BikeTechReview.com

  • Increase font size
  • Default font size
  • Decrease font size
Home Forum
Welcome, Guest
Username Password: Remember me

Suggestions for Field Tests
(1 viewing) (1) Guest

TOPIC: Suggestions for Field Tests

Suggestions for Field Tests 11 years, 1 month ago #10829

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
I think this is a great idea for a sticky thread! Slap your field test suggestions here, just make sure to talk about "correcting/adjusting" your data to a standard condition (if that floats your boat) to avoid confusion/trademark infringement.
-kraig

Re: Suggestions for Field Tests 11 years, 1 month ago #10832

  • triguy42
  • OFFLINE
  • Platinum Boarder
  • Posts: 357
  • Karma: 0
Definitely a great idea to have a generalized set of instructions, here's a few suggestions I got from my aero tests thread:

1) Windless days...probably goes without saying because of wind's ability to really mess with the averages. Since that doesn't exist in Floriduh, I'm kinda stuck with maybe more variability than I'd like in my own tests.

2) Running ~10 tests for each set, i.e. 5 each alternating if testing two options (LG Chrono vs Giro Eclipse helmets for instance). Makes sense, and would mean about 28 miles of very boring slow riding back and forth on my course.

3) Running shorter tests and more of them. This doesn't make a lot of sense to me as the average for each test will become less variable the longer the test runs. Looking at the extreme, a 200yd test would be so short it couldn't possibly be valid. I've been using a 2.8 mile loop near my house as it's flat, convenient and relatively free of car traffic.

4) Riding tests at higher speeds, maybe 24-25mph instead of lower speeds. This would make the aero portion more important and maybe make data more consistent as road conditions, tire drag, powertrain losses and "Egromo watts" vs SRM watts is less of an issue. Downside is that it's easy to hold a very consistent 20mph but harder to ride 25mph+/-0.2mph.

5) To me, correcting data for a difference from targeted average speed is a requirement. Comparing 175W at 20.1mph to 175W at 19.8mph doesn't say much...but once corrected to 20.0mph average it's a difference of 179.7W vs 172.7W. Without doing the correction it's like comparing apples to VW bugs.

6) Flat course, must be out and back (no circular loops). This was a suggestion from a friend not on this board...I don't know if specifically doing a loop vs out and back is an issue. I'd definitely think that any elevation changes need to be minimal, and ending at a different spot from the start would probably be a bad idea given how much small grades affect output wattage.

Comments...suggestions? Anyone have a good suggestion for a minimum total test distance? We need a good statistician to figure it out.

Re: Suggestions for Field Tests 11 years, 1 month ago #10834

  • jkheycke
  • OFFLINE
  • Platinum Boarder
  • Posts: 956
  • Karma: 0
I do trials of 0.5 miles each way, out and back. I think this is sufficient. I can usually take even smaller subsections (say .2 or .3 miles) and they usually match the overall average of a .5 mile run quite well.


-jens

Re: Suggestions for Field Tests 11 years, 1 month ago #10840

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
I haven't done this, but I think something that would really lend confidence in the results of any kind of testing protocol would be to attach a known shape (with a known drag coefficient) to your bike in as clean of air as possible.

For example, if claimed CV's are in the 1-2% range, a reasonable start would be to attach a 5cm x 5cm square piece of 1/4" plywood to the top of your helmet (normal to the flow)- this should produce an approximate 0.003 increase in CxA.

Maybe attaching it to the front dropout would be another location... Dunno - anybody else have any other ideas on this?

One should be able to reliably pick up that increment on any given day - say, do it for 6 occassions and measure the increment each time (preferrably without knowing whether or not the plywood is installed or not - "blind" so to speak). That would be a cool (and well defined) objective, goal.

If it turns out one can't pick up that increment, try doubling the size of the square piece of plywood and repeat. This kind of approach should give a clear, caveman-based , idea of how much we can believe the results.

If you have access to a router, bevel the edges of your square, so the sharp corners are on the leading edge, so to speak (this should eliminate any reynolds number/speed dependence).

