Post anything that doesn't belong in any other forum, including gaming and topics unrelated to motorsport. Site specific discussions should go in the site feedback forum.
Hmm, since they've admitted to writing off their first 3 years to playing catchup to the established teams, their strategy may be to save wind tunnel money for when they have enough knowledge to maximise its effectiveness.. 2013 may be their first serious attempt at F1 with validation
Any post(s) made by this user are (semi-)educated opinion(s), based on random fact(s) blurred by the smudges of time.
Any fact(s) claimed by this user will be supplemented by a link to the original source of said fact(s).
So many of you must be familiar with Horace Lamb's quote - "I am an old man now, and when I die and go to heaven there are two matters on which I hope for enlightenment. One is quantum electrodynamics, and the other is the turbulent motion of fluids. And about the former I am rather optimistic."
Turbulence is totally, poorly understood, and a complete bitch to simulate. Chris, please consider putting the technicalities of your post back up. It was well put and succinct.
autogyro, consider what you really think you're simulating in CFD, and what it takes to do so. Teams have big supercomputing resources, but they're finite, and lots of concepts to evaluate. Direct numerical simulation for a car at speed is impossible, we're at the heady age where it takes a moderately sized supercomputer to simulate a cube mounted in a ground plane in a relatively low-speed freestream. There are going to be shortcuts if you want to simulate flow around a car in anything less than a few years, seriously.
So you take shortcuts and you end up with a simplified turbulence model. And that gives you... flow around one car, at one heave, for one wind condition. I've never seen a race car do a whole lap as if frozen in space and time and never taking a corner. It moves, violently at times. So to parametrize what a car does on track you might want to simulate a few different heave positions, yaw, whatever else. Think of the possibilities testing a few static positions - reflecting braking into a corner, turning through it, accelerating out of it, high speed running, whatever. You'd need a few positions for each scenario if you were assuming everything behaved quasi-statically and without any hysteresis. So far, in CFD, for the one car configuration we're testing >10 positions. You want to add potential wind effects? Include more flow yaw, more CFD tests for the same condition. Headwinds/tailwinds? It goes on and on. And you can multiply that for every new part tweak you look at.
How quickly do you think you could run through a parameter space quasi-statically in a wind tunnel environment? Pretty damn quickly. You could get some idea of other factors too, hysteresis etc. So you'll develop a concept far quicker in a wind tunnel than you will in CFD, anyday. Wirth can say what he likes, and he should, as he's got sponsors and creditors to appease.
I've even seen some good presentation of research organisations connecting shakers to their stings and comparing quasi-static test results to fully dynamic shaking of the model. There are some good differences in data. Other organisations have devices in the tunnel able to simulate changes in upstream turbulence. Real cars on road don't have stings bolted to the top of them to hold them in place, either, and there's a lot of good ongoing research on magnetic levitation of test models. Research into all these technologies is all very new, very current and not yet at a commercial stage (e.g. you won't see it in F1 for a while). You can assume at the very least that there's a good bit of development yet in increasing the fidelity of the wind tunnel environment to the practical environment.
But so far, it's far closer to the practical environment than anything possible in CFD for a shape of that complexity that's deliverable in a reasonable timeframe.
So why have CFD?
It's a complimentary tool. You can do a lot in CFD that you simple can't do feasibly in a wind tunnel. You can try shapes before building them. You can get an idea of flows you don't have the facility to test for in an accurate, repeatable manner. Most importantly you can interrogate flows and obtain data that are not feasible to acquire in a wind tunnel.
If you're looking for force data and surface pressure, you can get this data pretty quickly in a wind tunnel environment. Flow field interrogation is another story. It's time consuming for one. The quick methods (PIV) aren't suitable for large installations and are a PITA in an air-based environment where run times are relatively short (they're also not especially developed for dynamic flows). It can also be intrusive, changing the flow field and introducing bias. Point-based inspection (pressure, hotwire, whatever) does the same, and to interrogate a large area is not time effective at all. LDA isn't intrusive in the slightest, but detailed tests take hours. Put this into a dynamic and not mean-static context and it becomes far more complex; in many cases the dynamic behaviours of aerodynamic phenomena define the mean effect (a Gurney flap is a good example).
With CFD on the other hand, you mesh, you run your case, you get data out. In post processing you can interrogate whatever you want. Used correctly, one can elucidate or validate key phenomena in flow that weren't previously visible. There are basic, simple, reference aerodynamic shapes who's flow structures are being better understood today thanks to very accurate, very time consuming - unfeasibly so in anything but research environments - computational methods. New, key flow structures are being identified on a scale and/or in complexities not possible with experimental tools.
