Wind Tunnel Superceded with Total Switch to CFD

[SR71]

Apologies for linking to 'thejudge' but I haven't seen this reported anywhere else. I'll switch the link out as soon as I can.

https://thejudge13.com/2016/08/22/wind- ... ch-to-cfd/

“We’ve asked the Strategy Group to look into a position where we can do a glide path down and a total switch to CFD,” Force India deputy team principal Bob Fernley told Motorsport.com.

“It’s on the basis that we can now foresee that in time CFD will replace wind tunnels, and we want to facilitate that, effectively.

“What we have today is a wind tunnel-biased formula, and what we are looking for is a CFD-biased formula that allows people to switch to CFD. It’s a completely different equation, and the current system would stay in place.

[Cold Fussion]

How much does a wind tunnel cost to run a year? I don't buy the cost savings because with open CFD regulations will simply start a computing arms race. Unless of course he wants to stick with the paltry 30 Tflop limit in which case we'll just stymie development.

[SR71]

How much does a wind tunnel cost to run a year? I don't buy the cost savings because with open CFD regulations will simply start a computing arms race. Unless of course he wants to stick with the paltry 30 Tflop limit in which case we'll just stymie development.

No clue. But have heard many teams say wind tunnel expenses are quite a real thing.

I don't know how it compares but there is no way simulation stays as expensive as it currently is NOR is there anyways simulation power doesn't double or even triple want we have now in the next decade (moore's law being the bare minimum) all while falling in price.

If you're correlation is strong then nobody could justify 1 part in a wind tunnel VS hundreds or thousands of permutations of the same part in the same amount of time VIA simulation.

[Paul]

That would be quite interesting because it would level the playing field somewhat, but with no wind tunnel validation I foresee lots of discarded carbon after free practices.

[Vyssion]

There is no doubt that eventually CFD will overtake wind tunnel testing as the best method for aerodynamic verification. Hybrid-LES turbulence models are being used at the moment in some of the top level simulation runs with several PhDs I know of looking at full LES simulations of old McLaren F1 cars and the like. DNS I feel will still be a long way away...

The number of Cells you need must be larger than the Reynolds Number raised to the power of 9/4. So for example, if you took a 1m chord wing and flew it through normal air at about 16m/s, your Reynolds number would be about 1.06 Million. So in order to perform a DNS simulation and capture all possible flow eddies, you would need 3.6x10^13 cells (36,000,000,000,000 cells)!!!! Given that the largest simulation that I know of was solved over a million cores with a total cell count of 1.4 trillion (1.4x10^12 cells) we are still a long way off having enough computational power to know exactly what is happening...

However, with the advent of quantum computing coming to the forefront, perhaps not as far away as one may think...

You are right in that teams which own tunnels and run them will begin to pour their money into clusters and programmers, however, the difference between a simulation which is run at 2 billion cells compared to 200 million cells is only about 2x better resolution in 3D space but will require (based on the typical CFD recommendation of 1Gb RAM per 1 million cells) an additional 1.8 Tb RAM along with a huge number of additional cores in order to still get results back in similar times to the about ~3hrs it takes a 200 million cell sim to execute across 96-cores.

That gap in performance, I feel, is much less than that of the current set up where wind tunnel testing is so far ahead of the equivalent terraflops of CFD that teams which don't have as much money fall behind at a much faster rate. Given that this is going to happen anyways, I am sure that with all the development into interpolation schemes and convergence theorems that teams use currently along with just the overall optimisation of solving something over a computer, we will see F1 not only bring technology to cars but to general computing as well.

[FW17]

The number of Cells you need must be larger than the Reynolds Number raised to the power of 9/4. So for example, if you took a 1m chord wing and flew it through normal air at about 16m/s, your Reynolds number would be about 1.06 Million. So in order to perform a DNS simulation and capture all possible flow eddies, you would need 3.6x10^13 cells (36,000,000,000,000 cells)!!!! Given that the largest simulation that I know of was solved over a million cores with a total cell count of 1.4 trillion (1.4x10^12 cells) we are still a long way off having enough computational power to know exactly what is happening...

If that is the number for a wing code what is it for the whole car?

[Vyssion]

The number of Cells you need must be larger than the Reynolds Number raised to the power of 9/4. So for example, if you took a 1m chord wing and flew it through normal air at about 16m/s, your Reynolds number would be about 1.06 Million. So in order to perform a DNS simulation and capture all possible flow eddies, you would need 3.6x10^13 cells (36,000,000,000,000 cells)!!!! Given that the largest simulation that I know of was solved over a million cores with a total cell count of 1.4 trillion (1.4x10^12 cells) we are still a long way off having enough computational power to know exactly what is happening...

If that is the number for a wing code what is it for the whole car?

It tends to vary over the entire car. The Reynolds Number is defined as the ratio between the inertial forces divided by the viscous forces; that is "Re = (local velocity * characteristic length) / kinematic viscosity" usually. So depending on the part of the car, this number will change. So for a front wing, Re will vary on the suction surface vs. the pressure surface due to it being a slightly longer distance, and also inducing a faster local velocity as well. But it is still a short distance when compared to the length of underbody tunnels and diffusers which tend to also have quite high velocities through them.

Essentially, the way turbulence works is it begins as large whorls which slowly break down to smaller whorls. These big ones are mostly dependent on velocity. Once they break down to a small enough scale (i.e. within the Sub-grid Scale defined in LES simulations where it is assumed that all flows are identical) then viscosity forces are dominant and that is how turbulence eventually dies out. The number of "steps" [?] between a big whorl breaking down over and over to one which is then damped out by viscosity is mostly dependent on velocity and the size of the fluctuations which are present in the flow. A surface which is close by will affect this breakdown cycle as there is the no-slip condition on the surface and boundary layer flow which induces additional shear forces.

