I'll admit I did the same thing as well, after I somehow thought the LSWT data wasn't corrected...I confirmed it was.

We're both so weird ;-)

---

http://bikeblather.blogspot.com/

Not a direct reply to anyone, but just a general question. Would someone like to take a shot at compiling a "best practices" list for this method? I've read almost the whole thread, but don't want to reinvent the wheel etc.

I finally have a power meter and looking to do some aero testing this year.

Protocols?
Lap length and distance?
Types of things that work well to test?
Software?
Equipment?

Thanks in advance,
Kyle

I agree that (CdA)_x is the most appropriate component to report.

For fun, consider that to balance a side gust, a cyclist must lean towards the wind until m*g*dy= D_y*dz= (CdA)_y*dz*(1/2 rho (U cos(Beta))^2) ... and the lateral tire force will equal D_y.

I only typed it out but didn’t sketch it so correct me if it’s wrong lol. I feel the angle can be significant in cross winds from experience, and i’d guess the magnitude is a something like cornering friction increases at the same angle (measured from center of mass to contact patch in dy and dz horizontally and vertically).

Best practices vary a little depending on what sensors you have, the type of test course you're using, and your racing speed. VE is pretty flexible so versions of it will work pretty widely, but sometimes you want to hone in on one or two things specifically so your protocol may differ.

That said, it's always good practice to use a dedicated speed sensor with the proper wheel rollout, to weigh yourself and all your equipment, and to track air density. If you're using a power meter, you want it to be accurate, not just consistent, so you'll want to have checked its calibration ahead of time. Then you'll want to make sure to let the PM match ambient temperature and zero the torque between runs. Wherever you choose to test, check for traffic and try not to use your brakes. If you're doing laps, make the laps short and vary your speed not only within laps but also across laps. For your baseline, find a position that you can hold well and that's easy to replicate. Practice, a lot.

Ugh, I am finally back to testing after a couple years off and had glitches in the virtual elevation and speed. Sigh, I realized I completely forgot to use add the speed sensor to the new bike I am testing.

Is all of the data suspect as a result or is something salvageable in the results? Just trying to make some rough decisions on angle of extensions and stack.


Screen shot of the results from Golden Cheetah (Looks like it uses the baseline - laps 2/3 and Lap 5 are same position).

Runs are all 3 laps on a loop:

Run 1 Lap 2/3
Run 2 Lap 5 (same position as run #1)
Run 3 Lap 8
Run 4 Lap 12
Run 5 Lap 18
Run 6 Lap 20

EDIT: I forgot to adjust ETA, so updating the screen shot. the CRR looks fairly close with the revised estimate. Running vittoria corsa speed (decent wear to it), latex tubes, smooth surface on the loop

Another EDIT: pretty convinced the data is junk, did another run with the speed sensor. still analyzing....

Below is what we did for some wheel testing in a small scale tunnel back in 2017. At the time it made sense to me, but I could probably be proven wrong or change my mind in the future!

---

Chris Morton, PhD
Associate Professor, Mechanical Engineering
co-Founder and inventor of AeroLab Tech
For updates see Instagram

---

Chris Morton, PhD
Associate Professor, Mechanical Engineering
co-Founder and inventor of AeroLab Tech
For updates see Instagram

My take:
- multiple strain/force sensors at every contact point between the rider and the bike enables you to determine the drag force (and lift or side force) in real units (say Newtons) of the rider themselves (not the bike).
- Thus, you have the drag produced by the rider, but not the drag on the bike.
- I can see a rider making changes to their position in close to real-time and directly seeing changes to the drag forces acting on them, which is neat.
- Unfortunately you do not have a normalization you can use for this drag force without measurement of the wind vector, i.e., you need to infer or directly measure the wind vector so that you can normalize by the dynamic pressure and produce a CdA value for the rider (not the bike).
- Once you have that (all these force sensors + wind sensor), you still do not know the drag produced by the frame, wheels, water bottles, or anything that is not carried by the rider. So you do not exactly know the total system drag (bike + rider).
- Interactions between the bike and rider are not captured directly here - perhaps a specific hand position reduces the rider drag but exposes the bike and increases the bike drag and hence total system drag remains unchanged or perhaps worse!
- You can infer the drag using an assumed Crr and re-arrange the force/energy balance, but then you are right back where we started, using virtual elevation, Chung method, other methods to infer drag. Which are great. I like them :)

- More sensors does not always mean better information, it often just means more information... which can complicate things further.

