Air temperature or tire temperature? Where to get the temperature dependence of CRR of individual tires?

You are extremely stable on a bike. I don't get +/-.003 in the A2 wind tunnel. I now have something that measures some body positions changes so am trying to improve this. But I think +/-.001 is beyond what most people can do even if we put all "sensor errors" aside.

On your 7min test, how many times do you go through the same point, ie how many laps and/or what is your lap length.

When will your device be available for purchase?

I've mentioned this before but evidently not clearly enough: I'm not a huge fan of statistical inference (despite having grown up and trained at Statistical Inference Central) but one of the reasons why laps are handy is that if you do it right one "trial" looks like AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA.

On a recent podcast, Endurance Innovation, almost as good as yours :-), Chris said they would be working through aero testers/fitters

All joking apart, I aero test almost every day and your podcast, as well as Endurance innovation are on my top 5. Josh from Silca is also there.

Couldn't agree more on the fact that some are very stable, others (like me), it has taken quite a lot of work to get down to +/-.002 or less. It doesn't mean that the sensors are not good enough to measure changes on the order of +/-.001, it just means that I personally seem to be incapable of maintaining position such that I obtain a cluster of laps (we should probably at some point clarify the difference between a lap and a trial, if there is one) that are all within +/-.001. I have seen it many times from some pro triathletes (e.g., when we testing Brent McMahon we saw this level of repeatability at +/-.001). Everyone will be different when they first start out, and this is why its important to understand on an individual level where you stand.
Interesting that you were not able to get +/-.003 in the A2 tunnel. Was that on different days? or was that you simply getting in and out of the tunnel at separate times on the same day? I have looked at the investigations of both, and on different days this is not as surprising. If it was the same day just at different times of the day, this is a bit more on the higher end, but not out of the realm of possibility.

---

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

As a confirmation, yes, the first AeroPro sensors are going to Coaches, Fitters, Pro Teams, Engineers, and Research Scientists.

---

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

Depends on the protocol.

I think until relatively recently, people were treating these tests as experiments and trying to do statistical inference on them, so they assumed a familiar "trial" framework, including that the test errors were iid. But tests in the field clearly don't have iid errors and they got predictably poor results.

One way to handle the problem is to add sensors so we can avoid the omitted variable problem. I didn't have other sensors, so to add structure to the estimation, I added laps. Laps add complexity, but it's only complexity at the computational level, not so much out in the field at the test execution level. I was trading off more complex computations to get a simpler field test. In an ideal world, sensors would make that unnecessary -- or, at least, redundant. In an ideal world, you could JRA and the sensors would handle all the messy bits. In the meantime, laps+sensors could be a good belt+suspenders test protocol.

Everyone on ST is one of the above. When do I get my sensor ? :-)

Nope! I test many many riders, not just myself, with the same precision. I'm positive I could test you to that level of accuracy. Remember that +/- 0.001 is the average over 7 minutes, so you need a stable average not stability in each moment. Sensor errors also average out. 7 minutes is a long time. There are lots of other details of course: calibrated trustworthy power and speed sensors, head unit that records the data properly, correcting for air density, temp adjusting CRR, etc.

I have by far the best results with out and back courses where I clip out the turnaround. Total lap length is about 3mi -- 1.5mi each way. Each point is traveled once in each direction.

I do agree that you can achieve +/- 0.001 on many riders. That said, I have some certainty (from direct experience), that every rider has a 'warm-up' period where their results are rarely going to be +/- 0.001, and they also have a fatigue limit where their results are rarely going to be +/- 0.001 (fatigue limit can be either being too aggressive on a given lap because they are all excited about testing, or simply testing too long through the day). I agree completely on the Crr temperature correction to get closer to +/- 0.001 (something that is often overlooked). My tests with out/back are typically between 400m and 800m out and back, and so about 2-4 minutes max of riding per lap. The data storage rate and convergence rates will differ depending on if you are using a higher speed power meter or if you are logging on slower 1sec intervals. Correcting for air density is also fun, especially on sunny days when your measurement is directly impacted by radiation and swings in barometric pressure.

---

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

Marcag, You are very kind, I will have to check out that episode. Also looking forward to a chat with you on the podcast to summarize with is going on with these devices. I did contact Andy Froncioni and all he could say is he is employed by Garmin. I can only imagine the level of NDA's in place for these things.

Here is a link to the show I did with the Notio folks 146-the-notio-aerometer

They outlined a test session that was basically four hours long in the show but I am hoping to get some results by using shorter runs and repeating them over different days. The aero test thingy on the phone app asks for a near 4 mile test, but there is a minimum test of 1 mile out and 1 mile back. From an effort standpoint, it is much easier to repeat 5 to 6 minutes than 10-11. The Notio app also has a lap feature that I have not played with a great deal. It would be great to integrate training and testing if one is fortunate enough to train in an area that is suitable.

For all I know the devise is working perfectly and the deviation I am seeing run to run is just my own inability to hold the same position. I am going to put mine back on the base bar, against their direction (they want it in the extensions). My results have been much more consistent with it there. It also gives one the option of doing cockpit adjustments without recalibrating.

I think the biggest thing needed is a much more clear set of directions for the "casual user". I think they are working on that. It seems pretty clear however that some coaches are getting some really precise data with it, primarily in a track setting.

It would be helpful for me to just get "macro" results (large changes like position/helmets etc.). Micro changes (which sock or skinsuit) might be a lot more problematic unless one really spends time with it.

How big are you seeing ?

Yup +1 on that too. First run is usually not as good. Fatigue takes a while and I usually don't test long enough for that, but I have seen that too. My goal is to teach riders to test themselves (and send me the data) so the testing becomes a constant part of their training, rather than a one-off long session testing everything at once.

