CFD - Computational Fluid Dynamics, Motorsport, Formula 1

[dren]

Unlimited CFD for 2021? Should be interesting.

http://www.racecar-engineering.com/arti ... mited-cfd/

I see $$$ and a widening gap for the top spenders if the cost cap isn't put into place.

[DarkNight00]

Thanks

[jjn9128]

Unlimited CFD for 2021? Should be interesting.

http://www.racecar-engineering.com/arti ... mited-cfd/

I see $$$ and a widening gap for the top spenders if the cost cap isn't put into place.

There's not much detail in that article as to what this may be... I read it as a situation like the development for the 2019 rules where teams were allowed to run CFD outside their normal car development to feedback data to FOM/FIA on the progress of the rules. So FOM/FIA will send teams a draft rule or a particular geometry which they're free to simulate so that the FIA know what sort of numbers and wake profiles teams may be getting - but beyond the definitive rules for 2021 the CFD limits are back in effect. I could be wrong though (and I usually am).

[wpsiatwin]

Unlimited CFD for 2021? Should be interesting.

http://www.racecar-engineering.com/arti ... mited-cfd/

I see $$$ and a widening gap for the top spenders if the cost cap isn't put into place.

There's not much detail in that article as to what this may be... I read it as a situation like the development for the 2019 rules where teams were allowed to run CFD outside their normal car development to feedback data to FOM/FIA on the progress of the rules. So FOM/FIA will send teams a draft rule or a particular geometry which they're free to simulate so that the FIA know what sort of numbers and wake profiles teams may be getting - but beyond the definitive rules for 2021 the CFD limits are back in effect. I could be wrong though (and I usually am).

Would most teams be using openFoam with custom solvers?

[jjn9128]

I see $$$ and a widening gap for the top spenders if the cost cap isn't put into place.

There's not much detail in that article as to what this may be... I read it as a situation like the development for the 2019 rules where teams were allowed to run CFD outside their normal car development to feedback data to FOM/FIA on the progress of the rules. So FOM/FIA will send teams a draft rule or a particular geometry which they're free to simulate so that the FIA know what sort of numbers and wake profiles teams may be getting - but beyond the definitive rules for 2021 the CFD limits are back in effect. I could be wrong though (and I usually am).

Would most teams be using openFoam with custom solvers?

Nope. Most (maybe all...) teams use Star.

[wpsiatwin]

There's not much detail in that article as to what this may be... I read it as a situation like the development for the 2019 rules where teams were allowed to run CFD outside their normal car development to feedback data to FOM/FIA on the progress of the rules. So FOM/FIA will send teams a draft rule or a particular geometry which they're free to simulate so that the FIA know what sort of numbers and wake profiles teams may be getting - but beyond the definitive rules for 2021 the CFD limits are back in effect. I could be wrong though (and I usually am).

Would most teams be using openFoam with custom solvers?

Nope. Most (maybe all...) teams use Star.

Interesting, what are the reasons for this? Also how different to openFoam is starCCm and how difficult is it to learn?

[jjn9128]

Would most teams be using openFoam with custom solvers?

Nope. Most (maybe all...) teams use Star.

Interesting, what are the reasons for this? Also how different to openFoam is starCCm and how difficult is it to learn?

Not sure. A lot of teams use Siemens NX so there's an easy work-flow from there. That said McLaren and Renault use Catia. I'm not sure if teams use the built in mesher - but OpenFoam only has a very simple block mesher so do get a nice mesh requires external software, again that might not be any different to the internal practice teams already have. It could be that 15-20 years ago Star really pushed their product for the marketing angle - "F1 teams use it".

I wouldn't say OpenFoam is easy to learn... it has no GUI so from that point of view you have to create text files to do all the work. But if you have an understanding of CFD then you can fairly quickly get the idea of what it's all doing.

[wpsiatwin]

Nope. Most (maybe all...) teams use Star.

Interesting, what are the reasons for this? Also how different to openFoam is starCCm and how difficult is it to learn?