Anyway, just a quick thought that ran through my mind on my ride today...
-kraig

Re: Suggestions for Field Tests 11 years, 1 month ago #10845

  • andy497
  • OFFLINE
  • Gold Boarder
  • Posts: 289
  • Karma: 0
kraig wrote:

For example, if claimed CV's are in the 1-2% range, a reasonable start would be to attach a 5cm x 5cm square piece of 1/4" plywood to the top of your helmet (normal to the flow)- this should produce an approximate 0.003 increase in CxA.


Very cool, but wouldn't a sphere be easier? With a lollipop on your head, you wouldn't have to worry about staying perfectly perpendicular and all that. It seems like somebody talked about this before, and the problem was getting undisturbed air. How far do you have to get in front or to the side of an object in a flow to get clean air?

Re: Suggestions for Field Tests 11 years, 1 month ago #10846

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
andy497 wrote:

Very cool, but wouldn't a sphere be easier?


Maybe, I'm just a bit concerned about reynolds number stuff. if the protocol is at different speeds, this could pose a problem.

I like the way you are thinking, though. Maybe a roughened cylinder would do the trick.

on top of the helmet (maybe 6-12 inches up?) seems clean to me - at the hands, hmmm.... not so good:


http://www.biketechreview.com/aerodynam ... nation.htm

maybe offset low (and out), towards the ground, from the front fork dropouts is another option?

not much interference down there, I reckon.
-kraig

Re: Suggestions for Field Tests 11 years, 1 month ago #10856

  • SteveB
  • OFFLINE
  • Platinum Boarder
  • Posts: 779
  • Karma: 0
kraig wrote:
I think something that would really lend confidence in the results of any kind of testing protocol would be to attach a known shape (with a known drag coefficient) to your bike in as clean of air as possible.


That makes all the sense in the world. It's basically calibrating your measurement setup.

Re: Suggestions for Field Tests 11 years, 1 month ago #10858

  • SteveB
  • OFFLINE
  • Platinum Boarder
  • Posts: 779
  • Karma: 0
triguy42 wrote:

3) Running shorter tests and more of them. This doesn't make a lot of sense to me as the average for each test will become less variable the longer the test runs.

5) To me, correcting data for a difference from targeted average speed is a requirement. Comparing 175W at 20.1mph to 175W at 19.8mph doesn't say much...but once corrected to 20.0mph average it's a difference of 179.7W vs 172.7W. Without doing the correction it's like comparing apples to VW bugs.

6) Flat course, must be out and back (no circular loops). This was a suggestion from a friend not on this board...I don't know if specifically doing a loop vs out and back is an issue. I'd definitely think that any elevation changes need to be minimal, and ending at a different spot from the start would probably be a bad idea given how much small grades affect output wattage.

Comments...suggestions? Anyone have a good suggestion for a minimum total test distance? We need a good statistician to figure it out.


#3 - I can tell you from my professional experience that in general, when you have too much "stuff" grouped together, without tightly controlling for confounding factors, you can often end up with "mush", that is, no distinct result. In this case, a longer course might allow for wind, traffic, other issues to creep in to your data and skewing the result or what I like to call "smushing the data" - you've introduced so many other factors that all the runs come up looking similar or you end up with seemingly conflicting results.

Sure that can happen on a 1 km course or a 2.8 mile course, but if you carefully select the 1 km course to mitigate confounding factors like wind, traffic, dog walkers, ice cream vendors, blah, blah, blah, you'll be better off.

One of the advantages of a larger number of runs is that it gives more opportunity to throw away data that is obviously erroneous. (That never happens in field tests though, correct? ) Even if you don't exclude the data, it lets you at least mark the data point as "different" and you may wish to apply some judgment as to how to use it. If an erroneous "section" is included within one of your 2.8 mile runs, you are less likely to know about it.

If you have been able to control the confounding factors, the more runs you make, the more confident you can be in the result. That's basic stats.
You note, "the average for each test will become less variable the longer the test runs" and you are right. But if you've introduced errors, all bets are off. If you've controlled the confounding factors, you should see an even more distinct pattern with more runs.

