I probably created a revolution in aerodynamic testing

Triathlon Forum
1

UPDATE: I’ve reached my max for replies as a new user and must wait 6 hours (it’s 10:23 am CST here) I’ll respond later when I can.

First post here and I couldn’t think of a better community to discuss this creation with.

The idea is, from a 20 second video with any phone I can generate a 3D model and then perform a full-on CFD test to determine, with seemingly quite high accuracy, CdA (aerodynamic drag)..

I have a service that has just barely launched (but I’m not posting a link unless given a greenlight, as I read the rules very carefully, it is in my profile though).

wind (1)

Oops, it appears since this is my first post I can only add one picture/gif. I have like… 30 more with explanations and 3D models that seem to embed directly into this type of forum, but perhaps I can answer questions with images.

So, what is it? This service (really a Shopify website with an upload form), which uses technology that barely exists in a commercially viable form (most of it is still cobbled together from various research projects, cutting edge AI (NeRF) etc., but it works), enables anyone with a standard camera, to be filmed by a buddy while walking around them in one circular loop, with emphasis on the frontal area, to then be tested by the most powerful computational fluid dynamic software I could get my hands on (Open Foam) at 7.5 degree yaw and 22.3mph windspeed.

How accurate is it? Seemingly shockingly accurate, finding a .29 watt loss by removing cables from the front of a bike. It finds a 2 watt difference for every 5mm of additional tire diameter.

Why aren’t I showing a TT bike? I have more gravel bikes than TT bikes, the reason I built this was to optimize my and my roommate’s setups, and I ended up automating most of it, and made a whole service supporting it, with one roommate supporting the server (he’s a sys admin), and the other building the website, he’s an aspiring frontend engineer.

As for me, I’m Eric Semianczuk, I work full time and build a stupendous amount of software projects aimed specifically at cycling in my free time, all of them free, except this one, as it has some manual work here and there. The others range from a world routing site for cyclists, with all sorts of fun features, like calculating tailwind, showcasing weather systems, etc. to a giant cycling physics simulator that works on mixed surfaces as well as whole Ironman courses, predicting everything from speed, to electrolyte loss.

I have completed a Full IM (Chattanooga, really helped with my sub-par swim), have a TT bike, and have done tons of Tris, I’ve just caught the gravel bug of late and might try an Xterra Tri at some point.

All of the tools I use for this cost me/us nothing, either Open Source or my custom tools, and the price reflects this (I don’t know if it’s okay to mention price, I just launched, days ago, and it’s extremely cheap at the moment, and will adjust as needed).

As an interesting case study, our first user (so exciting!) tried three different positions, two of them were only a few watts a part, and one was over 30 watts faster than the rest two! When you try to be aero, it can seriously make a difference.

The goal here is to get some feedback, I’m no salesman (obviously, look at my ramblings), I’m a developer/triathlete/gravel cyclist/terrible mountain biker. If anyone has business model advice or performs bike fittings or anything of that nature and want to connect/partner, let me know, all I want to do is grow this idea.

UPDATE: I’ve reached my max for replies as a new user and must wait 6 hours (it’s 10:23 am CST here) I’ll respond later when I can.

8 Likes
2

Hey Eric,

That is really cool! I am doing similar work and also believe CFD offers tremenduous potential for athlete optimization.

There are a few similar services like
STAC (STAC Virtual Wind Tunnel)
Falcon Pursuit (https://falconpursuit.com/)
and a company in Belgium that has slipped my mind.

Most world tour teams utilize a service like this at this point, so you are on a good track. I think what is interesting in your case, is that you offer the scanning as DIY service instead of having to go to a location.
Have you tried scanning the same position multiple times to see waht the resolution is?

Always happy with an exchange of thought. Great direction!

Ingmar

1 Like
3

Yes, but more in-so of taking the same scan and making high-resolution and low resolution, as well as slightly perturbing the the model to mess with it:

output

the difference always come out to be a bit less than 2% variation.

Also, I don’t use the full video’s frames, really just around 300 frames, so I have sampled from different offset frames and easily been within 1% repeatability.

2 Likes
4

The regular price is $149 and you get how many positions/gear changes for that price?

