Bunch of Aero tests! Tons of Gravel Wheelsets, and some outlandish ones (for fun!)

[Eric_Semianczuk]

multibike

I tested the same scan of the same rider/bike (5’7" 145lbs rider, size 51 felt breed with fox sus fork and dropper) with a variety of different wheelset combos, all with 45mm tires + 40mm external rim width, but with varying rim depths (I also enjoy making 3D art, and threw together the above quick animation showcasing the models, red is slowest to green which is fastest).

I’m the creator of the site, https://wind-tunnel.ai, which has gone from only an idea (which I first mentioned here, when it was barely a concept) of a quick “20 second video of a cyclist on a trainer to a 3D model to an Aerodynamic Drag test (using CFD)”, to a full on beta service!

Testing around 22.6mph/36.4 km/h (10m/s wind at 7.5 degree yaw approximates to 22.6mph “effective” frontal wind velocity) deriving both the CdA (smaller is better) and “Watts it takes to sustain said speed, not accounting for drivetrain loss/rolling resistance” in the “Wind Tunnel”:

(That’s just a stylized render I made in a program called “Blender3D”, I am testing the same bike/rider models though, the real CFD test in OpenFoam just uses a big boring rectangular prism)

Depth Front	Depth Rear	CdA	Watts (needed to sustain “sustain” 22.6mph/ 36.4 km/h)
17mm	17mm	0.272027	170.967
60mm	17mm	0.270629	170.088
60mm	60mm	0.264655	166.334
60mm	180mm	0.263797	165.795
60mm	Full Disc	0.263488	165.600
180mm	Full Disc	0.257849	162.056
Full Disc	Full Disc	0.238443	149.860

Now for the whacky one, I couldn’t help myself…

I tested a REVERSE MULLET Double Disc Wheelset!
26in aero gravel disc in front and 34in aero gravel disc in back!
CdA: 0.231983
Watt: 145.800

Here are some shots from a visualization tool, ParaView after testing:
Note: I tested more than these, and some are pictures from “both sides”.

17mm depth, both wheels

60mm front, 180mm rear

60mm front, full disc rear

other side of the last one.

Now we’re getting a bit silly, but don’t you want to know?? 180mm front, full disc rear.

TWO DISCS, 40mm wide, with 45mm tires! The DREAM, fast, unstable, and what… could… be faster and more unstable… hmmm…

THE REVERSE MULLET, 26in aero gravel disc in the front and 34in aero gravel disc in the rear!
4.06 watts faster at mortal-ish speeds! 10/10 don’t recommend though…

So, there you have it, I post experiments like this occasionally on the site’s brand new Insta https://www.instagram.com/windtunnel_ai/

This was like 2 days worth of effort (fine, I didn’t need to make cool one-off animations and renders, but, I couldn’t help myself).

I hope you enjoy the results.

[RChung]

Very cool, very interesting!

[Eric_Semianczuk]

Hi RChung, quite appreciated, there’s so much to test…

[ABG]

3w saved going to 60mm

Gains are gains

[rac1ngsnak3]

Need to validate the model and then you’re golden

[Huegelreiter]

170 Watts for 36 km/h seems mighty slippery, even considering the aerobar position.

But interesting site. Probably tons of difficulties had to be overcome to turn this into a service. And probably tons of caveats to be aware of when using it. Do you plan to benchmark this suite somehow with example setups tested via this method against true wind tunnel testing or other means like chung virtual elevation method?

[Eric_Semianczuk]

Agreed, got to balance it against stability and weight though.

[Eric_Semianczuk]

We’re working on getting into a Wind Tunnel as fast as we can for validation, but, to be fair, it’s more of a rubber stamp anyway. I didn’t write the CFD software, OpenFoam, the software that I’m using, is an industry standard that has been validated in every capacity against the real thing, from high to low speed.

So long as the model is scaled properly, which I accomplish by scaling it to a 622mm/700c rim (I have a circle sitting there in 3D of this exact diameter, I take the model and scale it, focusing on matching the rim to this circle precisly) it should be as validated as the state of the art CFD software.

I am going to get a comparison made, but this is along the logic of a bike company using CFD to test a wheel’s aerodynamic properties, but then testing to see how accurate the CFD software was by then re-testing in an expensive Wind Tunnel.

This type of iteration is done, get close enough with CFD, then further refine and using “real world numbers” with a Wind Tunnel, and in the case of my service, this can get the rider’s position and gear greatly enhanced so less time needs to be spent in the Wind Tunnel.

[Eric_Semianczuk]

Well… I love being aero, and the model is my twin, on my bike, with aerobars clamped under the bar on a -30 degree stem.

