front wing, barge board like structures, side pods, rear wing flow structures are not difficult to model and once the baseline is there, then errors in real world aerodynamics because of inaccuracies in the tyre model can be quite reasonably corrected for during in season development with just small adjustments to wing elements and side pod shape, especially by the top teams.

There seems to be the thought by quite a few that the side pods are the main driver of the W13’s problems and subsequent performance. I don’t think it is and instead attribute all of the porpoising on the floor design and the team’s original concept which didn’t take into account the continuous downforce build up as ride height decreases with increase in speed and the sudden stall or choke. There is no doubt that front wing, barge board, side pod etc. all influence floor performance but if there was a fundamental issue with the side pod then it does not explain the W13’s problem which is actually too much downforce. Because tyre wake issues and interaction with side pod flow structure and problems associated with that are more likely to introduce unwanted losses under the floor well before the downforce could build up. Instead, what we had with the W13 was quite an effective floor sealing where downforce built up rather too well, compressed ride height, further sealing the floor edge and drastically driving downforce higher until the floor was simply too close to the ground and the flow structure under the floor broke down.

The W13 design was completely in the wrong operating window. The aerodynamicists not accounting for this and instead designing for a window where they would get very high downforce with a very low ride height. When the porpoising revealed itself and they realised they simply could not run so low, then they had to raise the ride height but their floor design and indeed other detail design above the floor intended to seal the floor edges was poor for operating at the higher ride height so they lost significant downforce. To compensate we saw bigger rear wings and this dirty downforce was very inefficient adding much drag.

Understanding how to design the floor and build tuneability into it is the difficult challenge, where it can be run at the lowest possible, and allowed, ride height such that it achieves the potential downforce at that ride height but does not then shift into the phase where the downforce ramps up quickly overcoming the suspension stiffness and further reducing ride height into the unstable flow structure zone. An ideal scenario would be to have a type of aerodynamic “vent” where downforce builds up with speed until it reaches the maximum downforce achievable with a stable flow structure, and then the downforce levels out either by leaking in just the right amount of losses under the floor to stop the downforce from continuing to build (but not too much as to break down the flow structure) or some other vent. The floor design should also be resilient to an external ride height compressor such as a bump such that a sudden ride height compression again should not result in a sudden high downforce that generates further ride height compression and break of flow structure. An aerodynamic “vent” that can relieve any excess build up of downforce from ride height compression would be ideal.

Designing a floor that lets in certain losses from the floor edge to limit the downforce, after a certain speed and corresponding ride height, seems difficult but I think quite achievable by the teams. However this doesn’t address the external force input from a bump I mentioned previously. Such a compression is likely to overcome any of the intentionally introduced losses from the floor edge and result in a condition where the floor seals and downforce builds up uncontrollably again. If the bump is big enough, it could even compress the ride immediately into the zone where the flow structure breaks down. So I really do wonder how Red Bull have achieved this? An aerodynamic “vent” could in theory stop the build up of excess downforce. However it still doesn’t answer the problem of a bigger bump that immediately puts the floor into stall conditions. So they are able to keep their platform very stable, all the while running suspension that visually seems notably more compliant than their competition. Even with the 2021 regulation changes prohibiting hydraulic control of the suspension and further 2022 restrictions on inerters and other control mechanisms.

I don’t have any solutions to the above but I believe these broad principles drives the performance characteristics and limits of the 2022 regulations. I do believe that Red Bull have specifically solved these described challenges to a large if not the full extent.

I think downforce builds slower than the spring rate increases. So I don't think the rate of increase of downforce is an issue.

The cars are actually slowing down in the corners!

The porpoising on the straights though, again is not a fast buildup for a suspension whose response can be measure in kilohertz.

I think it's just a matter of a balancing act... Go to close and the car just bounces. Remember the tyes are one part of the suspension that has a stiffness limit. You could use a solid steel column for springs... And the car will always bounce according to the tyre sidewalls!

This is why I would say that the bouncing has to do with just finding that balance and having the aerodynamics, as you so rightly say, work flexibly.

[PlatinumZealot]

I now absolutely NOTHING about yaw and wind tunnels. I was speculating.

However I am a regatta sailor (skipper) and I know a thing or two about how wind angles change with speed and wind.

If you have a 100km/h corner (I guess that's on the high side for a slow corner) with wind, it's really simple vector calculation stuff.