Default is 7.5 degree of yaw. What percentage of time in real world do you find that yaw angle

And is that both left and right?

5

The price is floating, it’s currently $30, and may end up being $50, if it gets overwhelming, I’ll either cap the amount of “stock” (pending training my teammates on the pipeline/touchups), or up the price (my probably smart marketing guy wants it higher, like $150, but I like the idea of general availability, I could also offer different levels of resolution, etc., I have the final say in these matters, but that’s why the site says that.).

So, right now, $30 1 test, you are provided a form, via email when ordering, in there you can specify any changes to the default, different windspeeds, different yaw angles, it would take a small amount of additional effort to do left and right.

Currently, I simply rotate the model 7.5 degrees to the right to simulate that yaw.

Regarding the yaw angle, one of our teammates has extensive experience with cyclists in many wind tunnels and broke down that that particular angle was one of the best, especially for TT for show subtle differences between changes in gear/position/etc. and suggested we offer it as a default.

The current plan is to get enough sales to get this vs a real Wind Tunnel validation.

1 Like
6

Very cool. Excited to see how this gets developed.

Are you looking for investments in parallel with the app development? Or just really a hobby for now?

7

It does exist, just barely launched, I just didn’t want to violate any forum rules by posting a direct link in my article, I was also only limited to one photo, as this was my first post here (fair rule, I’d do the same if I had some a nice forum).

If this isn’t allowed, just tell me and I’ll remove it, I’m just excited to talk about it, I have a marketing guy who’s working on advertisements and such, so, that will be marketing this I see really as a discussion into the idea, I’ll take any suggestions, insight, etc. happy to give out free promo codes if anyone wants to try it, I’m just excited to start running more videos through the pipeline! The site is https://wind-tunnel.ai.

In addition feel free to abuse the heck out of my totally free other side projects:
https://sherpa-map.com (world wide routing website to compete with Komoot, RideWithGPS, etc.)

You guys in particular may find this site interesting too, it’s an extremely powerful physics simulator for TT/Gravel/MTB courses:
(removed link do to new user restriction (can only have two))

I even pre-loaded the Kona course. Play around with body size, FTP, tire width, bike type, etc. run the sim and see how fast you’ll be, run it with weather and it will even take into account forecasted weather.

Again, totally 100% free, I don’t make a dime off of it, both projects sit on a server in my apartment basement.

Heck, to make a point, I even scanned an entire race course’s sat imagery in order to determine where there would be exposure to the sun (using custom trained AI), then added forecasting… turned into a wet sandy race that destroyed my drivetrain, but I finished!
(removed link do to new user restriction (can only have two))

These are some of a pretty large amount of projects I have lying around… I’m even working text prompt to route i.e. “give me a route that is 20mi long, stops at a gas station, goes predominantly north, loops back to my location, and has 2000ft elevation gain”, it’s a WIP.

Back to your specific question. I’m just trying to leave my job, build a software company in at least one of these or other, unmentioned (you don’t want a novel) pursuits.

I’m currently the definition of “coasting” in my current job, not even software engineering, just “Data engineer”, it’s extremely easy, automated everything, but I really want to make at least one of these “hobby projects” into the real deal (not that I don’t try to move up, there’re never any software engineer roles open and I don’t have time to apply elsewhere).

So, would I partner with an investor? Sure, if I can build these things while staying employed, no idea what I’d be capable of if my only job was to build and improve them. Additionally, if this one takes off, it’d likely require more workstations and some employees for 3D “touchups” (I’m a sucker for perfection).

I’m also trying to collect funds to get access to a wind tunnel and derive two things A. what percent difference are we from the WindTunnel, and B. how perfect do my models need to be?

Upon checking out STAC they show imagery of their models, even missing decent portions, and claim to be within 2% of a WindTunnel (could be, the body means so much and the bike is mostly there, I’ve gotten similar results through testing higher and lower quality models)?

If that’s the case,… my process can even define mesh to the point of seeing the knuckles on someone’s hands:

image
image951×772 34.3 KB

STAC, not so much:
(couldn’t add image because of new user restriction (can only add one))

Nothing against STAC by the way, seems like a great company, they even proved the business model.

