Nate552

Just wondering, how much precision are you expecting from a power meter to yield a result more precise than a wind tunnel? And which powermeter were you using? When was it last calibrated?

I'm still waiting to see a working device that allows for field testing past 0 yaw. I know it's theoretically possibly, but until I see it, it's results backed up with data, and I can buy one, it's vaporware. (and no, I'm not building one myself)

This is a real problem. At the tunnel, I try to measure everything on a rider, including head position and hand position to help identify and prevent subtle changes between runs. But as soon as the door shuts, a rider can move their head, or shoulders, etc. I have seen very experienced TT guys have pretty poor position repeatability. Between runs they will get back into their TT position and their head is 5 cm lower. They are often amazed at how hard it is to get back into the exact same position. I can only imagine how bad it could potentially be in the field.

Then sometimes you come across someone like Racer Ex, or this kid Sergio that is rock solid in their position, and can nail it every time. Every time I measured this kid he was spot on. I was pretty impressed.



It is extremely rare to find a windless day in Texas. Like you, l have had to wait a long time for a windless morning in past. I'm actually waiting right now to do some field testing on a new helmet.

echappist

Nate, what size crank is he running? I assume 165mm?

Nate552

165. No way could that kid do that position with 172.5. There's a thread on the triathlon site about him (Sergio). Super nice (and super fast kid). He was quite disappointed that he couldn't beat my drag numbers, but I laughed and told him I would gladly trade his power numbers for my drag numbers.

echappist

yeah, i remember the thread. That was his road bike, with a pretty normal looking crank. 165 definitely stands out.

also, what's the consensus on the tradeoff between lower initial profile (like Sergio Jr.) vs a flat back (like most time trialists)?

Nate552

There is none. Traditional thought was that by lowering the front too much you increased surface area by exposing the back. testing might look like this :
1) - lower bars = less drag
2) - lower more = even less drag
3) - lower even more = Eh, kind of the same drag so lets stop lowering and work on helmets.

A few years ago Cobb decided to keep going and eventually, you kind of "break though" a wall and the drag starts dropping again. Sergio is an example of this. Of course, that position is not for everyone, and as always, what works for one person.... Cobb published a mini-paper on this that he passes out at the F.I.S.T. camps.

RChung

That's my point: there is a substitute right now for both. The non-zero yaw piece isn't generally available but the high precision part is.

Well, I paid a tad less than $700 for my PowerTap. What do you mean by "go to the tunnel a bunch"?

That's why field testing is interesting: you think of that noise as pollution but I see it as giving a real world context to the precision you need on a practical basis.

Well, above when I mentioned a sd of .0006 m^2 or a CV of 0.3%? That was done with an old PT Pro, the one that records at 1.26 seconds. It had been checked with a static calibration sometime shortly before the data collection, I don't remember exactly when, but it had also been checked several times over the preceding months and was also checked after the test runs, and over all static checks, both before and after, had been spot-on. [Edit:] BTW, what's the sd on the drag you measure at A&M?

Nate552

What do you mean by generally? Is it available at all? I'm still waiting to see this.


I see where you are going with this, but noise can also be a detriment. If you are comparing, say, 2 helmets, and you have various noise in your tests, but different kinds of noise, how do you make a sound decision based on your tests?


So you're saying it was tested a lot, but you don't know specifically when, with regards to your tests. What was the temperature difference between your static tests and your field testing?
My question was in regards to how accurate the power meter was. If a powertap is only accurate to +/- 1.5% how do you account for that in your precision testing? Also, what if the testing was done with a SRM? Should you adjust your findings because the power is derived at the crank and not the hub? There could be a 2% variation due to that. How do you account for these potential variations?

As for the tunnel sd, I'll need to check with them, but I'm curious now.

RChung

Not generally available means not available to the general public unless you make your own or know someone special.

Yup, but there are two kinds of errors: random and systematic. If you've been following this type of analysis you can often tell one from the other and put bounds on the systematic component. That's the real value of VE, btw: there's actually a closed form solution but the VE profiles give you a way to diagnose poor estimates and separate the random from systematic components of error. As I may have said elsewhere, this stuff isn't my day job: my day job involves, among other things in my field, getting robust estimates of some parameter or another. Estimating a parameter is good but figuring out how to diagnose and evaluate the quality of the estimate is even better. That's why I showed the VE method in the way I did (although I discuss the closed form solution and the direct assessment of variability in the accompanying notes).