---

Chris Morton, PhD
Associate Professor, Mechanical Engineering
co-Founder and inventor of AeroLab Tech
For updates see Instagram

As you say Chris, more is not always better. In my experience with multivariate modelling it is the same thing, one can think that more model components looks better until the data is over fitted.... so more is not always better and as in some cases actually worse. I suspect in this sort of modelling it will be similar. Everyone is trying to find the holy grail of real time aero and it sounds like a tough nut to crack. I have been following your progress and it looks interesting, but I gather that you feel that it is necessary that the user be trained to be effective so "not mass market" yet.... Do you think that it will be "mass market, plug and play" any time soon? I was an early adopter of the second generation velocomp newton "power" meter. What I still like about it is some (be it as limited as it might be) idea of the head/tail wind effect. There are days when it just felt nice to be validated when I looked at the wind effect and it verified how I felt. I don't have that aspect any longer since I went dual sided 4iiii. Too bad there wasn't just a simple add on wind gauge one could use and the software to indicate the wind vector force over the ride if for nothing more than entertainment value and validation of those days when the wind really was the dominant force during the ride.

A good measure of how close/far we are from real time is the number of constraints people place on protocols and the quality of the data being produced. It would be nice if more people posted their data so we could see the evolution of these devices. Pierre (Bugno) is one of the best for sharing his data. Have you seen data from any other device on ST ?

IMO, it won't be mass market until they get the price significantly down. These devices need to be sub $200 which is hard given the economics to develop them. There is potential for big things happening here.

If the purpose of the sensor is for entertainment value and basic understanding if you are in a head vs tail wind, yes, such a device could manufactured relatively cheap (it would be a small scale wind sensor only, contain no other internal sensors aside from something for ANT+ or Bluetooth communication and perhaps the basic stuff needed for an air density estimation). Thus, the wind sensor is not doing any complex calculations, it is not taking in data from auxiliary sensors, it doesn't have any intuitive 'training' on what is happening in the surroundings. So now you have a situation where a hardware + wind sensor company is making these sensors for dirt cheap, but they rely entirely on head unit companies to adopt them and make them useful for a consumer. When about 5% (possibly less) of the consumer market wants these things, the motivation of the head unit companies to move on something like this becomes less attractive and the wind sensor company is left waiting... burning cash!

If things work out for the wind sensor company, what gets calculated algorithmically and displayed to the rider based on this data is then up to the head unit manufacturer (or the 3rd party that adopts the collection of the data, like Strava or some others). The reliance of the wind sensor on other sensors to back out drag forces and coefficients leads to the following conclusion in this situation: The quality of the data you get is now dictated by the head unit sensors, auxiliary sensors, and some fancy algorithms, the ability of the algorithms account for everything needed to make the device "plug-and-play".

There is a lot more to it than just what I have above since I did not even get into the details on what it will take to make the device "plug-and-play". Its sufficient to say that yes, I do think its possible to make it mass market and plug and play based on my experience in developing our sensor. Are enough consumers at the stage where they will see value and make use of such as sensor on their own? Do I think it is economically feasible for a wind sensor company to have a successful business model based on what I described above? The answers to these questions are left as a fun exercise for us all to ponder :)