If you have 100km/h winds from the side, you have 45 degrees of yaw.

For 6 degrees of yaw you need 10,5 km/h side wind. That's 5,8 knots, 2,9m/s.

Depending on location, the average wind speed is somewhere 9 -12 knots and on most locations 15+ knots is a likely circumstance.

I'd take your word 6 degrees of yaw is what they test, but to me it seems that above 6 degrees of yaw happens on track.

Like I said earlier. There is probably a difference in what yaw range you design for and the yaw range you test. Maybe one would not design anything for +6 degrees yaw, but If I was evaluating a concept, I would at least want to know nothing catastrophic happens above.

Some considerations..
The ground boundary layer....
And also the car itslef acts as an air dam. A 60km/hr wind won't flow at 60km/hr into the side of the car.
The body of the car, the shark fin, endplates stop this.

[Mostlyeels]

The AMD stuff is marketing. I'm actually surprised they are using CPUs rather than GPUs. GPUs are much faster.

It's a bit of a head scratcher, to be sure. CUDA supports double precision, same as regular CPUs. Very old GPUs had limited double precision performance (only one execution unit, from memory), but modern ones can do lots of FP64. Unless CFD solvers need multi-precision arithmetic (but that would be slow on all types of execution unit), but why not just scale the values for doubles?

The special Williams CPU build was interesting though: https://www.cpu-world.com/news_2012/201 ... _6275.html

Custom made processor was Opteron 6275, named "Fangio". This CPU has the same specifications as the 6276, except that its Floating Point unit is limited to 2 double-precision FLOPS per module. As a result of FPU capping, the maximum theoretical performance of this chip is 75 GFLOPS. Integer performance and memory throughput of the 6275 are identical to Opteron 6276.

They talk about limits of data centre space, but you'd think GPU cards would have a higher FLOPS per volume of space than CPUs too.

But keep in mind that CUDA isn't used with AMD GPU's. Since AMD is the sponsor, even if they could use GPU resources, they would likely remain AMD products. And without CUDA FLOPS decrease quite a bit. Either way, for high end work I'm sure they would need some server grade CPU's to feed the GPU's.

I started off writing GPGPU and thought it might be a bit obtuse (or ROCm); I thought CUDA would be more well-known. Certainly a GPU would have several orders of magnitude more FLOPS than a CPU though, so even a slower implementation should be much faster than a CPU equivalent (also many many times more execution units). You're definitely right on the last point though, they need good infrastructure to keep them fed.

Beyond that, there are limitations to both CPU and GPU folding with how they function and what they excel at. Often distributed computing projects use both to harness the strong points in each. I know with Folding@home they must use CPU for certain reasons at times, otherwise they would always use the quicker GPU's.



At least AMD managed to use "Interlagos" and "Fangio" for a project.

Yeah, I found it too difficult to speculate on the possible edge cases (it's so specific to the code), and also too out of date with everything to really say. It was very nice to see "Fangio" used as a code name though.

[S D]

Vanja #66 or anyone else that is knowledgeable.

Perhaps you could help guide us into a more sensible discussion.

Designers use wind tunnels and computational fluid dynamics to determine the performance of the car design.

CFD helps designers analyze the fluid flows and tweak their designs. After that they create scale models and take them into the wind tunnel to verify what actually happens. Hopefully when they build the real car then it performs on track as they simulated and tested in the wind tunnel and try to correlate the actual performance to tweak the models to real life.

There is a lot of talk about floors, suspension stiffness, wing down force etc.

My question is this what do the wind tunnels and CFD help with beyond what most of us already know and what are they not capable of doing? In other words, what are the limitations other than correlation?

[Henk_v]

CFD is very rarely done on the entire car I assume. The CFD's done here are nice illustrations but useless for development.

A CFD study is mostly done on siplified models of small parts.

Also, a large part of CFD is mesh optimization and investigation into meshing dependencies.

CFD is always an attempt to approach reality with concessions necessary to keep them executable.

Running an entire car with a mesh fine enough would require immense computing force.

Any small error will propagate down.

In addition, CFD is mostly static. Time-dependent studies require at least one, preferably two orders of magnitude more computational power. So simulating yawing or bouncing of suspension is very, very difficult in terms of the required computational power and to keep small errors from stacking up to make the result useless.