Oh, it was within a couple of weeks but I no longer remember exactly when. Static calibration checks bracketed the test so I'm not too worried that it magically went out of calibration and then magically went back into calibration between the checks.
Because accuracy and precision are different things. In any event, what I was interested in for that particular test was the difference in CdA between two test conditions that were performed right after one another (and the temperature was recorded for each test run, btw). So even if the absolute estimates of CdA were biased, the difference in CdA between the two runs appears not to have been (you can tell that by examining the differential over the course of the test, which is what I did).

Yes, you should. Same for a Quarq. That's part of the reason why I prefer to use a PowerTap for this sort of thing. And, the drivetrain losses from a SRM or Quarq can be larger than 2%. I've measured that, too, on bikes with both a PT and SRM or Quarq mounted simultaneously. [Edit:] However, if all you're interested in is the change in CdA between two test conditions, then you don't really have to estimate the SRM or Quarq drivetrain losses as long as you keep it in one chainring/cog combination. That just biases the estimate but doesn't affect the change or precision.

Racer Ex 

There's seldom high precision in field testing (see external conditions)..."real world context" by your definition makes noise a good thing, which it's not. It's dog poop on the rug. "Real world" is, to paraphrase Chung, a red herring unless you apply it to yaw. Which you can't field test.

Having seen the variability of results from 3 different tunnels, and having run 100+ field test runs with a couple of riders, it's my experience that it's pretty rare for the field to hit the repeatability of the tunnel (and this is coming from someone who doesn't pee or drink during tests to keep the rider weight as static as possible). Can it be as precise, or at least close? Sure. Every time on any given day? Absolutely not.

The kid is good. I'm old and stiff.

botto 

that's what you get for taking those little blue pills.

Racer Ex 

Those pills make you old?

It's worth it.

RChung

1. I'm absolutely not saying that all field tests are always better than any tunnel test every time on any day -- that's a straw man because you'll never find in any thing I've written that I've ever made that claim. I'm saying that properly done field tests (and "properly done" is key) can be high precision. If you've been getting dog poop field test results, then your field test techniques are dog poop. How do you explain a CV on the estimated CxA of 0.3% otherwise? That's a sd of .0006 m^2, i.e., for that test I could reliably distinguish a change in CxA of around .001 m^2 -- "reliably distinguish" in the conventional 95% hypothesis testing sense. That's about the size of my thumbnail. When you go to the wind tunnel, do they give you a 95% confidence bound on your results? What is the CV on your wind tunnel estimates? After you tell me that we can do a little more discussion on noise and dog poop.

2. I'm also saying that equipment exists that allows one to test at variable yaw, although that equipment is not generally available.

3. As I've said elsewhere, a wind tunnel works when the conditions outside are not conducive to field testing; a wind tunnel is faster than field testing; a wind tunnel can (often) be more convenient than field testing; if you don't have the right equipment, a wind tunnel lets you test at non-zero yaw; and a wind tunnel is way more expensive than field testing. I've also said that the key to a good testing protocol isn't just that you can get an answer out of it -- it's that you can tell how good (or poor) that answer is. If you've done 100+ field tests and got dog poop results out of them then I'm thinking maybe those methods you're using aren't the best use of your time.

4. BTW, I can produce the same level of field testing precision even without a power meter -- the testing is slower and more tedious but the precision is the same.

Racer Ex 

Which is why it's not a substitute, which was my original statement. You're at nature's whims and while you'd now like to describe noise as a good thing, we both know it's not.

Well, you're inferring above that you can get dog poop results from field tests on a bad day, which we both know to be true. So you're really arguing both sides. Just don't let that escalate to a fistfight. If it does make sure you post it on Youtube.

I'm guessing that somewhere in the deep dark recesses of LabChung's supercomputer that there's a deleted spreadsheet or three. The people that stand head and shoulders above the crowd are usually standing on a pile of failed experiments (paraphrasing Erv)

I'm pretty OK with my protocol BTW. If you'd like to peer review it I'd be happy to forward it.

Don't need one. My confidence comes from repeatability on a controlled set up during the tests. But I'm using the tunnel for comparative analysis, not for the aero equivalent of the Ewang FTP chart.

I've got zero concern with the absolute accuracy of the derived number which is probably where and why we're parting company on this discussion. The fish I'm frying is winning bike races.