---

Chris Morton, PhD
Associate Professor, Mechanical Engineering
co-Founder and inventor of AeroLab Tech
For updates see Instagram

thanks for the insightful answer. I know that my newton was a fun device with some certain intelligence built in. I still use it as a add on when riding my road bike because I still take the 4iiii as the power device but the air resistance part is interesting... so I did not toss it when I went DFPM. If they dropped their price further on the pod I would be tempted to use it as an add on, but I always felt it was a limited use device at best. The fun part was seeing how close their estimate was to the 4iiii data and truthfully for my use case it was in a general sense not too bad. When I went to my gravel bike though it lost its basic utility since there is no real way for it to deal with changing rolling resistance going from pavement, to smooth dirt, to fine gravel... etc over and over again and well wind is much less of an issue when you are enclosed with trees for the most part.

Hi Chris,

Thanks for the response and apoligies for taking a bit to reply.

My main objection is regarding the normalization, classically "dynamic pressure at infinity," Q_inf = 1/2 rho Vr^2.

Lets consider a cyclist riding at Vc=Vx, and relative to earth, there is a perpendicular wind velocity Vw=Vy, such that Vy/Vx = tan(beta), and Vr = Vx/cos(beta), Vy = Vx tan(beta).

Our wind, as defined, does not add a tailwind or headwind component to the rider, so Fd = -Fx in your Equation (3).

To properly apply Equation (6), we should use Vr = Vx/cos(beta), making it:

Fd = (CdA)_x 1/2 rho Vx^2 /cos(beta)^2

So far nothing has changed from your equations, we just ignore the complication of law of cosines. But many people, particularly those without sensors, incorrectly think of the data as representing

"If I'm riding at 50kph, and the wind appears to me as coming from an angle beta, then my drag is

Fd = (CdA)_x 1/2 rho Vx^2

which should emphasize that Equation (6) hides its inherent beta angle in Vr, but the data was acquired under one specific beta angle... so if we instead choose to be explicit in defining (CdA)_x = (CdA)_{x,beta_i}/cos(beta_i)^2, then the end user merely digests the formula as they imagine it, and further, (CdA)_{x,beta_i} is self documenting as being valid at a specific beta angle. I don't think anyone meaningfully digests the data as side wind + head/tail winds, so analysis and aero specialists can use whatever they like since they're less likely to develop a false intuition from the equations.

One thing I've personally witnessed, is people being surprised by the disconnect between coefficients and the forces they think they represent. Drag force will not be zero for a rider at 50kph and an extreme side gust, and I believe considering CdA_{x,beta_i}/cos(beta_i)^2 communicates that better than the alternative does.

Hi, I have noticed through some of the treads asking when the AeroLab sys would be available for US Fitters. Yes, we have formally launched in Europe but we are taking orders from US Fitters as I’m seeing a couple Fitter sensors going to the US.

If you have interest I encourage you to check out the YouTube channel which has a boat load of technical goodies. https://youtube.com/...DP9LRmT4szD3_6QaTjQg

Also, the website https://shop.aerolab.tech/ will give you some options for leasing.

Thanks for letting me flag a couple postings request.

Completely agree with you there Cody. This is part of the reason the wind vector was so important to me in developing the AeroLab sensor.
If you are not normalizing by the right dynamic pressure (your cosine beta squared correction), then you run into a challenge of not knowing how to interpret your data because you are simply missing key information.

---

Chris Morton, PhD
Associate Professor, Mechanical Engineering
co-Founder and inventor of AeroLab Tech
For updates see Instagram

Yup. People thought the errors were random, so if you had wide variance you could just enlarge the sample size. That works when the errors are random but not when they're systematic. Extra sensors (properly used) give you a way to separate the random and the systematic.

Wait...you mean if I have unmodeled and/or unmeasured variables that might change systematically, that I can't just take a bunch of samples and then evaluate the results using statistical tools designed to judge the effects of random errors...and then claim "no difference" when the null hypothesis can't be rejected?

Who knew? ;-)

---

http://bikeblather.blogspot.com/

That's the "if I don't know what causes it, I can assume it's random cuz that's SCIENCE!" way to analyze things.