Any CFD engineer confronted with a request will at first break down the problem and start to simplify the model of the problem. People think you just throw some CAD 3D file in CFD and press calculate. That's not how it works. He will first study the promblem by modelling small parts to verify stuff like boundary conditions and the required turbulence models are correct. This often involves taking away 1 or 0,5 dimension (2D or 2,5D simulations). After its established he'll work out how fine the mesh needs to be for it to not affect the outcome.

Doing this for an entire car is time consuming and repetitie with iterations.

The difficult thing is that CFD can be very, very wrong without anyone noticing. It will always produce result. The whole point of good CFD is to make sure the result is not misleading.

At some point, questions are just better answered by sticking a physical model in real air.

I am NOT a racecar engineer, but have done a lot of complex CFD. I can just imagine.

[Vanja #66]

At the risk of sounding a bit naive and knowing this is about the RB19;

I can't shake the thought that RB has linked the floor beams to the suspension. They can deform the floor with much higher force tha aero could circumventing stiffness regulations and they can flex it opposing to aero forces.

Rolling in a corner increases the downforce on the outside corner side and reduces the downforce on the inner corner side. This increases the rolling force on the car that needs to be countered with suspension. It also increases the load shift to the outer wheels.

If the suspension is linked to the floor and keeps the floor level while the car rolls, the inner corner tires take more of the load, reducing the load on the outer corner rear wheel. The aero does expert less rolling force, allowing for les stiff suspension setting.

But maybe thats just dumb...

That would be attaching parts of chassis directly on wheels (as unsprung mass) and very illegal. Also not a very easy thing to do in a good way.

Vanja #66 or anyone else that is knowledgeable.

Perhaps you could help guide us into a more sensible discussion.

Designers use wind tunnels and computational fluid dynamics to determine the performance of the car design.

CFD helps designers analyze the fluid flows and tweak their designs. After that they create scale models and take them into the wind tunnel to verify what actually happens. Hopefully when they build the real car then it performs on track as they simulated and tested in the wind tunnel and try to correlate the actual performance to tweak the models to real life.

There is a lot of talk about floors, suspension stiffness, wing down force etc.

My question is this what do the wind tunnels and CFD help with beyond what most of us already know and what are they not capable of doing? In other words, what are the limitations other than correlation?

Limitation is accuracy of simulation. CFD alone on rigid geometry is very limited if it's not paired with proper fluid-structureinteraction (FSI) model, especially for front wing design. It won't affect whole lot if you did your job well, but you need to make sure of that.

Wheel-ground interaction is critical, modelling the tyre squirt correctly is of great importance for accurate results, you need to model the tyre wake just right if you want to make sure it doesn't have a negative effect on the car. Other than that, the biggest limitation of CFD from calculation and accuracy side of things are turbulence models - once you've made sure your geometry is accurate enough model of how the car will be on track. Meshing is critical to be able to start the solver (the actual calculation) and also if you make a mesh to coarse you will miss a lot of phenomena, separation might be at the wrong place etc.

CFD is very rarely done on the entire car I assume. The CFD's done here are nice illustrations but useless for development.

A CFD study is mostly done on siplified models of small parts.

Also, a large part of CFD is mesh optimization and investigation into meshing dependencies.

CFD is always an attempt to approach reality with concessions necessary to keep them executable.

Running an entire car with a mesh fine enough would require immense computing force.

Any small error will propagate down.

In addition, CFD is mostly static. Time-dependent studies require at least one, preferably two orders of magnitude more computational power. So simulating yawing or bouncing of suspension is very, very difficult in terms of the required computational power and to keep small errors from stacking up to make the result useless.

Any CFD engineer confronted with a request will at first break down the problem and start to simplify the model of the problem. People think you just throw some CAD 3D file in CFD and press calculate. That's not how it works. He will first study the promblem by modelling small parts to verify stuff like boundary conditions and the required turbulence models are correct. This often involves taking away 1 or 0,5 dimension (2D or 2,5D simulations). After its established he'll work out how fine the mesh needs to be for it to not affect the outcome.

Doing this for an entire car is time consuming and repetitie with iterations.

The difficult thing is that CFD can be very, very wrong without anyone noticing. It will always produce result. The whole point of good CFD is to make sure the result is not misleading.

At some point, questions are just better answered by sticking a physical model in real air.

I am NOT a racecar engineer, but have done a lot of complex CFD. I can just imagine.

