Escolar Documentos
Profissional Documentos
Cultura Documentos
A REPORT
JUNE 2017
The triathlon bike manufacturing and design trend that began with the Specializeds Shiv Tri and Felts IA both market-
leading, UCI non-compliant, triathlon-specific bike platforms introduced in recent years has accelerated as new breeds
of radical frame shapes have emerged from Cervelo, Diamondback, Ventum, Dimond, TriRig, and others. In addition,
several companies have launched triathlon bikes with disc brakes, and smaller internet-based upstarts like Premier Bike
now sell value-packed options directly to consumers. Fully integrated systems with hydration and storage solutions
abound.
Historically, such questions have been answered for athletes and consumers by the manufacturers with the resources to
undertake comparison studies. But this project was funded and designed by athletes and for athletes. With support from
the community, we set out to independently evaluate some of todays finest and fastest time trial bikes. We tested the
Cervelo P5, Cervelo P5-X, Ventum One, Premier Tactical, Diamondback Andean, and Felt B Series in the A2 Wind Tunnel
in Mooresville, North Carolina on April 18, 2017, and this is the report detailing the results.
II. TEST SETUP & PROTOCOL
The test procedure was designed to be compliant with industry best practices and was informed by the teams collective
experience with controlled-environment testing and aerodynamic equipment optimization.
All bikes were tested at a wind speed of 30 miles per hour (mph), which is the de facto industry standard and a
speed that allows for detection of minor differences in drag that are scalable mathematically to a range of real-
life speeds during back-end data analysis.
Drag measurements were recorded at angles of -10, -7.5, -5, 0, 5, 7.5, and 10 degrees, which angles account for
the overwhelming majority of relative wind encountered by a typical athlete on a typical course and day. Sweeps
were performed with the angle of attack always increasing from zero or decreasing from zero. The specific
ordering 0, 5, 7.5, 10, 0, -5, -7.5, -10, 0 degrees was used to diminish the effect of hysteresis during sweeps.
We tested all bikes with an athlete on board and used the data for our high-level analysis and time-course
simulation modeling in this report.
For a clean comparison, all six bikes were also tested without a rider in the large front chain-ring and the same
cassette sprocket, with the pedals removed and with the drive-side crank-arm rotated forward and horizontal and
taped in position. We removed between-the-arms hydration accessories for the bike-only tests. Wheel speed was
also set at 30 mph for the runs with no rider.
To assess variability and repeatability, we pre-ran the control bike, a Felt B Series, through a series of baseline
tests at the beginning of the day, which were repeated at the end of the day.
At the A2 Wind Tunnel, software was used to overlay video and pictures for each test relative to the baseline run
to monitor and ensure position consistency throughout the day. We pointed cameras at and projected live video
in front of the rider, where lines were drawn in marker of both the bike and rider from the first test run. The
position was monitored throughout each rider-on run for body and head position compliance, and adjustments
were made or runs repeated when deviations were observed visually or detected in the live data.
The only tares that were removed were wind-off static forces so that aerodynamic loads could be isolated. No
other forces were removed.
III. BIKE SETUP & PROTOCOL
We held every aspect of each bike as constant as was reasonably possible. All key elements of the fit were aligned,
including pad width, stack, and reach; cockpit rise and length; and saddle height, angle, and setback. Care was taken to
normalize variables while working within the limitations of the equipment. In certain cases, proprietary interfaces
precluded total congruence, and the bikes were tested with stock components. The component choices were based on a
typical, high-end stock or after-market setup. Two wheelsets were used one pair of Enve SES 7.8 wheels with rotor-
based braking, for the Diamondback Andean and the Cervelo P5-X; and one pair of Enve SES 7.8 wheels with rim-based
braking, for the others and set up with identical cassettes, tires, tubes, valves, and tire pressures. Other aspects of the
bikes, such as drivetrains and saddles, were also held constant from bike to bike.
