Bottle Positions CFD Tested: Bottled Speed

[ijaero]

I run a lot of CFD simulations for my work with Team USA so I developed tools that allows me to quickly test how different things affect the airflow around athletes. Last week trying2hard asked me if I had tested bottle positions, so I tested a few different set ups both in front and behind the rider.

To keep simulation time and work load manageable, I only simulated athlete, helmet and bottle, I did not include a bike, or bottle cage. I could see specific bottle holders affecting drag on their own, so keep that in mind. The simulation parameters all match best practices outlines in academic literature, and I wind tunnel validated the simulation approach myself last year and got good alignment.
.
.
.
Baseline position: CdA 0.222m²

I recreated a position based on my own 3D model, but in a position inspired by Matt Steinmetz excellent articles on the Taupo and Hawaii positions.
The CdA without a bike is 0.192m², which comes out around 0.222m² with a pro level tri and tt bike. This matches the numbers I would expect for a relatively upright position (I mostly work with pursuiters and WorldTour TT riders, which don’t have to hold their position as long, and have significant more resources allocated towards their aerodynamics)
.
.
.
Front Bottles

Every BTA bottle was faster than no bottles:

Regarding the orientation of the bottle (forward vs back facing), I would take the results with a grain of salt, as it probably depends on your exact bottle shape.
.
.
.
Rear Bottles

I tested a variety of rear bottle positions and the results were clear: Get the bottles as close to your body as possible and horizontal. If you ride the nose of a long saddle, your bottle carrier is most likely too far away.
The best result came actually when placing the bottle directly against the back
and low, like in a (low) jersey pocket, with the potential to save almost two minutes over the ironman distance.

The bottles in the back work as a spoiler to promote flow seperation on the lower back. If you look at the streamlines over the back of an athlete you can see that someair gets bent almost vertically down. This is creating a lot of lift and induced drag:

Placing a bottle on the lower back can disrupt this flow pattern and encourage the air to leave in the same direction it came from.

[beachedbeluga]

This is really interesting. Thanks for doing work! I can’t wait for Ironman to ban having a bottle in your back pocket, and people 3-D printing bottle cages to put there.

[Lurker4]

@ijaero

Placing a bottle on the lower back can disrupt this flow pattern and encourage the air to leave in the same direction it came from.

ButtBladder™ patent pending!

[kariscout]

ButtChug™
Knibb as first sponsored athlete

[DoronG]

Finally, Faris is vindicated!!

[ijaero]

The best aero decision he ever made.

[adgatri]

BootyPop™

Just for lols, what about the 3 or 4-bottle TriRig setup?

On the rear, what about the Wove-style 2-bottle stacked setup (i.e. one on top of the other vs. side by side) or the Ditlev 3-bottle setup?

[desert_dude]

I just finished some wind tunnel testing at A2 last week w/a pro triathlete. We tested a lot of bottle configs.

The surprising one was that removing the highest bottle in front was actually faster than having a bottle in. It allowed a better head position.

Great work putting this out.

Athletes, before you go making changes you need to understand that what works in CFD may or may not work for them IRL.

I’ve tested some other parts in that session that CFD claimed was going to be faster only to see it 4w worse IRL.

[marcag]

@desert_dude

I

Athletes, before you go making changes you need to understand that what works in CFD may or may not work for them IRL.

I’ve tested some other parts in that session that CFD claimed was going to be faster only to see it 4w worse IRL.

There is an excellent episode on Escape Collective Geek warning podcast (Dec 10th 2024) where they sit down with the engineers that designed the Colnago Y1Rs. They give a VERY candid opinion of CFD vs tunnel testing and the applicability to the real world. I went into it thinking “here we go again”, but was pleasantly surprised with their honesty.

[desert_dude]

Yikes, I let my Escape subscrip lapse during the holidays.

[ijaero]

Hey Brian,

Thanks for weighing in and sharing your experience from wind tunnel testing.
I think you make some excellent points.

The end goal is to help athletes go faster in the real world, come race day. The ultimate measurement of success are race results, but we can’t put on a statistically significant amount of Ironman World Championships every time we want to test a new helmet.

So the tools we have available are outdoor field testing, velodrome testing, wind tunnel testing (with life riders, or mannequins) and CFD. Each tool has its own strengths and weaknesses, and you are absolutely correct that we need to be aware of the limitations:

The validity of wind tunnels is affected by the blockage ratio, the flow profile across the test section, the main stream turbulence intensity and calibration and thermal set up of the balance. The main challenge in wind tunnel testing though is the repeatability of the rider. I have seen that run the whole spectrum. A world class time trialist or pursuiter can often repeat within 0.0005m^2 (one SD), many WorldTour time trialists are 0.001-0.002m^2, For amateur athletes, first time in the tunnel I have sometimes seen numbers worse than 0.005m^2. That being said: Wind tunnel testing, when done right, with enough repeats and at a quality facility is a highly valid approach to improve positions.

The validity of CFD is affected by the 3D model of the athlete, the mesh resolution, the turbulence model and convergence criteria. The biggest upside of CFD testing is its excellent reproducibility: Running the same position twice will give the exactly same result, dropping the head one cm drops the head exactly one cm and changes nothing else. The biggest downside in my opinion is that you can test things that you might not be able to do in reality: Bottle floating in air behind the rider, without any cage , or change your head positions without any accounting for range of motion, forward visibility, or what downstream affects this might have to the position.

