Tom Suddard
Tom Suddard GRM+ Memberand Director of Marketing & Digital Assets
3/23/22 4:26 p.m.

A phrase commonly tossed around at the track: “They’ve got tons of aero.” And while that may be useful shorthand, it’s a bit of a misnomer as every car has an equal number of aerodynamics: one. 

As in, every car affects and is affected by air as it travels. 

The laws of physics apply to every car equally, but when …

Read the rest of the story

Colin Wood
Colin Wood Associate Editor
3/24/22 1:56 p.m.

It's pretty wild how much computing power is required to generate the results.

Sure, I could probably plug some numbers into my computer, but I'd probably be waiting weeks before getting any results.

Still really cool, though.

tuna55
tuna55 MegaDork
3/24/22 2:06 p.m.

That's a neat explanation of how the car gets modeled. Are there DIY solutions? I've done CFD multiple times for internal flow, and each time been quite accurate. I've never had a reason to do external flow, but I suspect it is not much different. The model is the hard part.

VolvoHeretic
VolvoHeretic GRM+ Memberand Reader
3/26/22 12:48 a.m.

Nice article. Too bad I don't have $4000+. 

Tom Suddard
Tom Suddard GRM+ Memberand Director of Marketing & Digital Assets
3/26/22 7:12 a.m.

In reply to tuna55 :

Thanks, I'm glad you liked it. Sadly, there aren't any solutions to modeling that don't require time. You could do every bit of modeling yourself in free CAD software (here's my primer on CAD for first-timers), but the expertise and free time to do it is something I just didn't have. 

tuna55
tuna55 MegaDork
3/27/22 8:12 a.m.
Tom Suddard said:

In reply to tuna55 :

Thanks, I'm glad you liked it. Sadly, there aren't any solutions to modeling that don't require time. You could do every bit of modeling yourself in free CAD software (here's my primer on CAD for first-timers), but the expertise and free time to do it is something I just didn't have. 

I have tons of cad experience, but I think the main concern is the accuracy, especially on the little bits like panel gap and trim and such. Without scanning, just starting from scratch, I would have a lot of doubts unless you could confirm with a coast down or something. 

Tom Suddard
Tom Suddard GRM+ Memberand Director of Marketing & Digital Assets
4/6/22 2:14 p.m.

In reply to tuna55 :

Yeah, there's the rub. I've been working for more than a year on a story about how to quickly/accurately/cheaply scan parts at home, but it's proving incredibly difficult to get the resolution I'd hoped for. If/when I figure it out, though, I'll make sure to write about it. 

tuna55
tuna55 MegaDork
4/6/22 2:34 p.m.
Tom Suddard said:

In reply to tuna55 :

Yeah, there's the rub. I've been working for more than a year on a story about how to quickly/accurately/cheaply scan parts at home, but it's proving incredibly difficult to get the resolution I'd hoped for. If/when I figure it out, though, I'll make sure to write about it. 

I'm not at the tip top state of the art in that area, but I believe it (scanning accurately into 3D CAD) to be beyond the DIY, and likely more expensive than just renting a day at a friendly wind tunnel.

 

At a previous employer, we paid a scanning company to scan the job site into CAD. It was much bigger than a car, but much less detailed. I think the resolution was several inches, whereas for an accurate CFD of a car you would need two powers of ten better than that. It cost tens of thousands of dollars for the scan. One day on site and weeks of post processing. This was a while ago, but it's pretty far off of what I imagine the DIY racer needs.

stafford1500
stafford1500 GRM+ Memberand Dork
4/6/22 4:17 p.m.

It is not completely crazy to use 3d models made by others and found online. You will very likely need to modify/fix them for use as a CFD model, but it get you so much closer for not much effort. More detail can be generated using the photogrametry method Tom discussed, using free-ware and some compute time. All scanners produce is a set of surface points. The trick as he noted is a seasoned CAD driver to convert the "POINT CLOUD" into a set of useful surfaces. All CFD winds up using the surfaces to generate the bounding conditions. Gaps typical of production cars add lots of computational requirement, so bridge those gaps where you can, especially for a first draft.

On the tunnel versus CFD comparison: CFD gives lots of on AND off surface data and can be a very good tool for direction. If you need absolute pressures/loads/moments you will need to spend a good bit of time setting your model up. The compute time and the follow-on post-process willl likely be days for all this data for a single configuration in most DIY applications. Tunnel testing on the other hand only provide gross loads/moments of the entire assembly. Local pressures can be collected, but not to the resolution of the CFD. Test time for these results is minutes, vs days. The prep time is not insignificant and preparing a car for a tunnel test should be considered the same as preparing it for a track test - aero loads can be destructive quickly... Tunnel testing can be had for anywhere from ~$250 per hour up to the $1 per second rate. CFD testing is generally costed out as $ per CPU per computation time ($/CPU hour) and is not a small number in typical usage. It can be done on stout home machines, just budget days of computation time for even simple models.