Comparison tests are inherently challenged and limited in that it is impossible to test and control every variable of
interaction between a rider and a frame across six bikes. Further permutations of frame sizes, athlete morphologies,
ancillary equipment choices, and fit and position configurations, among others, are too many to capture and optimize in
a single day of testing, if ever however, we worked to control or eliminate as many as possible for this one test. The
details of the bike setups as tested are as follows:
Modern triathlon bikes are not just frames but racing systems, and we set out to test each system the way it is ridden by
discerning athletes on race day. The specific hydration configuration choices for our testing were made with an eye
towards best-practice aerodynamics as well as function and utility, but even then opinions differed as to the ideal setup.
We know from previous studies that athletes can make
choices to limit or eliminate drag penalties associated
with these choices still, we wanted frames that
incorporated hydration and nutrition solutions intothe
design to benefit from core integration relative to
products with no such option. There is no industry
standard for what Cervelo calls Ironman setup and
Ventum calls race trim in aerodynamic testing
setups vary as widely among researchers as they do
among athletes but we included at least 24 ounces of
fluids (i.e. one large round bottle) to reflect the realities
of how the equipment is used. We utilized a between-
the-arms configurations one standard cage and
round bottle was mounted horizontally between the Cervelos Ironman setup test rig in the San Diego Low Speed Wind Tunnel
extensions and above the stem for all bikes except
Ventum, which offers a large integrated 1.4-liter fluid reservoir, and Premier Tactical, which recommends a Profile Design
between-the-arms hydration product consistent with results from previous wind-tunnel testing.
Separately, we completed two rider-on runs with different behind-the-saddle hydration setups to quantify impact of an
additional bottle, but neither setup made a measurable difference aerodynamically. We tested the best-practice setup
observed in previous aerodynamic study by Cervelo and Specialized the bottle was tucked as closely as possible behind
the saddle, affixed diagonally relative to the seat post and tightly secured by zip ties to the saddle rails which did not
differ in drag relative to control. The riskier setup, tested with the bottle and cage well behind the back of the saddle,
tracked closely to the bike-only margin of error in magnitude.
IV. RIDER-ON TESTING RESULTS
(22-26 mph) is roughly equivalent to 10 watts. So .01 m2 = ~10 watts and .001 m2 = ~1 watt.
To weight the output data from the wind tunnel research
Simple Weighting of Yaw Angles
appropriately, we need to determine the effective yaw angle, which is
the vector of the wind experienced by a cyclist. Yaw probability
30%
distribution functions differ widely among cycling industry analysts,
but the consensus in recent years from analytical studies is that in 20% 20%
2,411 2,400
CdA (m2 )
A human rider introduces data uncertainty, but simplistic comparison of our control run relative to our baseline run
indicates uncertainty, at any given angle of attack, is at most 73 grams of drag at 30mph, or +/- 36 grams similar to the
variability introduced by a pedaling mannequin, which Trek stated was 48 grams between duplicated tests of the first
generation Speed Concept. Absolute uncertainty in our test measured as the difference between control and baseline
runs at any given angle of attack was 41 grams of drag on average, or +/- 20 grams. For ease of comparison relative to
other research, and because figures here are not meant to be applied to an individual athlete, we illustrate the baseline
and control run results in a chart with the units in grams of drag.
2000
302
1500
202
1000
73
46 53 42 41 102
500 22 31 22
0 2
-10 -7.5 -5 0 5 7.5 10 Average
Beta Angle (Degrees)
Some portion of the uncertainty we can attribute to movement and variability in the riders position and some to the
removal and replacement of the equipment, or other setup differences that imperil repeatability. We detected 11 grams
of absolute uncertainty on average between baseline and control runs with a bike only, compared to 41 grams observed
with a live rider on board, so approximately 75% of the uncertainty (30 grams or ~3 watts) is due imperfect reproducibility
by a live pedaling rider.
500
400 102
300
200 24 52
12 9 9 11 10 11
100 3
0 2
-10 -7.5 -5 0 5 7.5 10 Average
Beta Angle (Degrees)
On a more technical level, we used the data from the multiple rider-on runs of the Felt B series and the Cervelo P5-6 to
calculate standard deviations for those bikes by yaw point. The data from those two bikes is sufficient to assess the errors
inherent in the tunnel and protocol and to chart confidence interval ranges by yaw point.