However time is of the essence in a commercial or otherwise competitive environment autogyro, and Chris is right. The more simplified your turbulence model becomes away from full NS, the more constants there are to assume in any given model. Knowing what to use and why for a given stage of a given solution is something a good CFD operator will know to do, however being able to prototype a solution exactly is largely impossible, and thus the need for good assumptions. There are very few explicit solutions for NS equations, if there were more Wirth could design his car on the back of an envelope and the CFD industry would go to ---. But there aren't, and trained assumptions are as good as it gets. I understand where you're going with this - for every good CFD operator that knows the maths and implications behind his turbulence models there's at least 100 that knows fck all and generates solutions not much better than bum fluff - but even for the best (even for those authoring turbulence models) part of the art is knowing what to assume and why.
In that (and that for some cases of solution with some turbulence model, exact solutions are impossible) validation is a key requirement of any CFD program. The CFD program can then work successfully exploring concepts, knowing that the computational research program bears good resemblance to practical flow. It also allows greater liberties to be taken with shortcuts, which is important - once you've hit a critical relevance to the practical concept and understand the data you require specifically (qualitatively accurate flows? exacting loads? etc) then you engineer the CFD model for computational throughput... and get more done. There are ongoing, very serious efforts to validate CFD for very simple 3D shapes, far less complex than an F1 car (many good papers explaining the complexity of as much available for free too).
If Wirth isn't bullshitting about his model never having seen the inside of a tunnel, then it has none of these advantages. Unless he has a supercomputer tucked away a few orders of magnitude greater than his oppositions' (would seem a waste if he intends on joining FOTA, and you'd know about it on the '500 register) then his team is simply not running higher-order turbulence models, and his team isn't going through nearly as many concepts as his opposition's.
If it's true they're not using a wind tunnel currently, it doesn't exclude the possibility of not ever having done so.
Some information i found reading about the Acura ARX-02a...
I think that maybe all this digital solution, "only CFD" thing is just marketing and they will use windtunnels after all...
Acura's LMP2 car was widely hailed for its use of virtual development tools, and while the LMP1 car mostly followed this trend, physical testing of a wind tunnel model was implemented.
"We did a total of maybe 15 days in the wind tunnel, which is tiny compared to the amount of aero testing done in CFD. But certainly the wind tunnel was involved in the early aspects of the car"... (snip)
"As you know, the LMP2 car was completely developed in CFD. We had a wind tunnel model of the LMP2 car from back in 2006. What we decided to do was to convert that wind tunnel model into a development model for the LMP1. And so the LMP1 car in the initial design phases was done in a joint program between CFD and scale model. CFD was looking at some areas, scale model was looking in other areas. It was simply a matter of resources."
"That decision was taken because we could not replace...we had not got enough resources to replace the wind tunnel aspect of what we were doing in the early phases with CFD.
Essentially, it would have required another 50% or 60% more CFD capacity than we had. And so we took that decision to do a split program. But very quickly, within the program, we stopped. We got the information we wanted out of it, but then we stopped it. And then, certainly for the bulk of the program, we did everything in CFD."
Thank you SZ very educational.
It proves one of my theories that to build an F1 team, you simply need enough money and the right people working for you.
Wanna job?
autogyro wrote:Thank you SZ very educational.
It proves one of my theories that to build an F1 team, you simply need enough money and the right people working for you.
Wanna job?
Depends on the offer... are we starting the F1Technical F1 team? Because we'd need to put the tech director role on timeshare to avoid a total bitchfight
pipex has a bit of a point... seems Wirth has a wind tunnel, or at least access. Question is whether they've at least done some basic evaluation work.
This and this is interesting... capabilities are a bit vague though. They've done a bit of recent spending but you'd wonder how close it comes in capability (not mentioned) to other team's efforts in capability... hell of an investment all the same.
Can anyone put numbers to what kind of punch they're packing?
I would like to post some on this but I'm afraid I'd say too much.
What people must bear in mind in Mannor's case is that the complexity and dependancy of a Le Mans and Formula 1 car are totally different.
Its like saying we developed the aero of the road car fully in CFD, you wouldn't notice the different in your average road car whether it was done in CFD or wind-tunnel because of the aero dependancies.
Likewise, the Le Mans car is not nearly as complex as an F1 plus the aero is not as dominant.
CFD is always evolving, I wish people would stop talking it down simply because they don't understand it.
It is similar to measuring something, there is a trade-off between what accuracy is required and what can be done.
I write turbulence models occasionally, they are very specifically tailored to our needs, both in terms of detail and how they tie in with our infrastructure.
And for people that don't appreciate, meshing is a hugely important aspect of simulation, or distretization, whichever may be applicable.
Imagine trying to surface mesh each surface in a whole car model, we don't just click on a face and it meshes, that's cetrainly not sufficient.
Setting up relevant zones and surfaces also takes a vast ammount of time.
Any questions without being too specific about the work are welcome.
Ciro Pabón wrote:As I simulate traffic, the behav!ior of individual cars can be faithfully reproduced, throwing an random number to represent the aggressiveness of the driver: the behaviour of a driver when he follows another car and the opportunities or gaps he accepts to overtake are well known.