#5 - that's why some people like to analyze the data by linear regression. In theory, even if you are a little off on speed, the data should still fit the line.

#6 - Agreed. Starting and ending an out-and-back in different spots might not be a great idea from the stand point of replication. I think it might be helpful to do the same amount of work in both directions to be able to smoke out bad data.

Re: Suggestions for Field Tests 11 years, 1 month ago #10859

  • triguy42
  • OFFLINE
  • Platinum Boarder
  • Posts: 357
  • Karma: 0
An additional thought on #6...braking to turn around??? If done consistently at each time for the turnaround then this would be a non-issue, but if you slow down to 10mph one time and 14mph the other time to turn around then there would be a discrepancy in the data. I resolved this in my data by taking out the braking and acceleration segment from the data.

One good reason for doing the loop is that if there is a headwind (even 1mph can skew the data) then you account for it by going constant speed with the 1mph tailwind. On a loop though, 20mph is ~139W with no wind, 118.5W with 2mph tailwind, 162.3W with 2mph headwind. The calculated average power is still 140.4W if you ride at a constant 20mph in both directions, pretty darn close. Without doing an A-A loop even tiny amounts of wind would throw things off.

Regarding #3 I found this to be an issue in my tests, with cars being a major variable. Floriduh's ever-changing wind conditions makes it difficult to do things consistently...I threw out one set of data because the run included a section where I was hit with 30 seconds of super-strong wind that took over 315W to ride 20mph!!!

Anyway, I have a segment of my 2.8 mile course that is about 0.8 miles to the loop. I'll try it out tonight for a couple of laps and see how consistent I can get it in that part. It should be an interesting comparison to the 2.8 mile course.

Re: Suggestions for Field Tests 11 years, 1 month ago #10862

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
One other method that I'd really like to see someone try is the method that Robert Chung has described - loop course with variable conditions and whatnot - just ride around... The loop just needs to have a significant enough hill in it to use the gravitational term in the equation of motion to it's advantage.

Anybody else know the method I'm trying to describe? Links? Anybody tried this?
-kraig

Suggestions 11 years, 1 month ago #10867

  • dwp
  • OFFLINE
  • Senior Boarder
  • Posts: 55
  • Karma: 0
A couple more suggestions - most of which I learned on this forum. I've been performing all my field tests on a 0.5 mile flat side road with very few cars. I'll abort mid-run if a car shows up in either direction since I've learned the data for that run is garbage. I take runs in each direction. Download from powertap, and get my data from 'clicking and dragging' each run. I typically get 0.4 miles of steady state data in each run.

1. Always start and finish each run at the same speed. It doesn't seem to matter if speed fluctuates a little during the run. As long as I capture a run with equal beginning and ending speed, then the change in kinetic energy is zero and the only major cause for variation is wind. I really struggled with crappy data before I started paying attention to exact beginning and ending speed.

2. I use the Lim method of taking runs at 4-5 different speeds while I warm-up. This allows me to pull out a crr value for that day. Even just two speeds in each direction (say 17 mph and 25 mph) will let you ballpark crr if I'm short on time.

3. Lately, I've been getting my best data from averaging out and back runs at TT speed and power. I have the crr from the Lim method during warmup, and all that remains is CdA and some variability each run. 3 sets of out and back runs per position has been giving me useful data. I'll usually test 2-3 positions in each session. The wind seems to cancel out, especially if it's not too gusty. I know cross winds can raise and lower the absolute CdA, but relative changes in CdA from each position are typically identified.

Re: Suggestions for Field Tests 11 years, 1 month ago #10877

  • Ron Ruff
  • OFFLINE
  • Platinum Boarder
  • Posts: 894
  • Karma: 0
Great topic Kraig!

Well... in my opinion using power meter field test runs to determine Crr isn't a good idea... there simply isn't enough precision to nail that, unless conditions are ideal and the calibration is excellent. There are better ways to get a Crr value... like roller tests for instance. Doing low speed tests to get Crr values also wastes time that could be better used determining CdA at more realistic race speeds.

