This was a loaner bike that was not fitted to Ray. We just quickly adjusted the seat (nothing else), and off he went! That in mind (likely an uncomfortable ride) there were a few occasions coming out of aero and standing upright. There was also one instance where I believe the gap between the lead rider and Ray was not far enough (I think they came within 3 bike lengths momentarily). There was also one stop sign intersection which was not ideal, and some feathering of brakes may have occurred.

---

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

Hi Chris,

Thanks for posting here and for your work on this product. It seems very exciting!

I am fairly comfortable with GC & Aerolab, but I'm not sure the benefit of using it in conjunction with your Aerolab device. Maybe I'm missing out on the distinction, but I thought from your Aerolab I would get real-time CdA (and other) data which would mean that I would no longer have the limitations of GC's Aerolab to calculate aerodynamics after-the-fact. Plus, with your device I could test many more "positions" by making small changes on-the-fly and see how CdA changes rather than holding one position for several laps, then a second position for several, etc. (When I say positions, I'm referring to even small things like a shrug or hand position.)

The avatar idea is really good for big things like body position and gear, but ideally there should be a way to incorporate small changes, too, like I mentioned above like hand position. Maybe a way to photograph yourself and store the images along with the avatar.

I see myself, after making many big changes, doing some testing as follows:

I ride a circuit with a phone to record my voice and saying, "at 15:45 [minutes into the ride] I am moving my hands both in front of my BTA, right over left. [reading head unit] That looks to improve CdA to 0.241.... At 18:15 I am moving my hands over my computer, right over left. [reading head unit] That lowers CdA...." I would do repetitive tests of all sorts of combinations of hands and head and shoulders, etc. Then afterwards I would go back and examine the data file for the each position (using my verbal notes on which position was at what times). Then after I settle on a position, I could take a photo if I really wanted just to make sure I got all the nuances correct.

I hope that's clear. This is how I imagine I'd use your unit, but I'm a real data junkie (biologist). I'm not sure if it would be used this way by everyone, but I think this is how people would get the most bang for their buck.

Thanks again!
Mike

Actually...what Adam "taught" me is that it's better to do the calculations with a "work-based" approach (rather than "force-based", as described in Robert's talk outline) and you avoid problems with the acceleration term being a "half-record" behind ;-)

---

http://bikeblather.blogspot.com/

Hi Mike,
This would be the dream - to see these tiny changes in CdA with slight alterations to hand position etc. in real time. Is that realistically achievable?
The challenge, as with any sensor based measurement, is what is your instantaneous uncertainty in CdA that includes both random errors (noise in sensors) as well as bias errors (e.g., bias in power measurement, road slope measurement, etc.). It is my belief that some changes will be discernible immediately (e.g., a CdA change from 0.255 to 0.275 for example), while others require a longer sample of data or repeated trials to determine the change of CdA (e.g., a change from 0.2550 to 0.2555).
Keep in mind, experimental wind tunnel studies have shown that CdA can change with wind speed (i.e., Reynolds number effects), wind yaw (some wheels actually produce thrust at high yaw angles via a type of sail-effect - this will alter the CdA of the bicycle-rider system), crank position (yes, depending on if you have the pedals at 0 and 180 vs 90 and 270 will change CdA though most of this has been attributed to a change in 'A'), etc.

So, if you are riding along for a couple of minutes, monitoring your CdA, and then change your hand position slightly, you might see some change in CdA, but it will be impossible to determine if it was caused by you moving your hand, or caused by any of the above factors (I did not include all factors here).
I should also note, a change from 0.2550 to 0.2555 is 0.0005. For a cyclist at 40kph (11.11 m/s), and density of 1.2 Kg/m^3, this changing from 1925.4 grams of drag to 1929.2 grams of drag. The difference is 3.8 grams. Guess what weighs in at about 3 grams? A penny.

From a practical perspective, detecting a change in drag on that order of magnitude (3.8 grams) is out of the realm of possibility for most wind tunnel facilities (especially once you factor in the effects of repeatability). It will be unlikely to see any sensor system detect changes on that order of magnitude. We (at AeroLab) are scheduling full scale wind tunnel testing in the coming months to better ascertain the limits of the sensor system. It is a critical piece of information for any consumer.

---

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

GoldenCheetah is a free software project (free as in freedom) -- you are encouraged to contribute ideas, documentation or code.
There is a user group here: https://groups.google.com/...golden-cheetah-users

Altruism can be a two way street.