As with any testing, a single test case does not provide much information. You MUST test multiple configurations to generate direction for good/bad, better/worse, etc. The definition of those is also not an explicit formula, knowledge of the end use AND/OR the direction you want to go is a requirement.

The real challenge of DIY CFD is the shear amount of control inputs you CAN use/tune to make the model operate. Getting a model to run is not the hard part. Getting a model to run and produce useful results is the trick.

I spend time doing both CFD and tunnel testing so I come at this from the point of view that CFD provides great directional and relational information, but tunnel testing can generate results much faster per part change.

Warlock
Warlock New Reader
4/20/22 2:50 p.m.
stafford1500 said:

...CFD provides great directional and relational information, but tunnel testing can generate results much faster per part change.

And that's the basis of a long-running gag dating back to when CFD was in its infancy:  you spent weeks building your model, an entire night running the code, and you had a handful of runs around a half-dozen airspeeds for one attitude and one configuration.  Meanwhile, the primitive wind tunnel guys next door finished an entire test and were building up the next one. :)

For those lamenting the cost and availability of CFD and wind tunnel time, there's still lots of utility in really old-school free-air testing.  An afternoon on track or an old runway, a couple of video cameras, and a well-laid pattern of cloth tufts or oil droplets can get you inside your working envelope of aero shapes and positions, and even help generate some calculated results to check against if you have the opportunity to build a computer model later...and if not, at least you have proof you're in the right ballpark.

stonebreaker
stonebreaker New Reader
4/21/22 2:34 p.m.
Warlock said:
stafford1500 said:

...CFD provides great directional and relational information, but tunnel testing can generate results much faster per part change.

And that's the basis of a long-running gag dating back to when CFD was in its infancy:  you spent weeks building your model, an entire night running the code, and you had a handful of runs around a half-dozen airspeeds for one attitude and one configuration.  Meanwhile, the primitive wind tunnel guys next door finished an entire test and were building up the next one. :)

For those lamenting the cost and availability of CFD and wind tunnel time, there's still lots of utility in really old-school free-air testing.  An afternoon on track or an old runway, a couple of video cameras, and a well-laid pattern of cloth tufts or oil droplets can get you inside your working envelope of aero shapes and positions, and even help generate some calculated results to check against if you have the opportunity to build a computer model later...and if not, at least you have proof you're in the right ballpark.

That, and maybe some of those suspension movement sensors I've seen advertised in GRM.  Use the suspension sensors to calculate the downforce generated by the different aero configurations.

leegrx7
leegrx7 New Reader
5/20/22 8:50 a.m.

Great info and well presented (i actually easilly followed the logic Thru the article) but the true bottom line on the detail of aero improvements is a really complex problem . Resounding success CAN BE  a cliff edge VERY CLOSE to serious problems/failure. The current Mercedes  F1 design is the prime example of the “cliff edge” of aerodynamic success. Their wind tunnel, their ultra aero expertise and their almost limitless pile of $$ has  pushed  their new F1 aero design over the line of success into a porpoise (sp?) mode that makes the race car almost un-drivable  at speed .The up and down oscillation frequencies  are litterally “hammering” the driver. I expect The team will eventually solve the problem but what a mess in the interim!

RodrigoVEng
RodrigoVEng
5/21/22 11:36 p.m.

In reply to Tom Suddard

Hi, nice work.

Have you tried 3d scanners like creality cr scan or any similar?

 

 

 

Bluebayou22
Bluebayou22 New Reader
8/20/22 8:01 p.m.

Great article thank you.  What if Morlind offered financially challenged trackers a basic package.  Now that they have your Z scanned, he could save a basic model without worrying about control arms and stickers and just give us basic mods such as a 3 x 7 inch canard angled at 30 degrees placed "here". 

And same for wing  and splitter size.  What we average peeps don't know is why have a splitter that is 3 inches long or 8 inches long?  Looking at the cars from this year's Pike Rally....... their splitters are longer than most wings.  

Just a thought because I would be interested in a basic aero package.  "Don't let perfection get in the way of done"

dean1484
dean1484 GRM+ Memberand MegaDork
8/21/22 8:12 a.m.

As someone that works with 3d AutoCAD the amount of data from a point cloud with that resolution would be large. With GPUs getting much cheeper and the 4000 series on the way I can see a time in the not so fade away future (maybe a couple years) where this will be very feasible as a DIY at home project in terms of modeling.  I would be interested to see what two 3090ti cards could do with crunching the data as they are the only cards with SLI still. Yes SLI is dead for most "normal use cases but in the commercial realm being able to tie together that much computing power could be very useful. I don't know if AutoCAD can use SLI. I do know it can be told to use either the cpu or the GPU for rendering so I am sure people out there that are way smarter than me with the program suite can most likely make it work.   

You'll need to log in to post.

Our Preferred Partners
yKt2LvOUrB16qV84540ZmKu1RI0lwVknEiCKiHs9Zznj0wnu3xVV1uclkEYeX0ST