PS. I don’t have time to re-edit the whole reply, so I removed what I could, link and picture-wise, to get this through.

1 Like
8

Pretty sweet and I’ll probably give it a go. One thing I’m not clear on is how many variants do we get to test for the $30? Is it just results for a single position or do we get to upload a few to compare against each other?

9

^^^^^That’s the product to market. I really want to send 3-4 different setups (ex. different bottle positions or bar tilts) and see the deltas.

Would it also help if there was a smoothing algo post processing the scan so we’re not testing the “Rodin” version of the rider?

3 Likes
10

This is SO cool! Awesome work. How much space do you need around the model? My bike is currently about 1 arm length from a wall. I’d love to get an idea of this and see how it works. If it’s just one position though on this early testing, there’s no comparison for ourselves right?

11

You can compare yourself all you want, the idea is, video, recorded from someone walking around you in a circle for position 1, then, another video, for position/gear change 2, then position/gear change 3, etc.

This service gives a clear CdA as well as the amount of watts needed to sustain said speed, so, you could take a video, get a CdA and watt # and try to make it lower video after video.

1 m on one side, still doable, if I couldn’t make a model from the footage I wouldn’t charge, even if I messed around with it manually for a bit trying to get it to work, there’s still a good chance I’d get a good quality model.

12

I am also not clear on what is included for the $30

In a wind tunnel or outdoor aero test session it is not uncommon to test 25 or more variations.

Such as 5 dofferent helmets. A few trisuits. A few calf sleeves

A few different positions

A few different bottle setups

And then to try different variations of these

So let’s take a wild guess at 30 variations

How much would that cost $30?

Or 30x$30?

Thank you

13

What’s your validation of these results? You mention removal of cable housing, but no real validation testing (items with known coefficients, vs wind tunnel, vs Chung method, etc.).

Model quality here looks very coarse, maybe good enough to estimate body position, definitely not refined enough to evaluate helmets, bottles, etc.

You’re talking about charging money for something that is historically hard to get right. And something where being a little off could have people spending real money.

14

Totally understand where you’re coming from. Right now, my models appear to be far superior in quality to the mentioned STAC Virtual Wind Tunnel, and they claim to be within 2% accuracy of a wind tunnel,

Their models:

image
image1383×599 128 KB

I have an aerodynamic specialist on board who’s worked with cyclists in wind tunnels for years, every test we’ve performed is within expectations.

So, my models are at or better than Lidar quality from the competition (I went to the ends of the earth to build a non-existant points to polygon mesh algorithm that doesn’t exist elsewhere), they claim 2%, if I work to create very high-quality versions of my models, with every wire, spoke, etc. I have to degrade quite a lot to resolve a model outside of 2% of the high quality one in terms of different derived CdA values.

Given this, and my… many other tests… it would simply appear that yes, the body, body position, form of the helmet, bicycle, etc. contribute far more than anything else aero-wise.

Our first customer, a multi-time top 5 finisher at Garmin Unbound XL recently used it to hone his position, the two mildly relaxed but different positions barely made a difference, the shift to a true, hard to hold aero position was monumental… was about to upload imagery, but I’m at the limit of having a new account, one link, one photo. Oh well… … and I just got permission to use it… with numbers and everything.

15

One test gets you this:

output_2_0
output_2_01920×2485 301 KB

Additional tests gets you iterations to find your preferred form.

The tests are not fully automated, take an extremely powerful ($11k computer) 40 minutes of compute time and around 10 minutes or more of hands on fiddling to make it work.

It is not trivial, and it saves $$ in gas/petrol for driving to the nearest center that would offer such services with lidar, and doesn’t even compare, price wise, with a wind tunnel.

16

Here’s a compressed (colors and size) version of the process with time wrapping for the boring parts:

output

1 Like
17

6 significant digits?

18

this is all great. Can you please post the link for: “this site interesting too, it’s an extremely powerful physics simulator for TT/Gravel/MTB courses:” - it was cut off your last post due to ne user restrictions. thanks

19

enjoy

20

Seven.