A corollary is the "if I don't know what causes it, then you can't possibly know it either" argument.

One thing that had peeved me throughout my academic and professional life, is the inevitable and pervasive “we already tried X and the result was Y,” where upon investigation X was nothing like an actual X, and there could hardly be a “result,” much less a Y. The devil is in the details, and good science comes from scrutinizing these details, and making the conclusions as narrow as possible, which sometimes means there are no conclusions that can be made.

Actually, this a good spot for an Einstein quote:

Reminds me of a similar quote my dad, an experimental physiologist, told me decades ago:

“No one believes a theory — except the theorist. Everyone believes an experiment — except the experimenter.”


I don't remember if he attributed it to anyone then, but Google now suggests this might also have been Einstein.

---

Damon Rinard
Engineering Manager,
CSG Road Engineering Department
Cannondale & GT Bicycles
(ex-Cervelo, ex-Trek, ex-Velomax, ex-Kestrel)

I am diving into the world of Aero field testing and came across this great thread and spend the last day reading through it. While some of it is over my head it seems there is really helpful and talented group of people involved. I would really appreciate some feedback on my baseline run I have done to make sure I am on the right track.

I performed and ABAB test, where A was me riding in the Aerobars with completely relaxed shoulders and my head relaxed looking up the road, and B was me shrugging and holding the shoulders in and keeping the head as low as possible.

My set was as follows:
Cervelo P5 three
Training wheels with Aerojacket rear, and Bontrager R3 clinchers and butyl tubes
Giro Aerohead with visor
Bibs and Jersey (fitted but not aero)
No shoe covers.
Rear speed sensor & wahoo Gps
Favero Assioma dual sided powermeter pedals .

I found a 550m loop that has no traffic at that time of the morning and is pretty much completely flat with smooth pavement. There was no strong wind, and I did 5 laps A (laps 3 - 7), 5 laps B (laps 9 - 13), 5 laps A (Laps 15 - 19), and another 5 laps in the B configuration ( laps 21 - 25).

Using CRR 0.005 (a guess), total mass 90kg, RTA .98 and RHO 1.146 (from mywindsock).

From my analysis in GC I get the results I expected, that B is faster. From my analysis I get the following CDAs: A - 0.288, B - 0.255, A -0.274, B - 0.255. I suspect the first A run was worse due to an oncoming school bus on one of the straights during one lap, but can't remember which lap. All other runs were clean, no braking or coasting needed on any run.

I would appreciate any comments on my protocol? or if anyone wants to take a look at my file to see if I calculated correct that would be great. I just want to make sure I have the basics right before I start testing more subtle changes.

FWIW, all my laps in the first 20minutes or so will have a higher CDA. Is it me and a less relaxed position, the need to warm up or my tire or.....but it's always the slowest lap.

That's an okay protocol. Next time try to remember when a bus or car passes, but you can usually see an anomaly on one of the laps.

I don't always get higher CdA on the first lap, but I *almost* always do. I used to worry about why that was but eventually I gave up and usually I just do an extra lap (or, if the laps are very short, sometimes two) for the first run of the day and then I snip that out.

I guess if I just want to compare A vs B for the purposes of which one is faster, using a fixed, made up Crr works (unless I am comparing different wheels with different tires). But I understand that if I want to arrive at a very accurate correct CDA, I need an accurate CRR.

I am struggling to find the CRR for my training tires (Bontrager R3, with Butyl tubes), but a very similar model the R4 I see was tested at 0.0036 per bicycle rolling resistance.com. Am I correct that I would just double that number to arrive at the total crr? 0.0072 Seems kind of high when I see people for similar tests using crr numbers of 0.005 and it impacts the CDA by more than I had realized.

I am sure you have thought this conundrum over before and have a far better idea than me for sure, but is there any chance that it may be the tires achieving equilibrium state and arriving at a stable CRR? I know that Chris Morton mentioned that they had to control for tire pressure by using the ANT+ tire pressure monitors in their system.