I remember discussing this in my old office on campus several years ago. I think it's quite clever.

To be clear I'm not saying it doesn't work for you.

I am saying that for me, having measured the wind, there are many times it is not identical in the two directions and that throws it off (for me) without a wind sensor. It is often pretty close.

I use to "inject" wind into my model, with a bunch of different techniques, but would not get uber precise numbers. OK numbers, but not as good as with a wind sensor

But what I was unable to do doesn't mean someone else can't.

Or maybe our winds aren't as well behaved.

So, let me get this straight: you held your anemometer in exactly the same position but turned it around 180 degrees and got a different value? What does it mean for the wind to be different in the two directions? (Being a little tongue in cheek here but you get the point.) I have athletes test on roads all over the world and it regularly works. Sure you might find a road where it doesn't, but if done right the process is much more robust than you think.

Let me add one more thing. Here's a story I hear all the time: someone goes out and does some field aero testing and they get inconsistent results. They don't understand why so they *blame* the wind gusts. Fact is that's not usually the problem. It's usually that they have unreliable data. They are using a bad protocol that doesn't average everything out properly, and/or they are not using a reliable speed sensor, power meter, and/or head unit, and the data has noise and holes in it (that they may not even know about because their software is filling in the holes). Fix those problems and voila.

Imagine a perfectly straight head wind on the way out and a perfectly straight tail wind on the way back.d
It's +2km/h average on the way out and -2km/h average on the way back. That's "most of the time".
But there are 8 gusts on the way out where it hits 5km/h. And there are 4 lulls on the way back where it hits -1. Not very hard to imagine conditions. You will see that variability in your results.
The longer the course, the more opportunities for differences. The shorter the course, the less you use your odds that things will average out. The advantage of short loops is you can see the one outlier vs the 6 more similar patterns.

But you can absolutely model "theoretical" wind after the fact and get ok results. You can model a wind pattern to get you within .001 difference and say "yes, that wind pattern makes sense". But was that truly the wind pattern ?

If you have wind data you say "yep, that's it". If you don't, you say "there was probably a gust".

JUST an opinion, influenced by looking at hundreds of rides with wind data.

one of these days we should do a thread on wind patterns. My wife and kids hate looking at my charts, I need to find someone who appreciates some of these crazy experiments....not all on a bike

If that were a real example (12 gusts in 7 mins) then you might need more than 7 mins to get the averaging. That's not how wind usually works though. Basically you need to increase the time of the test until all the different shocks approximately average out. Honestly, if you're experiencing big variability from run to run I bet it's coming from something else.

No. The wind is measured and the Chung model fits almost perfectly. But it fits because the wind is measured.

I am not trying to explain errors in my data. There are no errors. It all lines up.

I am showing that the data model works really well IF you have all the data.

If there were errors, the Chung VE would not fit.
If I didn't have wind data, the Chung VE would not fit or at least would not fit as well.

Or, maybe you're not computing CdA in a way that properly averages out the wind when the wind is not observed. It's a complex statistical problem.

Here's a data overload post but ties several of the topics discussed in this thread

First of all, I don't like protocols. I like to ride, do my workouts and if I can get aero data, great.
I absolutely appreciate that protocols allow to get more accurate data, especially when no accurate sensor exists. The same start/end point gives a delta Elevation = 0, no need for altimeter. Out and back averages wind out, I get it......

But the more accurate sensor data we have, the less need for strict protocols. The more data, the easier to find and correct errors. There is value in sensors that can do this, IF they do this. Rant over.

This data is part of an exercise I am doing in preparation for another project. The idea is to explain what sensors can do.

Here is a ride I do regularly, out 7km, back. Rolling hills, and part East/West, part North/South. You get it all, wind, yaw, up, down..... It's also the course I race weekly and I love getting real race data at much higher speeds.

Chart 1

So here is a Chung VE plot with only the Garmin data. In green, the altitude, in white Aerolab's VE plot.
The altitude sucks. My 520 has a crappy barometer. The VE curve shows quite obvious head wind and tail wind.

BTW, I have a version of Aerolab that re-syncs VE to altitude at selected laps. 2 is an out, 4 a back, 6 and out, 8 a back




Chart 2
Overlay speed and airspeed (airspeed - speed = wind)
Notice speed varies considerably. No need to hold constant speed or power

Top green is speed, blue airspeed. We see the general trend of head wind out (blue above green), tail wind back (green above blue). You can see mirror of wind patterns on the out and backs but not always perfect





Now let's correct the altitude with a high precision barometer, accelerometers and 3 other sources of data. One of those sources is broken here, I'll explain why later. Correct altitude in yellow





Now let's calculate a VE using the wind and see how it matches the corrected altitude.
VE with wind in white. Altitude in Yellow. Laps 4,6,8 are much better




Let's correct the CDA for the first lap to show the sensitivity here. I just moved the CDA slider

ST is preventing more charts, so I'll continue in another post

Continuing here

As I said, I modified the VE slider b 0.002. Out #1 is now better, 2,3,4 are off
So the difference between lap 1 and the other 3 is .002. Not bad. As a matter of fact, most of that is what Chris mentionned, first lap is usually off. If I turn on CRR temp correction it drops to .001, but details




Now to date, everything has been done using Chung VE method from Aerolab but fed with wind data and accurate other data



If I calculate a "rolling CDA" using the "corrected Altitude", I get this rolling CDA chart in blue at the bottom. It has some very small lumps in it due to that 3rd sensor missing. IF and when this gets nice and flat, we are close to "real time CDA" although I guess processing 3 days later is hardly real time. Need to finalize that third sensor....




l