For the Felt, the 95% confidence interval for the zero-yaw condition in CdA is 0.2234 m2 +/- 0.0012 m2, which can also be
expressed as a range between 0.2222 m2 and 0.2246 m2. In grams of drag, that same interval at zero is +/- 14 grams, or
between 2,505 grams and 2,533 grams. This range captures all the zero yaw point estimates for the non-Cervelo bikes.
Rider On Testing
30 mph wind speed; A2 Wind Tunnel
0.2250
0.2200
0.2150
0.2100
CDA (M2)
0.2050
0.2000
0.1950
0.1900
0.1850
-12.5 -10 -7.5 -5 -2.5 0 2.5 5 7.5 10 12.5
ANGLE OF ATTACK (DEGREES)
Felt (with 95% C.I.) Andean Ventum P5-X Tactical P5-6 (with 95% C.I.)
For the Cervelo P5-6, the 95% confidence interval for the zero-yaw condition in CdA is +/- 0.0016 m2, or a range between
0.2155 m2 and 0.2187 m2. In grams of drag, that confidence interval at zero is +/- 18 grams, or between 2,420 grams and
2,477 grams. Those intervals across yaw nearly cover the P5-X point estimate data, but the drag difference between the
P5-6 and P5-X widens enough at yaw that we can be reasonably certain that the newest bike from Cervelo, as tested, is
faster.
The standard deviation of the measurements at higher yaw angles was greater than at zero yaw, which in combination
with the lower number of samples means the error range increases at yaw. Applying our simple weighting function to the
confidence intervals by yaw point, the Felt interval
Weighted-Average Drag widens to +/- 25 grams of drag, or to a CdA of +/-
30 mph wind speed; A2 Wind Tunnel
0.0022 m2.
0.2250
It is statistically invalid to calculate standard
0.2200 deviation data from one bike and apply it to
another bike that has a less robust sample, but we
0.2150
do so anyways in charting each bikes weighted-
CdA (m2 )
0.2100 average drag figure with error bars from the Felt
0.2186 (+/- 0.0022 m2). This is an illustration of how close
0.2050 0.2138 0.2154 these bikes are aerodynamically and how much
0.2134
0.2085 0.2103
0.2000
overlap there may be.
These overlays, statistically defensible or not,
0.1950
P5-X P5-6 Tactical Andean Felt B Ventum indicate that there are essentially two performance
Series groups rather than distinct levels for each bike.
Cervelos bikes are a touch faster, while the rest are
almost indistinguishable when accounting for variation and uncertainty from the tunnel, the protocol, the removal and
replacement of the bike on the platform, and the movement of the rider on board, all of which likely contribute about ~20
grams or ~0.002 m2 of variability in either direction of the drag data we report in discrete units.
Converting drag data into more meaningful real-world values requires additional inputs that are specific to the individual
athlete. Time assessments (e.g. seconds saved in a 40K or an IRONMAN bike leg) require environmental data that is
location and course-dependent. Because effective yaw angle is a probability distribution which is itself a function of wind
direction and speed as well as rider direction and speed, a separate model needs to be generated for each course and
rider.
David Bowden from velogicfit.com developed a proprietary model for interpretation of our results. This model was
constructed to accommodate the angles we tested so our output could serve as an input to a predictive model with real-
world location- and course-specific data from four IRONMAN and IRONMAN 70.3 courses. GPS and mapping data from
gpsies.com and weather data from wunderground.com formed the basis for the environmental parameters in the model.
This analytical approach has been tested and refined over the last ten years and has proven to be a reliable predictor of
rider performance and bulk trends in average race splits.
Despite the track record of these models, they are a best-effort representation of complex real-world phenomena and are
limited. Beyond that, the model we use for the analysis rests on discrete CdA inputs from our testing as a primary driver.
As covered in the previous section, our data is best conceptualized in ranges rather than finite points. However, we have
chosen to report time output data from the model discretely and to the nearest second. These figures should not be taken
as literal or definitive they are best viewed as a demonstration of what the differences in these bikes might be for a
course and athlete and condition, if we could be certain about the precision of the input data from our testing and the
accuracy of the parameters in the model.