He and I are 145lbs, 5"7’

on a regular roadbike, not in aerobars, just in the hoods, with the most aero position I could hold for around 2hours, on a bumpy mildly hilly 2m asphalt circuit, I can average 20mph at 151 watts, so, take out rolling resistance, smooth things on tire like tread, and use aero bars… we get pretty aero.

I really got to get my taller/bigger friends to pose for a few, so it seems more realistic to the non/climber+TT guy crowd.

… oh here it is.

Here’s an example from one of our first customers, Chris Mehlman (with permission to share this), he found us from a post and we performed several tests, he’s 6’2" and gunning to win Garmin Unbound XL (rides for Pivot Cycles), the first two positions were very close watt wise, but his “very hard to hold”, most aero position he could produce ended up being staggeringly more Aero:

Just some fun info, Also, I’m currently upgrading the test to follow Trek’s CFD protocal, automating 11 concurrent tests at 11 yaw angles, graphing it, averaging it, etc., so it’s only getting more accurate, too bad I have to redo that pdf again…

[rac1ngsnak3]

Will you be able to model different tread patterns? Or is it so insignificant that it’s not even worth it to go down that rabbit hole?

[Eric_Semianczuk]

first, yes, second, correct.

I already found that removing wires from the front of the bike (going integrated) saves around .29 watts at nearly 23mph…

However, I LOVE experimenting (I made this because, I I wanted to know so I will likely put together some tests with a bunch of different treads I model and see the difference, then make a post about it and probably add it to the insta and eventually somewhere on the site.

[Eric_Semianczuk]

Cool thing, I just finished make it so I can run 11 yaw angles for one wind speed at the same time and graph CdA over yaw and Watts over yaw, average and median them, I got the idea from a Trek whitepaper:

Real pain to setup, but it’s totally automated, so that’s cool.

[jsiegrist]

Really loving following along to this. Fascinating stuff!

[huges84]

@Eric_Semianczuk

We’re working on getting into a Wind Tunnel as fast as we can for validation, but, to be fair, it’s more of a rubber stamp anyway. I didn’t write the CFD software, OpenFoam, the software that I’m using, is an industry standard that has been validated in every capacity against the real thing, from high to low speed.

As someone that’s done a fair amount of 1D simulation, I can say the results aren’t as iron clad as you seem to think they are. There’s many ways to get inaccurate results from perfectly good simulation software. One giant caveat is always that the tool is simulating exactly what you tell it, which is almost always a simpler scenario than the real world. Did you simulate moist air or dry air?

Take the roughness (friction factor) of various surfaces. I assume all of your surfaces are treated as having the same surface roughness. But in reality a painted frame is relatively smooth, gravel tires typically are not, clothing and skin are different still, etc. So that means your results may not match a wind tunnel.

There’s also the fact that the wind tunnel is Assisi not perfect at representing the real world riding conditions. Air flow isn’t uniform. In fact, for a given “wind speed” there’s a lot of variation. Some of that is due to surges and lulls and changes in direction. But an even bigger factor is the friction of the air to the ground. So the air one nanometer above the ground is actually always at zero velocity. Then it exponentially approaches the actual wind speed as you increase in height from the ground. Actual weather reports are usually measured like 20 ft off the ground for this reason. Then of course there’s turbulence in the air due to it interacting with the terrain, vegetation, buildings and other riders. The wind tunnel can account for some of this with effort, but most of it usually isn’t.

Lastly, there’s always various configuration settings in the model and solver that affect the speed and accuracy of the calculations. If you’re not trained on OpenFOAM or don’t understand the underlying physics, you might use settings that give you improper results, with the error amount potentially being influenced by the geometry. That would mean that even “A is better than B” concussions could be false.

Now, I’m not knocking what you’re doing. I think it’s cool and useful. But please understand and be clear with customers that the data isn’t validated, but should be pointing you in the right direction. Even better would be to share your OpenFOAM settings and parameters that you use in the analysis to get some feedback and potentially make your results better.

[lazydog]

Totally agreed. I work on products where we use the best commercial grade CFD code for analysis but are still constantly surprised by real world test results. You can completely change your CFD results based on the settings used, and since flow separation is so dependent on surface friction and texture your model could be very incorrect based on how you model those.

You also don’t need a wind tunnel to validate your model when you can use Chung testing! It is very easy to do.

[Eric_Semianczuk]

All of these concerns are valid and many are the same concerns aimed at a Wind Tunnel.

I gave the post a whimsical nature instead of a serious tone so it may be interpreted that I am lacking some experience and understanding of these points, I am more than aware of each one and many more.

The truth of the matter is, if you really want to be particular, yes, the specific, outdoor nature of wind, from turbulent zones between sky scrapers in a city, to a lax, windless day in a prairie, can create such a dynamic range, it’s hard to say if any method could truly account for everything, Wind Tunnel, fancy equipment or otherwise. How can we even test outside if the same exact, down to the moisture level, conditions are practically never the same?

