CdA changes between headwind and tailwind - to be continued

Post Reply
Hein
Posts: 3
Joined: Sun Oct 21, 2018 7:09 am

CdA changes between headwind and tailwind - to be continued

Post by Hein »

I do my own CdA calculations based on different data-sources. For wind-speed I use AeroPod data, for ground-speed I use my garmin speed-sensor, for the altitude I use a high-precision laser-scan with an accuracy of 2-3 cm, for power, I use my DFPM, etc... I combine all those different data-sources and correct for known errors. As a result I can model my ground-speed and the model is freakingly close to reality. In the attached picture you can see the speed of one testride as measured by the speed-sensor and the modeled speed for the same ride.
simulation_vs_speedsensor.jpeg
simulation_vs_speedsensor.jpeg (19.6 KiB) Viewed 5523 times
Since the measured speed and the modeled speed are virtually the same for every second of the ride, for high speed and for low speed I think the model captures the relevant variables.
In all my testrides I get higher CdA numbers for headwind than for tailwind. Since this effect is very robust I wonder if it might be true. I have some first thoughts for an explanation. If there is wind, there are always turbulences. I think those turbulences hit harder in headwind conditions than in tailwind conditions. As a result they slow you down more in headwind conditions. Has anybody some second thoughts about this explanation or some scientific papers?
Velocomp
Velocomp CEO
Posts: 7793
Joined: Fri Jan 11, 2008 8:43 am

Re: CdA changes between headwind and tailwind - to be continued

Post by Velocomp »

It's not at all clear from your posting how you are modeling ground speed, so it's difficult to comment on your post. Also, it's not clear how you are measuring CdA, so your headwind/tailwind comment is difficult to assess.

It would be helpful, too, if you posted a .ibr ride file from your AeroPod.
John Hamann
Hein
Posts: 3
Joined: Sun Oct 21, 2018 7:09 am

Re: CdA changes between headwind and tailwind - to be continued

Post by Hein »

I did not write about all the details in my simulation, because there are lots of details and I cannot explain all of them in a short post. The pic was to show my simulation hits the recorded speed data pretty well.
However, basically my simulation goes like this:
I assume a body with a constant system weight, a constant CdA value, a constant Crr value and a few other minor sources of resistance.
This body moves over a terrain with known elevation for every centimeter. I assume a constant thickness of the air. The elevation of the virtual track equals the elevation of the physical test track. The virtual body is pushed with the power as recorded from my DFPM and the wind on the virtual track is calculated from the AeroPod wind-speed (with a temporal resolution of 1 second). The initial speed of the virtual body is taken from the speed of my speed-sensor. After the model is initiated, it goes ballistic. With the known power from the DFPM and all sources of resistance it calculates the speed for every second. I wrote an optimizer function in C++ to calculate the whole model. I start with one set of parameters and compare the modeled speed, second by second with the recorded speed and store the deviation to the recorded speed. Then I rerun the model with slightly different parameters. I change all free parameters over a plausible range. The most important free parameters are CdA and Crr. The parameter combination with the closest match to the measured speed is the winner.
In the pic you can see the comparison of the recorded speed and the modeled speed for one test-ride with a distance of ca. 1.5 km. My garmin stores the speed for every second and my model calculates the speed for every second.
Since measurement and model are virtually the same, I don't think there are relevant bugs in my code.

Now comes the important finding: In all test-rides with headwind and tailwind the model finds an optimal fit with high CdA values in headwind conditions and low CdA values in tailwind conditions. If the wind is orthogonal to my riding direction, the optimal CdA fit is the same for the out and back parts of my rides.
I do all the rides on my TT bike. Since my calculation seems to be ok and the finding is robust, I wonder if CdA numbers can vary with wind conditions. Assuming a laminar airflow, my findings are highly implausible. It would have been detected and reported in wind-tunnel tests. However, in windy conditions, there is no laminar airflow. So my first guess is about some weird interaction with turbulences. Since I am a data-scientist and not an expert for low-speed high-turbulence aerodynamics I wonder if some expert can make sense of my findings. It would be great if somebody could point to some technical literature about non airfoil-shaped samples in dirty air.
RolfK
Posts: 26
Joined: Fri Nov 02, 2018 4:47 am

Re: CdA changes between headwind and tailwind - to be continued

Post by RolfK »

Hello Hein,

I'm not a physicist or an aerodynamics engineer, so I can't help with reference papers. However, it seems well-established that CdA changes with wind direction, hence all those tests of wheels at different yaw angles (here's one example: http://bikeblather.blogspot.com/2016/03 ... party.html ). Headwind / tailwind is probably "just" an extreme case, and I have found the same differences in CdA as you have, using the Aeropod - you can find some old posts about it on the messageboard.

Best regards
Rolf
User avatar
lorduintah
Posts: 642
Joined: Fri Jan 11, 2008 9:37 am
Location: Plymouth, MN

Re: CdA changes between headwind and tailwind - to be continued

Post by lorduintah »

This appears to be model building with various forms of data. What are the number of variables in this model - and - do you know that these variables are independent? Going through the "experiments" have you built a set of orthogonally designed experiments (DOE) that provide adequate variations to allow parameter estimations with great enough resolution so that the overall model is significant?

Without control of the variables, you will need a much larger set of data to be able to assess the solutions you are seeking. Otherwise the random noise in the results is likely to confound any analysis - and right now your comments appear to be substantiating that problem.
ted.kelly
Posts: 27
Joined: Fri Apr 19, 2019 12:40 pm

Re: CdA changes between headwind and tailwind - to be continued

Post by ted.kelly »

Could you elaborate on your "high precision laser scan" for elevation?
Are you working on a velodrome or open roads? How do you get your elevation data?
Post Reply