IRONMAN Chattanooga 70.3: Our test rider, Kiley Austin-Young, rode the Felt B Series tested during the 2017 Chattanooga
70.3 to an official split of 2:16:32 over the 56.7-mile distance. In Tennessee and northern Georgia on race day, ambient
winds were light and largely in the direction of the rider, so yaw distribution was centered close to zero, with nearly 100%
of wind experienced by the rider falling between zero and 10 degrees.
IRONMAN Florida: The 4:04:39 ridden by Andrew Starykowicz at this event in 2012 set a new benchmark for IRONMAN
bike splits. If his bike at the time was equivalent to the Felt B series, Starky would have surrendered over two minutes to
the Cervelo P5-X. The conditions were windless in the first half of the ride, picking up later in the day. Thus, the yaw he
experienced was extremely low leading to an advantage for Cervelos bikes that is significant in the context of a record
attempt.
There are other limitations in this kind of modeling. The bikes tested were not weighed as part of the protocol, which
prevents us from incorporating the impact of equipment mass; however, the sensitivity to this omission is low: a 2 kg (4.4
lb) difference in bike weight would make a difference of 12 seconds for our 2:16:32 rider at IRONMAN 70.3 Chattanooga,
and it is unlikely that the difference in mass between the tested bikes exceeds 2 kg. Additionally, since the analysis is based
on weather station recorded data with a conversion factor for ground level wind speed, shielding from or swirling caused
by buildings and natural features are not accounted for. Finally, the higher the wind conditions modeled, the more prone
to error the analysis will be, given that the confidence intervals for our test data are wider at yaw than they are at zero.
VII. BIKE-ONLY TESTING RESULTS
750
performance, and advocates for a bike- 700
724
only protocol argue that the benefits of
repeatability outweigh the interactions 650 648
661
between the rider and the bike for the 600 597
same reason. On the other hand, some
550 552
companies like Dimond, in their 2013
research at the Faster tunnel in Arizona, 500
have found rank order to reverse course -10 -7.5 -5 0 5 7.5 10
Angl e of Atta ck (degrees)
in rider-on compared to bike-only tests.
Other companies in the industry have chosen to test in both rider-on and bike-only configurations and release one set of
data but not the other, which could be due to conflicting results vis--vis each other or lower performance relative to the
competitive set in one method.
As is evident in the chart, our results from bike-only testing are in several cases different from the rider-on output. These
differences are greater than the margins of error described above, so they cannot be solely attributed to rider variability.
We suspect Mr. Vroomens comments, made in 2007, are less relevant to todays products, which are designed as
complete systems while leveraging computational fluid dynamics (CFD) analysis that accounts for hydration, storage, rider-
bike interaction, and other elements that advantage a design for how a bike is utilized in the real world. Airfoil study in
the cycling industry has advanced considerably in the last decade, and Cervelos achievement in their development of the
P5-X was to engineer a system that minimizes drag in configurations athletes commonly use on race day. We think testing
a bike with a pedaling rider on board is important to account for the air flows dynamic interaction between the rider and
the bicycle. Despite the additional variability associated with this method relative to a bike-only or static mannequin
approach, aerodynamic interactions occur everywhere between body and bike and those interactions are critical for
assessing overall performance for both bike and rider.
There are additional reasons for
Bike-Only Testing
discounting the bike-only results or
30 mph wind speed; A2 Wind Tunnel
excluding them from our primary
Felt Andean Ventum P5-X Tactical P5-6
analysis: for unknown reasons, the
0.075
Premier Tactical and the Cervelo P5-X
0.070
were adversely affected during bike-
only testing. Aberrations occurred
0.065 during testing of the Premier Tactical
at a single yaw point, the first zero
CdA (m2)
Every manufacturer and third-party researcher takes a different approach to aerodynamic testing. Comparing data from
one wind tunnel test to another test is analytically indefensible. Even the smallest changes in testing protocol can skew
results. Tunnel architectures and statistical methods for interpreting the results differ. Even the same tunnel can produce
conflicting values on separate days. In rider-on testing, some athletes are more capable of repeating and holding position
from test to test, while others fatigue more rapidly during a day of testing.