See, this calls into question the entirety of anyone’s aerodynamic testing efforts in all capacities.

What I understand from experience, is I am faster at less watts on average with deeper wheels, even at slower speeds. Can I model this? Yes. Intuitively, having less cables on the front of the bike probably makes you slightly more “aero”, have I tested and seen the .29 watt difference? Yes.

On to your point about air density, humidity, etc. we’re looking at a comparative value from test to test in the elevation (denotes air density) with different setups, could this comparison be ever so slightly different with different properties? Perhaps, is it suddenly going to make an Aero helmet the wrong choice, likely not.

Regarding qualifications and OpenFoam, you got me, I don’t even have a college degree, no formal training on the software whatsoever, so obviously the results would reflect that? Well…

It actually was quite simple in comparison to my other pursuits to understand it, and fully automate it’s usage, even running many yaw angles simultaneously and combining the results.

With similar lack of experience, for the simple pursuit of finding gravel roads to bike on, I built a world routing engine from scratch in C++, a website, now used by thousands, that is a straight competitor to RideWithGPS: https://sherpa-map.com

I even taught myself enough in the geospatial realm to make an entire custom map with the highest resolution geography for cyclists, supporting the world, trained an ensemble of AI to classify road surface type that had not already been classified and integrate the results into the road coloring and routing engine:

In having a hard time deciding which bicycle amalgamation would be appropriate for a mixed surface bike race I built easily the most powerful physics engine for entire race courses, it even incorporates wind/moisture/exposure/etc. dynamic CdA, etc. GPX Route Speed Estimator for Cyclists: Multi-Surface, Weather, and Nutrition Strategy

I thought It would be cool and pioneered a custom pipeline to take insta360 video from a single ride through and turn it into an entire 3D mountain bike course Potree Viewer (best accessed on desktop, WASD to move), with plans to fill the whole that was FatMap, but I couldn’t host it.

These are just some of my public projects. I also managed to get a custom llm work offline on a android phone with a custom transcription service that could, in theory, replace a garmin, letting you record your activity with your phone and as questions like “what was my best 20second power so far in this ride?”, or “how long at this pace until the next aid station”? It totally works, just couldn’t beat the wind noise…

So, regarding qualifications, I need to know this software so well, and understand the science to such a degree that not only can I operate it, automate it, but be able to go on podcasts, speak to journalists, and experts in the field without looking foolish.

All of my results have been corroborated by experts I am in touch with, my OpenFoam test cases have never been disputed, and I mimicked Trek’s own CFD testing protocol which they claim is within 3% of a wind tunnel.

Will I hire an army of Aerodynamicists and CFD experts to take on some of this work if this starts getting traction? Heck yeah, do I have to take on the responsibility of teaching myself everything from the most cutting edge AI, creating novel algorithm implementations to generate high quality mesh for this test, to CFD software, totally. In the case of OpenFoam, it’s a relief to just be able to burn through a bunch of textbooks and not spend days upon days researching papers and coding until 2 am to “perhaps” make a breakthrough…

A little ranty sure, but I was just having fun with it for these tests, I’m extremely serious and dedicated, and am well aware of the points suggested, I do appreciate that you brought them up as it does give me a chance to address them.

[Eric_Semianczuk]

I wrote a giant reply to the other guy… so I’ll keep this one simple, I’m aware, everything has it’s flaws, I’m trying to determine which position is the best comprise between speed and handling on a bike, I’m not designing inlets for scramjets or ramjets that require precise shock positioning to achieve proper compression.

[RChung]

I think what you’re doing is both a good serious contribution and also very very cool.

[splash]

@Huegelreiter

170 Watts for 36 km/h seems mighty slippery, even considering the aerobar position.

I was thinking that, but I think this is only power needed to overcome aerodynamic drag. You’d have to add 50 W or so for rolling resistance and internal mechanical losses. Which would bring it up to 220 W, which feels ballpark correct to me.

In any case, I think the absolute values themselves are not the most interesting, but the trends are. Of course, aerodynamics can have funny quirks were trends don’t transfer, or suddenly invert or whatever, but modelling is a great way to work out in what direction to go for real-world testing.

[huges84]

Great to hear that you have a good grasp on the complexities and are trying to account for them. You just made it sound like it was a slam dunk that the simulation results and testing would match.

I wasn’t trying to imply that your simulation needs to account for all the real world factors, though I can see people interpreting it that way. Being close to wind tunnel results is good enough IMO. I was as much talking to others reading this as I was talking to you to make people aware that even the tunnel isn’t perfect.

While there infinite scenarios that can be modeled, there’s only so many that we need to draw good conclusions. Yaw angle is an important factor and including it and weighting the results the same as Trek are great steps. Doing a yaw sweep at one to three relevant speeds should tell us everything we need to know.