Ultimately, I believe wind tunnel or velodrome testing to be the most valid approach for experienced riders that can hold their position reproducibly. The big advantage of CFD is the ability to test positions 90% cheaper and with perfect reproducibility. So for athletes that don’t have the budget for wind tunnel testing, CFD can offer a real alternative. For a WorldTour or an Olympic Team, you can use CFD to scan dozens of positions to then bring the top candidates to a final validation to the tunnel or track.

I’d love to compare note some time on your experience with athlete testing. When I was at A2 last summer to validate the CFD approach Geoff mentioned you and spoke highly of your work.

[ijaero]

Thanks for sharing. I’ll check out the podcast and will let you know my thoughts afterwards.

[marcag]

@ijaero

Thanks for sharing. I’ll check out the podcast and will let you know my thoughts afterwards.

I have never run CFD myself but many times, colleagues would feed me CFD data that would end up getting tunnel or road tested. It’s a great tool that I always say I will learn more but man, there is just to much to learn.

My experience has been it is a GREAT tool to hypothesize what to test in the real world but it does not replace real world testing.

Would I make a decision based on what CFD tells me : no. Would I build my test matrix based on it : absolutely.

Very specific to your analysis above : did you/can you run them at yaw ?

[desert_dude]

@ijaero

I’d love to compare note some time on your experience with athlete testing. When I was at A2 last summer to validate the CFD approach Geoff mentioned you and spoke highly of your work.

Anytime. I love talking about this stuff especially with people who are out of my area of expertise but still within the field. I’m sure there is a lot you could teach me that could help me expand my knowledge base & be an even better aero tester. Hit me up anytime.

I’ve been answering questions on my IG stories (@accelerate3) that people have DM’d me after this last testing session. Ironically today’s question was about all the bottles upfront.

I really like working with Geoff. Such a great guy.

[ijaero]

Hey @marcag
Thanks for sharing the podcast with me. I like when a company is willing to push the boundaries and I like the things coming out of the “new” Colnago, one thing I would note is that the interview was with a product manager and an industrial designer. They seemed quite technical though, respects to Colnago.

I am glad they started first validating their CFD models, both against the tunnel balance and pressure tabs. Validating tools should always be a first step and the first thing I did before we started using it with athletes.

My viewpoint is that CFD now is where wind tunnel testing was 10-15 years ago: Lots of people are doing it, but the quality varies greatly - and I am not even counting people running “Toy-Grade” CFD like the things that come bundled with CAD software.

For what its worth, when we developed the Venge, SL7, Roubaix, and Shiv TT (when I was working at Specialized), the deltas to the bike comparisons matched within tunnel accuracy. Absolute numbers from memory where around 5% apart. If you look up peer reviewed validation data on Ahmed bodies, DrivAer etc. you find closer agreements than 15% as well, so higher accuracies than that are certainly possible. That being said, its also not hard to be less accurate.

[ijaero]

I did not run them at yaw. The sports I work with currently are indoors so the tool is set up for zero degrees, but this is certainly something I could add.

I can share my own validation data: I tested 30 positions at A2, and in the software. The black dots are the tunnel results, the blue ones are CFD. The error bars are representative of what I would expect 1SD disagreement to be if both datapoints came from a tunnel. Since I work with a range of sports, those positions were me standing in the test section, not on a bike, adopting many different positions, in a range beyond cycling positions, so keep that in mind. I have a few other validation tests I did, but that date isn’t entirely mine to share.

The simulation approach is based on validation studies done by Professor Blocken in TU Eindhoven (here and here). If you want a deep dive on CFD accuracy around cyclists, these two papers are the best I have found. Key findings the Shear Stress Transport two parameter k-omega turbulence model, without any wall models, and enough surface and wake mesh resolution, matched wind tunnel tests to within 1%
.

You can potentially get even better results by adding a transition model, but that is ongoing research and something I haven’t included in my “production” models yet.

[Engner66]

Thanks for sharing all this, this is great stuff. I am a big fan of CFD, been using it in our industry for over 12 years now. I agree with you that the quality of the model varies greatly among software and users, but when used properly is an extremely powerful tool.

[ryinc]

Thanks for sharing this. Are there interactions between the front and rear or do you expect the results “stack” additively?

[mathematics]

@desert_dude

Athletes, before you go making changes you need to understand that what works in CFD may or may not work for them IRL.

I’ve tested some other parts in that session that CFD claimed was going to be faster only to see it 4w worse IRL.

Do you think there’s an amount of gain that you can see in CFD that pretty much always translates to a real world gain? Surely wheels, tires, things far away from the rider are solid, but things like bottle placement seem rider dependent. I’m just thinking if the CFD shows a 2 watt saving it’s probably a toss up in testing, but if it shows a 20 watt saving it’s likely to pan out? Just margin of error, basically.

[Hanginon]

Very nice! Your “baseline” athlete seems positioned for long-course. Have you run any analysis for a rider positioned for a short-course position more like this -