2,350
2,358 2,352
2,300
2,250 2,292
The bikes in our testing all see a similar, relatively symmetrical drag arc across the sweep, except for Ventum, for which
there was materially higher drag detected at negative yaw than at positive yaw with the rider on board. No position errors
were observed, and the data sampling at zero yaw equals or exceeds that of any other bike in terms of consistency. We
can only conclude that the irregular flow detected is either due to the asymmetrical design of the frame itself or the result
of setup- or rider-specific variables that were unique to our test. The differences between our test results and those of
two other rider-on tests, performed on the same bike at the same tunnel, belies our confidence in the result, even as we
have no qualitative or quantitative reason to doubt the integrity of these data.
IX. CAVEATS
No testing is perfect. Aerodynamic studies and the statistical models we use to apply the results are useful in assessing
the merits of and making inferences about the worlds top race bikes, as we have done here. But the nature of yaw and
the complexity of real-world winds reduce our certitude with respect to the outcome. Wind speed and direction are
exceptionally noisy outdoors, and the bike is not fixed vertically as it is in the stable environment of the wind tunnel. The
meteorological data used in performance modeling data is imperfect, and does not account for the impact of objects at
ground level, like buildings or vehicles.
Wind tunnel testing is prone to other challenges and
sources of error. By way of example: even as we took
care to test both disc- and rim-brake wheelsets with
the same new tires, the Grand Prix TT (23 mm) from
Continental, the aerodynamics of the same type of
tire can differ out of the box. Chris and Jon Thornham
from Flo Cycling have completed extensive wind
tunnel research on wheels and tires and have found
deviations in the same brand and model of a new tire
that far exceed the ranges of error we saw in our
testing. Others have noted that the aerodynamic
properties of tires can drift during testing due to the
heat caused by friction from the rollers.
There are other sources of error related to equipment
that we accepted for reasons of practicality or necessity. We note the following with respect to our equipment and setup
choices:
Wheels: The disc-brake model rims from Enve were not the same as the rim-brake rims from Enve. At the time of
this writing in June 2017, Enve is the only company that offers different molds for disc-brake rims than its rim-
braking equivalent. We made this choice intentionally, because Enve has leveraged the design opportunity
afforded by disc braking to develop a new mold and product for the new road braking paradigm.
Extensions: The Premier Tactical had proprietary extensions as well as arm pads and cups. The Tacticals interfaces
are incompatible with third-party parts. The cockpit rise was thus significantly less on the Premier Tactical than
on the other bikes, which all used the same Zipp extensions with identical or nearly identical rise. Further, those
extensions were not all mounted in the same location relative to the handlebar or the pads. The Felt B Series was
equipped with an after-market Enve bar in an under-mount configuration, while the Andean and P5-X have
extension clamps on the same horizontal plane as the handlebar. Still, the extensions used allowed us to align the
position of the riders hands.
Hydration: The between-the-arms bottles for the rider-on tests were not in the exact same place relative to the
elbows. Certain bikes provide bosses, and others do not. We used the integrated bosses for the former and zip
ties where the bottle cage could be best secured for the latter.
Components: The Cervelo P5-6 comes stock with hydraulic brakes from Magura, and because they are well-
integrated into the design, we chose not to change the brake levers, which have a different aerodynamic profile
than the Shimano Di2 components we used for all other bikes.
Flat Kit & Tool Storage: The Cervelo P5-6 and the Ventum One were tested without frame-mounted tools or flat
kit materials. The Felt B Series was tested with a frame-mounted hydration accessory made by Torhans that
accommodates tool storage and comes stock with the bike, while the Cervelo P5-X, Premier Tactical, and
Diamondback Andean provide for storage in the frame itself or in a modular compartment designed to mate with
the frame. Like rear hydration, we know that these materials can be carried in a way where the impact is
aerodynamically de minimis or drag neutral.
Fit & Sizing: These bikes were all configured to a specific set of fit coordinates for the test rider. As such, the results
are most relevant to that rider or to a rider with similar fit coordinates or characteristics. Every bike in every size
has an aerodynamically optimal configuration, and every rider has an aerodynamically optimal position. What
works for one rider may not work for another, and these results will not universally apply to athletes of different
sizes and setups.
