F1technical.net

I've been taking a look on this article tweeted by ScarbsF1.
http://www.cd-adapco.com/pdfs/presentat ... /1-art.pdf
You can see more than a 20% of absolute error in drag. Is the error that big in a F1 Team?? I guess it's not... so, how big can it be?

I would suspect they struggle to get the turbulence right. This is steady state CFD modelling and there are numerous dynamic phenomena happening about a race car. The number of vortices, etc etc that they have to resolve and who says that there is a "steady" solution to this type of flow? They've also not specified the wheel velocity (if they can - maybe they could with Star 6?).

Interesting evidence of CFD not being the holy grail some make it out to be.

There have been plenty of teams been mired by faulty tunnel calibration including Williams, Toyota and Honda. So now it is Ferrari's term.

I wonder if the use of RANS rather than LES has a limit on how well the model resolves the drag (by limiting how well turbulence is resolved)?

Given the size of the RANS grid, I should imagine an LES grid was beyond their capabilities. You're right, however, I'm sure a good LES simulation would have given better results.

There's also not much talk of error in the presentation for either the physical or numerical model. There is just a straight comparison. I think their V&V could definitely be improved.

It's surprising that with such computing power they use a porosity model for their radiators instead of modeling the actual radiators.
There are many reasons the drag is so off, and it may not be the CFD itself, but how it is used. Assumptions and short cuts can really create a big disparity with results and reality.
3 hours to mesh is pretty long too. Goes to show the level of detail these guys are using.
20% is not something i think ferrari would have as a deviation though.

This article doesn't have enough information anyway.

JotaBass wrote:I've been taking a look on this article tweeted by ScarbsF1.
http://www.cd-adapco.com/pdfs/presentat ... /1-art.pdf
You can see more than a 20% of absolute error in drag. Is the error that big in a F1 Team?? I guess it's not... so, how big can it be?

+6s a lap, if you look at Virgin

I seem to remember something about the BMW cfd center using 16 processors and 50odd cores? That compares somewhat favorably to what these guys are using.

I'm not so sure the specific percentage of error matters as long as it's consistent. We know from fixed floor wind tunnels, which consistently underreport underbody and wheel effects, that useful work is still entirely possible.

I think current F1 big temas are around 1000 cores. They had more beofre limitations by the rules.

Big difference in drag prediction comes from RANS model. I do not think this can be circumvented even if they ran thier simulation with a much higher number of cells (100M-200M).
If it took them just 3 hours to mesh the entire car, their meshing tool is really fast!

shelly wrote:I think current F1 big temas are around 1000 cores. They had more beofre limitations by the rules.

Big difference in drag prediction comes from RANS model. I do not think this can be circumvented even if they ran thier simulation with a much higher number of cells (100M-200M).
If it took them just 3 hours to mesh the entire car, their meshing tool is really fast!

Presumably if they ran LES they'd get a better handle on the drag figure but at a somewhat higher resource requirement?

CFD has never been good at modelling drag, and it probably wont be for a while.

For racecar's of course drag errors are not so significant but for aerospace, most stay away from trusting the drag CFD outputs.

The modelling approach is an interesting one in F1...

You would hope one day there would be a system to model RANS for the car, whilst using unsteady LES for the exposed wheels. RANS for a 3D rotating tyre is bad enough imo and I would put the variance in drag mainly due to that.

Problem with LES though is that for downforce measurement it's the small scale stuff and it is mean values you would want. It seems to be all about compromise.

What puzzzles me is the question:
Correlation between CFD and wind tunnel results.

Why would you bother? Are they racing wind tunnels these days? The wind tunnel will ineveitably give results that have to be tweaked and fuzzed to correlate to
ONTRACK behaviour of the competition car and it´ss the same with CFD fudge factors..
CFD is another method to evaluate aero modifications and do aero research as
you do in a windtunnel .Both have their limits and both will not be entirely accurate.If you work enough in this environment and can VALIDATE your experimental data coming from the tunnel or CFD in your on track testing -AERO testing- you will over time find out that within set limits your cfd or tunnel work will correlate and you gain confidence in your judgement if things will translate to track behaviour without confirming it by actual tests.
Certainly there are limits of correlation and sometime your shapes will raise a new issue in reality not respected in your CFD or tunnel scaling and bang you are in the woods again.
Ferrari changing model size over winter and suddenly they cannot predict the worth of their work is telling you something...It´s empirical you don´t know when you leave the correlation window ...the longer you work in a useful frame the more precise your predictions will be=experience - not science.
the cloud those boffins are drawing can be quite big ,bigger than the car at times...

It's good for drag they just setup the thing too stingily for the computing power they have.
The radiator model is the whole problem with that simulation.

They could have a million trillion cells and it will still output garbage.

The model was setup to innacurately to begin with. They used porous surface, which is basically a solid surface that admits a portion of fluid and outputs a certain portion according to a mathematical model.
The model may be too simple and probably only holds under set conditions.

If you have that level of computing power, you should model the radiators fin by fin and all that is in the side pod and forget about porous surface boundary conditions.

That's where most of their error is.

ringo wrote:If you have that level of computing power, you should model the radiators fin by fin and all that is in the side pod and forget about porous surface boundary conditions.

It's already taken them 3 days to mesh it with a porous surface and you're saying they should model the radiator fin by fin and the complicated internal flow associated with such a mess? That would be over complicated for the sake of it if you ask me.

ringo you are wrong. The differnce between wt and cfd resukts has little to do with radiator modelling.
Even with higher computational powewr than theirs, you would not attempt a direct modelling of radiators; a lumped paramerters model is porven to work correctly