As for only testing when there is little or no wind... do you also only race when there is little or no wind? Because drag will vary with the angle of the wind, it makes sense to test in conditions that are similar to what you would expect to see in a race... ie, a wide variety of wind conditons. It is nice if you can take wind measurements on your course so you can possibly correlate your results with the wind conditions.

It is best to do comparitive testing (rather than absolute), and swtich back and forth between configurations several times, to help reduce the effects of random variations and drift.

That's all I can think of for now... more later. I'm in the middle of a many thousand mile move and am pretty distracted...

Re: Suggestions for Field Tests 11 years, 1 month ago #10910

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
Ron Ruff wrote:
Great topic Kraig!.


I wish I could take credit for it, but it was a suggestion from within a different thread!

Thanks for your thoughts - the intriguing thing, to me, about the Chung method is that you just ride your bike and let things vary as they normally would (kind of addressing the things you allude to) - granted, you have to stay in the same position, but nonetheless, it is an interesting approach that no one has attempted to my knowledge.
-kraig

Re: Suggestions for Field Tests 11 years, 1 month ago #10931

kraig wrote:
the intriguing thing, to me, about the Chung method is that you just ride your bike and let things vary as they normally would (kind of addressing the things you allude to) - granted, you have to stay in the same position, but nonetheless, it is an interesting approach that no one has attempted to my knowledge.
Would individual power pacing strategy start to come into play in this scenario (you know, go a bit harder on incline/headwind, less so on down/tailwind) - hard to control maybe?

Re: Suggestions 11 years, 1 month ago #10933

dwp wrote:
1. Always start and finish each run at the same speed. It doesn't seem to matter if speed fluctuates a little during the run. As long as I capture a run with equal beginning and ending speed, then the change in kinetic energy is zero and the only major cause for variation is wind. I really struggled with crappy data before I started paying attention to exact beginning and ending speed.
In the spreadsheet (hosted on trainwithpower.net) here it enables you to have different starting and ending speeds - it compensates for the differences in kinetic energy. But I agree about trying to minimise the difference. At least this way you aren't stuffing around trying to cherry pick data (which is why I added it after a while).

Re: Suggestions for Field Tests 11 years, 1 month ago #10934

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
Alex Simmons wrote:
Would individual power pacing strategy start to come into play in this scenario (you know, go a bit harder on incline/headwind, less so on down/tailwind) - hard to control maybe?


Maybe, but I don't think so. The potential beauty of the approach is that you don't have to control anything - just as long as things are not systematically biased:

http://mywebpage.netscape.com/rechung/w ... meter.html

Maybe because the fact that this approach is not so "rigorously controlled" has something to do with why no one has really given it an honest go?
-kraig

Re: Suggestions for Field Tests 11 years, 1 month ago #10938

  • Tom_Anhalt
  • OFFLINE
  • Platinum Boarder
  • Posts: 1007
  • Karma: 0
kraig wrote:
Alex Simmons wrote:
Would individual power pacing strategy start to come into play in this scenario (you know, go a bit harder on incline/headwind, less so on down/tailwind) - hard to control maybe?


Maybe, but I don't think so. The potential beauty of the approach is that you don't have to control anything - just as long as things are not systematically biased:

http://mywebpage.netscape.com/rechung/w ... meter.html

Maybe because the fact that this approach is not so "rigorously controlled" has something to do with why no one has really given it an honest go?


I guess you could say I kinda tried this approach by applying the equation of motion to my Polar data on a point by point basis. I gave up since it appeared that slight delays in some of the different readings (i.e. speed, or altitude) munged up the data....

That is, of course, assuming I understand what you mean by the "Chung Approach".