There may be other reasons to doubt the data. Without unlimited resources, every comparative bike study requires one
to make choices in protocol and configuration that will influence the results. Despite the caveats and limitations, we think
the takeaways from this testing are instructive for athletes and consumers as well as industry professionals alike:
Triathlon bikes are approaching aerodynamic parity. With a rider on board, just over one hundred grams of drag
difference or .01 m2 on a weighted-average basis were seen between the fastest bike in our study and the slowest.
Cervelos new P5-X is negligibly faster than the P5-6 that came before it, which was negligibly faster than the P4
that came before it. We as athletes and consumers now live in the age of peak aero.
Economy bike options and smaller, less-established products can be very fast. In some cases, they are faster
than newer or more expensive bikes. The Felt B Series which was configured with high-end after-market parts,
including an aerodynamic center-pull front brake from TriRig and an Enve handlebar performs as well as or
better than some of the top triathlon bikes, although it offers no integrated hydration or storage aside from the
Torhans unit. The Premier Tactical, an economical (for its spec), direct-to-consumer option with expansive
features and quality components, was the third fastest bike in the course simulation modeling, eclipsing the
Andean from Diamondback and Kevin Quan Studios. If the results of our study are any guide, retail price and
aerodynamic performance are not well correlated.
Cervelos continues its legacy of best-in-class aerodynamics and engineering. Like the P3C and P5 platforms, the
new P5-X is an achievement: this company produced a bike that is as aerodynamic or slightly more so than its
predecessor, which is widely regarded to be the fastest bike ever, while providing improved speed modulation
and greater stopping power from rotor-based braking, while offering integrated and modular options for
hydration and tool storage to meet the needs and behaviors of long-course triathletes. Our test lends credence
to Cervelos claims and to the integrity of their aerodynamic testing, which closely mirrored our own.
Comparative wind tunnel testing and subsequent data analysis is hard. This was an exceptionally difficult project,
and we are proud of our work and of the results, imperfect as they may be. Wind tunnel testing is an endeavor
that is at best flawed and at worst futile. Force measurement, even in a controlled environment, is prone to error
of both man and machine. Resource limitations can lead to statistically insignificant sampling, which reduces
certainty in the result. Models of the real world will never reflect the real world exactly, and even with our best
efforts and the most sophisticated quantitative instruments at our disposal, we still struggled to identify marginal
differences that were definitively larger than the ranges of possible error in our measurements.
This study was funded and supported by a community and an industry, designed by and for athletes who want to go faster,
and we learned about that here. Fast riding is a process that is never done we are always looking for gains, however
marginal, as we work to become better bike riders. At the highest level, the art of fast bike riding is all about the details,
and the closer you get to the limits of your physiology and your equipment, the more important the details become.
The bicycle itself is only one detail, and given the takeaways from this study, you may conclude that this detail commands
more attention and money than it merits. We love our bikes as machines and tools, but the performance gains possible
from this choice are narrowing: the aerodynamic arms race to optimize the speed of this piece of equipment is nearly
won. Welcome to the age of peak aero, where performance is no longer, or is much less, about the bike.
From the age of round-tubed, aluminum frames to Cervelos latest ground-breaking design, what is takes to succeed in
cycling has not changed. Tiny details and marginal gains are still out there, and for those of us who want nothing more
than to ride better, there will not be an end to the search. Even as some opportunities for improvement diminish, there
is an extra detail, another small gain. The new bikes may offer little improvement in speed, but there will always be
another way to go faster.
ACKNOWLEDGEMENTS
We are indebted and thankful to all who played a role supporting this project. Special thanks to:
Kiley Austin-Young is a competitive amateur triathlete and avid cyclist from Pennsylvania. His athletic accomplishments
include overall amateur podiums at the IRONMAN 70.3 distance, including one overall age group 70.3 title. Kiley graduated
from the University of Pennsylvania with a Bachelor of Arts in English and a concentration in Creative Writing and now
works in real estate development and technology in the Philadelphia area.
David Bowden is a competitive triathlete and cyclist with a degree in statistics and a keen interest in applying science to
endurance sports. Located in New Zealand, he is the founder of velogicfit.com and SPEEDTheory. A professional bike fitter
and triathlon coach, he also serves as technical advisor for Profile Design. A cycling equipment innovator, David was
instrumental in the development of the J5 and F40tt front end system from Profile Design.
APPENDIX