BTW, this is what I was referring to when I wrote about "by time you build it, you could have gone to the tunnel a bunch", not the zero yaw tests. Given that you're probably the most noted advocate and resource for field testing and you don't have the equipment or even an controlled estimate of time/cost/effectiveness of non zero yaw equipment...


Depends on what your time is worth.

As for the CV numbers from the tunnel, I'll leave that to Nate.

Nate552

Are you talking about force measurements? The Oran Nicks Tunnel lists the specs of the tunnel on their website. I don't know about the San Diego or A2 specs.

https://lswt.tamu.edu/info.htm

I'm having a tough time with the "thumb nail" CdA. A CdA diffrence of .001 m^2 works out to be, what, a watt? So your powertap, is reliable down to the 1 watt? Impressive. Have you published this test? I'd like to see your testing protocol, as it must have been really good.

Nate552

This. When it comes down to it, we're trying to figure out what is fastest at various wind conditions. A wind tunnel is superior than field testing for this purpose. I'm not saying field testing is bad, or worthless just that it has it's limitations.

RChung

And I didn't dispute that. I've said many times in many places that field testing is not a substitute wind tunnel testing; I've said it's a complement. The part I was disagreeing with was the claim that field testing *always* has poorer precision than wind tunnel testing, and that field testing *can't* handle non-zero yaw. I've shown that field testing *can* have very good precision, and that it *is* possible to measure non-zero yaw.

Evidently we both don't know that. In my day job I look at a lot of "noisy" data. The history of this field is that sometimes the data that we think of as "noisy" can reveal systematic variation in addition to the random component. Being able to tell one from the other can often lead to better estimates. That's why my field tests turn out to have much better precision than traditional field tests -- traditional field tests couldn't differentiate between the two.

Sure. Maybe start a new thread.

Actually, I don't think this is where we disagree. If the tare on a wind tunnel balance is off then the absolute drag is off but most of us are knowledgeable enough that we don't care about that. Most of us care about comparing equipment or positions so even if the tare is off the relative difference is still good. That's what happens with well-performed field tests. The absolute CdA can be off but the relative comparison between equipment or positions depends on precision.

Um, what makes you think that? I haven't made any specific comment on what I do or don't have available to me.

Nope, not exactly. I've known for a while what the claimed accuracy and resolution of the force measurements are. I'm interested in the measured precision, not the accuracy.

Of course I've written this up. As I pointed out above, you don't even need to use a power meter -- it's just faster and easier (much easier) that way. If you don't have a power meter then you have to do the tests at zero power -- reliably at zero power. You think that's impressive? But by your questions I'm guessing you're not trained as a statistician. For example, we can talk about average global warming as measured in hundredths of a degree even though any individual thermometer may not have precision of 0.01 degrees. The Central Limit Theorem is a pretty neat thing. As for the protocols, they depend on the type of venue and situation you have available. Even if you have the right venue using the wrong protocol can give you poor results -- just as using the right protocol but in the wrong situation. I've never claimed that a good protocol magically makes everything work (I've claimed that a good protocol makes things better than a poor protocol).

Of course field testing has limitations. Who has said it doesn't? Y'all have been making blanket statements. I've been making nuanced ones: I've been saying that dismissing field testing because "it's not precise enough" or "you can't do non-zero yaw" is incorrect.

Nate552

I have a State TT championship coming up in 3 weeks, and I want to know if the new helmet I have is faster between 10 and 15 degrees than my old one since that's the predicted yaw angles for 2 of the legs. How would I accomplish this using field testing with tools currently available to me?

RChung

Currently available to you? Hmmm. Find a huge parking lot where the wind is consistent from one direction then ride at a speed so the net yaw angle is between 10 and 15 degrees?

I want to test in a wind tunnel tomorrow but my checking account balance is watched carefully by my wife. How would I accomplish this with the bank balance and schedule currently available to me?

Racer Ex 

I haven't seen systemic variation. Not saying it's not a possibility, but I haven't seen it. Examples?

Because you haven't made any specific comment on what you do or don't have available to you. An array of iBikes and a flatbed truck with one of those rotating beds from the 60's? I made that joke because the imagery makes me chuckle BTW.

I would say that saying we've been making blanket statements dismissing field testing shows you haven't been reading what we've both been writing. I think the best cycling analogy would be that field testing is Carlos Sastre and the tunnel is Alberto Contador. Both are capable of winning a grand tour. One just does so with greater frequency. That's all I've been saying...Sastre is not (really) a substitute for Contador. I think we've agreed so we can move on to:

Heck, I'm all eyes too.