Re: Suggestions for Field Tests 11 years, 1 month ago #10940

  • Ron Ruff
  • OFFLINE
  • Platinum Boarder
  • Posts: 894
  • Karma: 0
kraig wrote:
The potential beauty of the approach is that you don't have to control anything - just as long as things are not systematically biased


If I understand correctly, it seems like Chung's task was more difficult than ours needs to be... I mean he didn't know *any* of the variables except power and speed... and maybe weight. Also, for most of the people on this forum the critical question to be answered by field tests is "does this configuration have less aero drag than that configuration?"... leading hopefully to "is this configuration *faster* than that configuration"... after some actual race simulation testing. It shouldn't be too tough to get a decent delta-CdA value no matter what the conditions are. If you can't do a series of tests on a windy day that show one configuration being faster than another, then I don't think you'll notice a difference in a race either.

I like the idea of fitting large data sets to complex equations. Back around 1989 I wrote a program to do a least squares fit of any equation (and set of variables) on any data set. Even then an IBM AT didn't have to labor too long on the stuff I was working on... similar programs would run almost instantaneously on todays computers. So now we could put in the equation of motion for a cyclist (with CdA, wind speed, and wind direction as the variables), with the data set being the standard PM output of power, time, and speed... plus the bike direction and slope. I thought about letting the slope be a variable... but the chances of success should be much greater if it isn't. Probably need to make the wind direction consistent, too... at least for each out/back pair. If your test area (and weather) is nice enough so that the wind speed can be considered consistent for the the whole course, that would make things a lot easier... and you could then make the slope a variable if you wanted, rather than part of the data.

I'm just thinking out loud here... it might be more difficult than I assume (or even easier?)... but I'd be willing to bet that it can be done and yield meaningful results. Unfortunately, it's been a long time since I've done any programming... and then it was Fortran and Basic...

Re: Suggestions 11 years, 1 month ago #10943

  • SteveB
  • OFFLINE
  • Platinum Boarder
  • Posts: 779
  • Karma: 0
Alex Simmons wrote:
In the spreadsheet (hosted on trainwithpower.net) here it enables you to have different starting and ending speeds - it compensates for the differences in kinetic energy. But I agree about trying to minimise the difference. At least this way you aren't stuffing around trying to cherry pick data (which is why I added it after a while).


Though I haven't used it yet or looked at it for more than about 20 seconds, it seems to be a nice spreadsheet. However, what I would add would be:

1) more sheets for data from more configurations under test
2) more traces on the existing plots for those configurations

I think a lot of people are looking at relative differences and the ability to see the data from two or more test configurations simultaneously in graphical form is quite valuable.

Re: Suggestions for Field Tests 11 years, 1 month ago #10955

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
Tom_Anhalt wrote:

That is, of course, assuming I understand what you mean by the "Chung Approach".


I seem to remember trading emails or posts on the topica list with him, and what sticks in my mind as the "big picture" of the approach is that the loop needs to have a significant elevation change. Force the instantaneous data to hit the same peak elevation on each lap by tweaking CxA.

This is essentially what he did with Demet's data at the world cup race.

FWIW, here's some old moment by moment corrected CxA data from our "flat" fiesta island TT:

-kraig

Re: Suggestions 11 years, 1 month ago #10962

SteveB wrote:
Though I haven't used it yet or looked at it for more than about 20 seconds, it seems to be a nice spreadsheet. However, what I would add would be:

1) more sheets for data from more configurations under test
2) more traces on the existing plots for those configurations

I think a lot of people are looking at relative differences and the ability to see the data from two or more test configurations simultaneously in graphical form is quite valuable.
Well, happy to make upgrades if people suggest things.

As for 1. if you right click on the sheet's name tab and select "Move or Copy..." and check the "Create a copy" box, you'll have a 2nd sheet instantly.

As for 2. well yes, that wouldn't be hard. How many data sets would people like to compare on one chart?

I'd also like to add in an indicator of precision to the results (like a +/- confidence factor) so you can tell if two results are significantly different or not. I might need some help to choose the most appropriate method for that though.