Not sure. A lot of teams use Siemens NX so there's an easy work-flow from there. That said McLaren and Renault use Catia. I'm not sure if teams use the built in mesher - but OpenFoam only has a very simple block mesher so do get a nice mesh requires external software, again that might not be any different to the internal practice teams already have. It could be that 15-20 years ago Star really pushed their product for the marketing angle - "F1 teams use it".

I wouldn't say OpenFoam is easy to learn... it has no GUI so from that point of view you have to create text files to do all the work. But if you have an understanding of CFD then you can fairly quickly get the idea of what it's all doing.

I have seen Mercedes ask for experience with Catia on job advertisements so I’m guessing they also use it. With openFoam what about snappyhexmesh?

In industry is it not common to use text files if a GUI is available?

[jjn9128]

I have seen Mercedes ask for experience with Catia on job advertisements so I’m guessing they also use it. With openFoam what about snappyhexmesh?

In industry is it not common to use text files if a GUI is available?

SnappyHexMesh is the really basic block mesher I was talking about. I would say ANSA is the industry standard for surface cleaning and meshing, certainly in the automotive industry, but I seem to recall star has some geometry cleaning tools in the pre-processing suite.

For ease of use and work flow I'd say a GUI helps... but everyone's different. You may find some people prefer to work from a command line, while others like to click through.

[FW17]

With a lot of CFD happening along with wind tunnel being used to verify, when will teams start developing AI to do the CFD process? or is AI already in use?

https://www.cfd-online.com/Forums/main/ ... e-cfd.html

[wesley123]

I personally believe that machine learning is in use. For example, Mercedes; Their car is so incredibly detailed, it is still years ahead in that regard. Also, their ability to develop, or counter others, is so incredible, I can't think of anything other than ML going on.

[Vyssion]

With a lot of CFD happening along with wind tunnel being used to verify, when will teams start developing AI to do the CFD process? or is AI already in use?

https://www.cfd-online.com/Forums/main/ ... e-cfd.html

In an ideal world, teams would use GPU solving techniques to speed up CFD massively, and then just create a design search and optimization matrix of all sorts of parameters and then hit go, and wait for it to spit out the best design.

The issue with this is mainly just due to 3 things: First, rules prohibit the use of GPU's. Second, rules prohibit the use of massive amounts of CFD solver time. And thirdly, the reason for these two issues, is mainly driven by cost. GPU's are ridiculously expensive when you get up to the enterprise grade stuff, and racking up hundreds or thousands of them to solve massive amounts, gives a substantial advantage to teams who can afford such installations.

With even more budget cuts coming in 2021, I very much doubt that you will see anything like that for a long time. It is true that the world of F1 is beginning to favour CFD more so than Wind Tunnel - again mostly due to cost - but it is important to know what the limitations of each analysis technique is. CFD can show you things you simply cannot measure in a wind tunnel but suffers from numerical and correlation issues. Wind Tunnel suffers from scalability effects, but it is real air - so what you measure, is what it is. On track testing, again, has its own limitations, but it gives you the best data of the kind it can give you. CFD, however, is very much on the cheapest side of these three techniques once the initial cost of setup is done. Hence why teams will begin to favour CFD more in the future with these budget cuts, but with the rules still strictly prohibitting larger teams from gaining a massive advantage over the lesser ones.

[FW17]

With the processing power massively increased over the last few years with introduction of EPYC chips and now the new Nvidia Ampear, when will FIA increase the 25 teraflops limit that it has imposed.

As part of cost reduction, should F1 teams set up their own data center together created specifically to their needs which all teams can then use together?

[Just_a_fan]

I doubt the teams would trust each other. Imagine one or more teams running naughty snooping software on the server to spy on the others. You can bet the teams would imagine that and so no trust. No trust, no go.

[FW17]

This is mainly for AI mainly, but has a powerful compute side too.

WRT security, I doubt the teams would actively employ people to snoop into other teams data.

Some of these central infrastructure could help in reducing costs.