Racer Ex 

Dissapointed.

If you wear pantyhose over your head when you go into the bank and speak in a loud voice, they will often give you more than you have in your account. During the cooler months you could substitute a ski mask.

RChung

Well, I was assuming he didn't have any other equipment available to him. If he had one of those Kestrel recording wind meters then there are other ways. In that case, it's sort of similar to a guy I know who was doing some experiments with a short loop behind his housing development that went up a little slope around an athletic field. Because he lived in a little bit of a valley the wind was pretty consistent in direction so all he was worried about was wind speed. He went around the loop several times at different speeds using the VE protocol then matched his position on the loop to the wind speed (and direction). Then he was able to get an estimate of his CxA at different points along the course and relate the different points to wind speed and direction.

But all of that requires a recording wind sensor.

[Edit:] I should add that although his estimates seemed reasonable, he couldn't afford to go to a wind tunnel so we didn't have any way to validate them. However, since Nate has been in the wind tunnel, if he was able to do something like this then he'd have an external validity check.

[Edit 2:] BTW, that same guy affixed a flat 5cm x 5cm plate to his bike and tried to estimate the difference in estimated drag. He estimated the difference in CdA at .003 m^2. The area was 5cm x 5cm = .0025 m^2, but the Cd of a flat plate isn't 1.0. Go look it up and figure out what the CdA of a 5cm flat plate ought to be.

RChung

The easiest one is slope, but there are others. I was recently looking at a file a guy sent me where he was disappointed in the results. I asked him if a vehicle had passed him going in his direction on the third lap right about *there* in his file. He affirmed it (it was pretty clear if you know what you're doing) and that's a variation that was systematic. Once I snipped it out the estimate improved quite a bit. Just to emphasize, he asked me to look at the file because he thought it was dog poop -- he didn't say there was a passing vehicle or where it was. If you want more, just ask. I've got several.

Yup. I haven't.

Racer Ex 

Lost me here...sytemic or systematic? In any case the car would seem to fall under the heading of random noise, unless it was a train that ran on the hour, or something of that sort, in which case I would consider it systematic. But I get where you're going...much the same as I'll abort or note runs when I TV test if there's something that I know is going to skew the results you clip out the noise.

I thought you were referencing something that occurred with regularity say, every third run or at 3:30 into the test.

I would say that using prevailing wind and a gauge to do yaw testing is gong to be very iffy to get a good control. Wind seldom travels in a straight path outdoors so you'd want a bike mounted velocity and directional gauge with a black box recorder, and even then you're subject to the whims of nature, the road builder, and terrain to "decide" on the "test" angle. You'd have to layer the data to determine the average effective yaw as wind and bike speed changes...which really gives you an average drag number at an average yaw number, and not an actual result.

RChung

Systematic, as opposed to random. But the thing that occurs with regularity is the other thing I mentioned, which is slope. VE mostly works by doing loops or laps over the same course a few times. The regularity you're looking for is that you know the hill you're riding on doesn't move around between laps so everything has to line up the same way each time you ride over it, no matter what speed or power you're expending at the moment. That's clear when the hill you're riding over is big but it turns out that even small dips and bumps in an otherwise flat test venue can add to the "noise." If you take it into account you improve the estimate's precision. I occasionally use a 1500 meter-long shoreline trail on SF Bay for testing. This is on the Bay so it's about as flat as roads get. I could spot the low point on the trail -- low point meaning it was about 40 cm lower than the rest of the mile-long trail. People were skeptical that VE could pick that up so the next time it rained I went out there with a buddy and showed him the puddle on the trail exactly where I'd predicted it would be.

Right, but the issue isn't control as much as it is measurement. That's why I mentioned the test loop went around an athletic field -- the wind was unobstructed over the field. I was surprised it worked as well as it did but, as I said, he couldn't afford to validate it with a wind tunnel trip so we couldn't be sure.
Now you understand why it's not enough to have a wind speed sensor -- you want to have a yaw sensor. That's what we've been talking about.

RChung

You mean, something like one of these?

https://www.bikeradar.com/gallery/art...news%2Farticle

https://www.bicycling.com/bikes-gear/...unnel-obsolete

https://www.grc.nasa.gov/WWW/K-12/airplane/tunp5h.html

https://nyvelocity.com/content/coachi...010/wang-chung