F1 cars are simulated always from the start in yaw, full model, front wheels turned, car rolling, etc. All attitudes of a car are simulated (imagine braking+cornering e.g.), including actual cornering simulations (semi-ring domain with a ground rolling with angular velocity law). As far as I know, all, and I mean all, openings on the car are there in the simulation, although radiator cores are simplified with very accurate models and airbox plenum is a subdomain with a specific sub-pressure. Internals are simplified, but internal ducting is an exact geometry. Even exhaust gasses are introduced to fully replicate everything that can happen, even if exhaust position is fixed and has ridiculously small influence on overall car aero.

Meshing is almost fully automated, all solvers and turbulence models are bespoke, I think also every simulation is transient. Calculations are perfected and very efficient regarding teraflops since those are the limiting factor per regulation. If you think they are going too far, I can assure you I probably haven't listed every detail that every simulation has.

Regarding CFD accuracy and WT testing, criteria for approving a model for WT is very high and CFD results need to be good enough and there has to be a consensus among aero engineers that CFD results are accurate. WT can confirm or disapprove CFD and the best geometry ends up on the car. Friday FP sessions are then used to confirm if the design is working as intended. Sometimes it's an improvement, but not as big as expected. In any case, all the data from the track is fed back to turbulence model correlation and driving simulator. Everything is likely more complex than what I've written, but I think this can give you a good idea

[AR3-GP]

My question is this what do the wind tunnels and CFD help with beyond what most of us already know and what are they not capable of doing? In other words, what are the limitations other than correlation?

1) Wind tunnel model uses different materials and construction to the race car. Any deflection and flutter characteristics of the race car parts (wings, floor) won't be replicated in the windtunnel model.

2) It is very difficult to accurately reproduce the race car tire shapes under load in the wind tunnel model.

3) Wind tunnel model motion systems are slow. You cannot model any realistic transient behaviors (how quickly do loads change when the car pitches and rolls). It's difficult to understand if there is hysteresis in the aero map (if loads "stick" where they shouldn't). A few years ago, there was a rumor that Williams had a problem with flow re-attachment on their rear wing after closing the DRS which led to a crash at Silverstone if I'm not mistaken.

[S D]

At the risk of sounding a bit naive and knowing this is about the RB19;

Vanja #66 or anyone else that is knowledgeable.

Perhaps you could help guide us into a more sensible discussion.

Designers use wind tunnels and computational fluid dynamics to determine the performance of the car design.

CFD helps designers analyze the fluid flows and tweak their designs. After that they create scale models and take them into the wind tunnel to verify what actually happens. Hopefully when they build the real car then it performs on track as they simulated and tested in the wind tunnel and try to correlate the actual performance to tweak the models to real life.

There is a lot of talk about floors, suspension stiffness, wing down force etc.

My question is this what do the wind tunnels and CFD help with beyond what most of us already know and what are they not capable of doing? In other words, what are the limitations other than correlation?

Limitation is accuracy of simulation. CFD alone on rigid geometry is very limited if it's not paired with proper fluid-structure interaction (FSI) model, especially for front wing design. It won't affect whole lot if you did your job well, but you need to make sure of that.

Wheel-ground interaction is critical, modelling the tyre squirt correctly is of great importance for accurate results, you need to model the tyre wake just right if you want to make sure it doesn't have a negative effect on the car. Other than that, the biggest limitation of CFD from calculation and accuracy side of things are turbulence models - once you've made sure your geometry is accurate enough model of how the car will be on track. Meshing is critical to be able to start the solver (the actual calculation) and also if you make a mesh to coarse you will miss a lot of phenomena, separation might be at the wrong place etc.

Thanks for that. Do the WT scale models have sensors on them such as the real car would have? I.e. Can they measure pressure or loads in critical places or are they simply mechanical mockups? Example, can they see how much down force is generated by the front and back wheels?

Would they place the grid of sensors that we see behind the front or back wheels during track testing? I know that they use the paint or smoke to track the flow of air.

[Stu]

Moog have been working with Honda HPD (USA) on wind tunnel tech to introduce ride height control (very similar to active suspension) for their wind tunnel (for Indycar). It is not a huge jump to assume that such a system could be used to measure induced forces….

[Vanja #66]

Thanks for that. Do the WT scale models have sensors on them such as the real car would have? I.e. Can they measure pressure or loads in critical places or are they simply mechanical mockups? Example, can they see how much down force is generated by the front and back wheels?