So the Reynolds number on the front wing will vary pressure side to suction side, which will vary when compared to the rear wing, which will vary compared to the underbody................. you get the idea

[Cold Fussion]

How much does a wind tunnel cost to run a year? I don't buy the cost savings because with open CFD regulations will simply start a computing arms race. Unless of course he wants to stick with the paltry 30 Tflop limit in which case we'll just stymie development.

No clue. But have heard many teams say wind tunnel expenses are quite a real thing.

I don't know how it compares but there is no way simulation stays as expensive as it currently is NOR is there anyways simulation power doesn't double or even triple want we have now in the next decade (moore's law being the bare minimum) all while falling in price.

If you're correlation is strong then nobody could justify 1 part in a wind tunnel VS hundreds or thousands of permutations of the same part in the same amount of time VIA simulation.

It doesn't need to double of triple than they have now in the next decade, if there were no computing limits in place in the regulations, every team could go out tomorrow and order a super computer 1000x faster than what they have now.

[Zynerji]

Using pro level graphics cards to do the number crunching in Open CL could easily put them into hundreds of TFLOPS for a few thousand dollars.

A CFD only formula would necessitate some restrictions however, as with only 2 mega chip manufacturers in the world, there would need to be some rule that doesn't lock one of them to a single team.

[Cold Fussion]

A CFD only formula would necessitate some restrictions however, as with only 2 mega chip manufacturers in the world, there would need to be some rule that doesn't lock one of them to a single team.

It wouldn't be the chip manufactures it would be the supercomputer manufactures (like Cray etc). In any case I'd be surprised if any of them would sign exclusive agreements as it's not in their best interests to limit their supply.

[Jersey Tom]

To think that any type of simulation is a complete replacement for physical measurement is very naive, in my experience. Doesn't have to be CFD. Could be tires, could be vehicle suspensions, you name it.

Though it's entirely possible the balance is too heavy on the wind tunnel side of things (or maybe it isn't, I'm not privy to that). But there's a difference between rebalancing time vs. eliminating something.

[bhall II]

If the smaller teams were smart, they'd pool their resources together and build/buy a wind tunnel. Instead, it seems nothing can shatter the false notion that restrictions beget equality.

[hardingfv32]

This will not be an equalizer. The teams that spend the most money on validation activities will always have the advantage. Even at this point there are leaders/winners as far as validation is concerned.

This is a cost control pipe dream!

Brian

[SameSame]

There is no doubt that eventually CFD will overtake wind tunnel testing as the best method for aerodynamic verification. Hybrid-LES turbulence models are being used at the moment in some of the top level simulation runs with several PhDs I know of looking at full LES simulations of old McLaren F1 cars and the like. DNS I feel will still be a long way away...

The number of Cells you need must be larger than the Reynolds Number raised to the power of 9/4. So for example, if you took a 1m chord wing and flew it through normal air at about 16m/s, your Reynolds number would be about 1.06 Million. So in order to perform a DNS simulation and capture all possible flow eddies, you would need 3.6x10^13 cells (36,000,000,000,000 cells)!!!! Given that the largest simulation that I know of was solved over a million cores with a total cell count of 1.4 trillion (1.4x10^12 cells) we are still a long way off having enough computational power to know exactly what is happening...

Why are the number cells Reynold number dependent? Is there a proof for that or was it determined by empirical observation? So theoretically you could solve the flow exactly around your 1m chord wing at 16m/s provided the air had a high kinematic viscosity? (e.g. Air at extremely high temperatures for 1 atm)

[SR71]

There is no doubt that eventually CFD will overtake wind tunnel testing as the best method for aerodynamic verification. Hybrid-LES turbulence models are being used at the moment in some of the top level simulation runs with several PhDs I know of looking at full LES simulations of old McLaren F1 cars and the like. DNS I feel will still be a long way away...

The number of Cells you need must be larger than the Reynolds Number raised to the power of 9/4. So for example, if you took a 1m chord wing and flew it through normal air at about 16m/s, your Reynolds number would be about 1.06 Million. So in order to perform a DNS simulation and capture all possible flow eddies, you would need 3.6x10^13 cells (36,000,000,000,000 cells)!!!! Given that the largest simulation that I know of was solved over a million cores with a total cell count of 1.4 trillion (1.4x10^12 cells) we are still a long way off having enough computational power to know exactly what is happening...

However, with the advent of quantum computing coming to the forefront, perhaps not as far away as one may think...

You are right in that teams which own tunnels and run them will begin to pour their money into clusters and programmers, however, the difference between a simulation which is run at 2 billion cells compared to 200 million cells is only about 2x better resolution in 3D space but will require (based on the typical CFD recommendation of 1Gb RAM per 1 million cells) an additional 1.8 Tb RAM along with a huge number of additional cores in order to still get results back in similar times to the about ~3hrs it takes a 200 million cell sim to execute across 96-cores.

That gap in performance, I feel, is much less than that of the current set up where wind tunnel testing is so far ahead of the equivalent terraflops of CFD that teams which don't have as much money fall behind at a much faster rate. Given that this is going to happen anyways, I am sure that with all the development into interpolation schemes and convergence theorems that teams use currently along with just the overall optimisation of solving something over a computer, we will see F1 not only bring technology to cars but to general computing as well.

Amazing. Thanks!