Re: Suggestions 11 years, 1 month ago #11011

  • SteveB
  • OFFLINE
  • Platinum Boarder
  • Posts: 779
  • Karma: 0
Alex Simmons wrote:
SteveB wrote:
Though I haven't used it yet or looked at it for more than about 20 seconds, it seems to be a nice spreadsheet. However, what I would add would be:

1) more sheets for data from more configurations under test
2) more traces on the existing plots for those configurations

I think a lot of people are looking at relative differences and the ability to see the data from two or more test configurations simultaneously in graphical form is quite valuable.
Well, happy to make upgrades if people suggest things.

As for 1. if you right click on the sheet's name tab and select "Move or Copy..." and check the "Create a copy" box, you'll have a 2nd sheet instantly.

As for 2. well yes, that wouldn't be hard. How many data sets would people like to compare on one chart?

I'd also like to add in an indicator of precision to the results (like a +/- confidence factor) so you can tell if two results are significantly different or not. I might need some help to choose the most appropriate method for that though.


I guess 3 sets might be enough.

On my spreadsheets, I've been putting confidence intervals around the CdA and the Crr estimates. It puts things into perspective to see these. You can get the formulas for confidence intervals of slope and intercept out of most university-level statistics textbooks for math, science and engineering students. While Excel has the capability to do confidence intervals for point estimates (IIRC), I don't believe that intervals for slope and intercept are readily available in the Data Analysis pack or plain-vanilla Excel.

The confidence intervals for the intercept, and hence the Crr, will be very wide. That's the nature of the beast (the intercept).

You could also add error bars on the plot to show the confidence intervals.

Re: Suggestions for Field Tests 10 years, 11 months ago #12303

  • Ron Ruff
  • OFFLINE
  • Platinum Boarder
  • Posts: 894
  • Karma: 0
kraig wrote:

FWIW, here's some old moment by moment corrected CxA data from our "flat" fiesta island TT:


I've been "gone" for awhile and just realized that this never went anywhere... and I'm confused. What factors does the "corrected" CxA take into account? How flat is it really? And it was calm? So why so much variation?

Basically, I guess I want to know what the variables are in the equation you are using. The concept is intriguing but I just don't understand...

Re: Suggestions for Field Tests 10 years, 11 months ago #12317

  • kraig
  • OFFLINE
  • Administrator
  • Posts: 3285
  • Karma: 4
Ron Ruff wrote:
kraig wrote:

FWIW, here's some old moment by moment corrected CxA data from our "flat" fiesta island TT:


I've been "gone" for awhile and just realized that this never went anywhere... and I'm confused. What factors does the "corrected" CxA take into account? How flat is it really? And it was calm? So why so much variation?

Basically, I guess I want to know what the variables are in the equation you are using. The concept is intriguing but I just don't understand...


I used the equation of motion of a cyclist. I reckon that I used (it's been awhile..) the known power and speed data to solve for CxA on a moment by moment basis.

A course description can be found here:

http://www.fiestaisland.com

though, I think that there is probably 10-15m of elevation change per lap.

Even in calm conditions on the flat, when you are going au bloc, power data is noisy - which I reckon adds to the noise of the extracted CxA.
-kraig

Re: Suggestions for Field Tests 10 years, 10 months ago #12352

  • Ron Ruff
  • OFFLINE
  • Platinum Boarder
  • Posts: 894
  • Karma: 0
kraig wrote:
Even in calm conditions on the flat, when you are going au bloc, power data is noisy - which I reckon adds to the noise of the extracted CxA.


At least we can assume that your actual CxA didn't change much. If there had been a steep hill this wouldn't have been true... but on hills the aero drag is not too important unless there is a stiff headwind. It seems like Chung mostly used this feature to fit the data in his analysis.

Concerning what Chung did, I'll quote his conclusion from the link you provided: "My conclusion? The next time someone asks me how good the HAC4 could be at measuring power, I'll say it's about as good as using a powermeter to estimate altitude gain."

I can't think of how to use this approach to get a good CxA number from field testing, though. If you can correct for slope on a fairly flat course (ie you know what it is and know where it occurs in the data stream) you should be able to determine CxA and wind effects. Your data seems pretty noisey... but I wonder what (if any) smoothing you used...?
Time to create page: 1.33 seconds

Poll

Which type of tire is more aerodynamic?