Would they place the grid of sensors that we see behind the front or back wheels during track testing? I know that they use the paint or smoke to track the flow of air.

There are all kinds of sensors on WT models, including pressure, force, torque, pitot tubes, the whole lot. You'll find a lot more in this video

[sucof]

Remember in the previous regulations, RB had the highest rake, and then people talked about that the most, and they tried to copy it but no one replicated it perfectly. The key was the suspension, and probably a bit of floor geometry.
The same happens now as well. It would be silly to find out that there is one small element, solution which allows them to have a different way of their suspension to work like it does, and that wins them so many races and championships.
I wonder if any suspension specialists left RB for other teams bringing the ideas with them?

Or it is simply about a better simulation of vehicle suspension dynamics? This could be a huge differentiator as well. As teams tried have the best CFD and expensive wind tunnels, they might did not pay the same attention to suspension simulations?
This could also explain why RB went out on the first day of testing with a quick car, and Ferrari decided to play with the setup all along, without ending up anything close to good?
Imagine how many elements shall such a simulation have:
-Tire deformation regarding speed, tire heat, pressure, etc.
-How suspension elements work exactly, springs, dampers, geometry...
-How suspension, chassis, floor elasticity is happening...
-How parts of the car moving around affecting ride like moving dampers/weights...
-How car weight changes during a race...
-And you have to simulate the whole car in one piece, you can not have results only simulating one wheel only...
-And you have to add track surfaces, slopes, kerbs and so on...
the list is kind of endless... And you might want to include CFD as well as the two works hand in had, especially in the new regulations.

As a programmer, I can see that such a simulation could be a lot harder to make then a CFD, as it has so many different variables. Perhaps a lot more than CFD... Once you have one set incorrectly, the whole simulation falls apart.
Think about the rumours that Ferrari saw their cars in simulations a few tenths faster... it was perhaps not the CFD they were talking about.

[Tommy Cookers]

the above video is very good
....

Moog have been working with Honda HPD (USA) on wind tunnel tech to introduce ride height control (very similar to active suspension) for their wind tunnel (for Indycar). It is not a huge jump to assume that such a system could be used to measure induced forces….

it is a huge jump

wind tunnels were and are primarily for force measurement
the problems were and are with the WT and with the model - not with the force measurement

[Stu]

the above video is very good
....

Moog have been working with Honda HPD (USA) on wind tunnel tech to introduce ride height control (very similar to active suspension) for their wind tunnel (for Indycar). It is not a huge jump to assume that such a system could be used to measure induced forces….

it is a huge jump

wind tunnels were and are primarily for force measurement
the problems were and are with the WT and with the model - not with the force measurement

The Moog collaboration is all about model control.

[S D]

Thanks for that. Do the WT scale models have sensors on them such as the real car would have? I.e. Can they measure pressure or loads in critical places or are they simply mechanical mockups? Example, can they see how much down force is generated by the front and back wheels?

Would they place the grid of sensors that we see behind the front or back wheels during track testing? I know that they use the paint or smoke to track the flow of air.

There are all kinds of sensors on WT models, including pressure, force, torque, pitot tubes, the whole lot. You'll find a lot more in this video

Excellent video. Thanks for that.

So over the winter, Ferrari get their software models, feed them into the simulator, come up with scale designs that are built and then tested in the wind tunnel, and go back and forth until they like what they see.

They then built the cars and take them to preseason testing and either determined or confirmed suspicions that they need more FW and RW down force, especially RW.

Based on the testing feedback they needed to adjust the models in software to reflect the tested reality and rebuild the test model by implementing new FWs and/or RWs. They must also have adjusted the measured requirements from the sensors to account for the extra needed down force.

Once satisfied with their new designs or forced by time constraints based on the race calendar, they manufacture new front and/or rear wings (possibly multiple versions) and bring the new wings to the next race and hope that they are an improvement.

So this is why it takes a few races to bring new parts. There wasn't sufficient time after pre-season testing to bring a proper RW to Bahrain and it was rushed into service where it failed.

Hopefully they now have some sense of whether the RW concept works or needs more redesign. Even though if it did what they wanted it to achieve aerodynamically they still need to reinforce the RW structure.

Is this right in your opinion?

[Vanja #66]

Well it's always a back-and-forth process, rear wing in particular was pushed hard on its diet and weight saving, causing structursl issues. They didn't get enoufh time to validate its aero design for sure, so they need strengthened parts before they can do that.