Mark

What are you views on getting accurate values in typical turbulent air along typical roads?

I've made a similar device* and have struggled to get good precision on days with wind. I think the underlying issue is that the pitot tube is only measuring air pressure at one point in the vertical air column, and moving air and wind gradient in the last couple of metres next to the ground makes it difficult to infer what the total air force is.

*http://forum.slowtwitch.com/...F_P6481386/#p6481386

I certainly followed that post and ST thread. Your device is damn small and sleek.
This is a great discussion.
So, in making your device, you struggled with good 'precision' on days with wind. Basically a lot more scatter in your average values of CdA, but the scatter still surrounded your expected value (i.e., accurate but not precise from run to run)?
To me this is not particularly surprising, as the rate of convergence of mean quantities will likely scale with the turbulence level and the integral length scale of the turbulence (i.e., a longer statistical sample is required to obtain a converged mean quantity in the presence of stronger fluctuating components and low frequency fluctuations).

Did you do an analysis of the wind measurements (mean, RMS, spectral energy content, and integral length scale) to see if each run had the same statistics? I believe this would be needed to determine if your scatter is attributed to experiencing different turbulence levels or attributed to statistical convergence issues.

Now, when it comes to atmospheric turbulence, the integral length scales are significantly larger than the size of a rider. I haven't actually ever estimated the kolmogorov length scales (dissipation scales) expected for atmospheric turbulence, but I would propose that the typical length scales of turbulence encountered by a rider are significantly larger than the rider. And so, the fluctuation in wind is a global fluctuation that is felt across the entire vertical column. Certainly the no-slip condition ought to hold true for the ground plane, and so we anticipate a velocity gradient to exist in the presence of wind. I have been stopping quite a bit along various routes to measure the velocity gradients in heavy wind conditions (>10kph). My experience is that the severity of the gradient was always dependent on the local terrain, so it was difficult to have a single perfect model fit. Nevertheless, we have a basic model that we have created from experiments on the road (we apply a modest gradient to the velocity field in a polynomial-type fit). I am still playing with this to see if CdA results turn out better when applying such a model - basically you need to compute the velocity of the wind in a fixed reference frame (vector subtraction of the rider velocity), and then apply the model, and then add the rider velocity back. There is still a challenge of deciding what velocity you use to compute the CdA since the frontal area is a function of the y-normal distance from the ground plane. It has been a bit of a rabbit hole for me.

---

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

Is Ray really really bad at holding a TT position?


Ray's brilliance is in his passion and deep knowledge of the training technology sector. I doubt there is anyone else around who has the knowledge and info about this area. Ray's always been my go-to when I have questions about tech and gear in this area.

That he's not so good in the aero position - not so much of a problem for me.

---

---

Steve Fleck @stevefleck | Blog

It sounds like you've put a lot more legwork into the issue than me!
No I'm not doing any post processing other than just for determining average wind direction. I basically just do a variation of the virtual elevation method and spew out the results direct to the garmin. I got as far as pondering that I might need an algorithm for quantifying the pressure variations and and either alerting the user that the data is probably a bit rubbish, or just dropping significant digits off the cda value. It was more a case of remembering what an old paraglider instructor said about always visualise the air flowing over the land like water round an object in a stream. And while cycling along, visualising how the wind was flowing over the adjacent hedge next to me, and seeing in mind all the vortices swirling across the road, I decided that I probably needed to live somewhere with less wind and hedges to get nice values :)
But yes spectral analysis certainly might be the answer to a least giving the user an index as to what level of confidence they can have in the numbers.

Sorry to be late to the party.

Yeah, GC/Aerolab can be a PITA if you're doing non-standard things. Unfortunately, I don't speak C++ so I can't really contribute to GC, other than to make suggestions.

In general, there are a few PITA things that happen quite frequently when you want to analyze rides.

1. Excise a turnaround or braking incident or a spoiled lap and then paste the file back together.

2. Split a ride and mirror reverse the VE for half of it, like for an out-and-back.

3. Overlay repeated laps or portions of laps, with the laps synchronized on distance and starting (or ending) at the same elevation.

I've written little R functions to do these three things. I had, at one point, hoped that I could learn enough C++ to do that in GC and use the mouse to do this but reality eventually set in.