You need to use a network model.
Background info - My company simulates people flow for major public events, including how they jump a queue, or decide to go to a different queue. We used to use a mesh approach with individual people. It was a bit like a very coarse CFD model with very large particles. We now use adaptive network models so if we see that flow goes from A to F, we model a network from A straight to F, not individual cells for A, B, C, D, E, F.
You can also use this to optimise static structures. I'm not sure it would work with air flow, although a colleague has used a similar approach for large eddy flows and got very good correlation with conventional CFD and wind tunnels around buildings. (Large eddys use a fraction of the computing power of CFD meshes)
F1_eng wrote:Its like saying we developed the aero of the road car fully in CFD, you wouldn't notice the different in your average road car whether it was done in CFD or wind-tunnel because of the aero dependancies.
There's plenty of very serious road car aero development done in wind tunnels because it simply can't be done in CFD. If you're trying to use this as an analogy by degree, I understand where you're going, but seriously, road cars and open wheel race cars are completely different problems aerodynamically. There's a lot on a road car that you can't do accurately in CFD.
F1_eng wrote:CFD is always evolving, I wish people would stop talking it down simply because they don't understand it.
It is similar to measuring something, there is a trade-off between what accuracy is required and what can be done.
You could probably take every argument questioning the accuracy of computational methods, swap 'CFD' with 'wind tunnel' and still have the statement ring relatively true.
Both seek to repeatable approximations of a given practical domain within known limitations, and both are evolving.
F1_eng wrote:I write turbulence models occasionally, they are very specifically tailored to our needs, both in terms of detail and how they tie in with our infrastructure.
Of interest - in which languages? C/C++? Fortran?
F1_eng wrote:And for people that don't appreciate, meshing is a hugely important aspect of simulation, or distretization, whichever may be applicable.
Imagine trying to surface mesh each surface in a whole car model, we don't just click on a face and it meshes, that's cetrainly not sufficient.
Setting up relevant zones and surfaces also takes a vast ammount of time.
Granted.
I only dabble in CFD but possibly (I'm actually getting into a very new meshing package myself) those regularly meshing could give their experiences on available packages for those interested - would go a long way.
If this could extend into recommendations for suitable paid and freeware meshers, all the better.
It's all relative, what do you mean by "very serious aero development" SZ?
I can guanrantee its not 24 hours a day, 7 days a week, 360 days a year, which is what most F1 wind-tunnel run. all of the time on the same vehicle. Incase anyone mentions the rules regarding wind-tunnel testing limits each year, the rules are hopeless.
"but seriously, road cars and open wheel race cars are completely different problems aerodynamically."
Please, I don't need you to tell me. Sure there are things on a road car you can't model very accurately, but its still accurate enough. One of the biggest differences in F1 and road car aero is aerodynamically generated acoustics which I admit, can be a very difficult area.
Wind tunnel testing is fairly solid these days, we achieve great repeatability in the tunnel over a very long period, and tunnel results transfer very well to on track performance.
The key is to keep on monitoring every possible parameter of the tunnel over long periods to keep an eye on features that can drift. And have very good data storage to compare to previous data if things are suspect.
For sure its not replicating the track, but thats not the intention.
C/C++ are my prefered language.
What meshing package are you learning at the moment?
There are a lot of very good meshing packages out there, most of then often unheard of.
Serious road car aerodynamic development doesn't necessarily imply that a comparable amount of time is spent on a single project relative to F1. Time doesn't define the seriousness of an aerodynamic project. A road car project, by it's very cost nature, doesn't feature nearly as much test matrix parametrization. However a great deal more flow interrogation is generally undertaken, for which a far greater depth of analysis is undertaken.
Aeroacoustics was indeed where I was going with this primarily, for which CFD becomes a very limited tool; throughput (let alone accuracy) is greatly heightened when testing experimentally. Testing simulated realistic upstream turbulence is another area where road car aerodynamics currently explores very realistic aerodynamic problems not quite possible to do feasibly in CFD, but that has shown to be an area of growing prominence and importance for road car manufacturers.
F1_eng wrote:Wind tunnel testing is fairly solid these days
As it has been for some time. Wind tunnels are just a controlled industrial environment, after all... the most impressive of which are not used in F1.
F1_eng wrote:For sure its not replicating the track, but thats not the intention.
Mentioned this earlier - it's to provide the most repeatable environment with the best relevance affordable.
I'm doing a little part time development work on something new. I'm not the developer, so it's not for me to divulge, but it's got some nice features. Otherwise I'm familiar with the usual suspects, though I'm keen to see what people use that's freeware, as the tools I work with I certainly can't afford for personal use. I would imagine many are in a similar position.
For example, I've seen Gmsh but am yet to play with it; any feedback would be great.
