Aerodynamic Optimisation of a Vintage Commuter Bike

The Test

The test itself was devised as follows:

• Take system weight.
• Tire circumference roll out (one that required a system upgrade to add to the data analysis).
• Warm up – while unrealistic, I did choose to wear a skinsuit for this test. For engineering righteousness in comparing to my road bike, I also wanted to eliminate any noise from a flapping dress shirt or surface roughness interference from a misplaced sequin.
• Baseline – 2 baseline repeats of 6-8 out and backs of an approximately 700m course. I wanted to collect as much data as possible and settled on multiple runs of a shorter track than less of a longer distance. I predicted some position instability compared to what would be seen on a road bike so I figured using more runs to restabilise between each one, and have more measurements recorded per configuration would help data robustness.
• Position optimisation. This was going to be the most difficult part – would I be a consistent enough test subject; on a bike I have only been riding on for the last month?
• Baseline repeat. Easy.

The test area. A flat and straight back road, early enough in the morning to ensure most of the office commuters and school runs had been taken care of, and the spectacle that was about to commence would stay relatively out of sight. For testing integrity as well as protection of my own ego. Cyclists get enough slander as it is, and I was certainly a case to add to the library in this test state.

Some caveats, assumptions and already established test comparisons.

• GCN performed a performance analysis of a Brompton vs a Cannondale SuperSix EVO. This test uses an assumed CdA of 0.401 for the Brompton. Seeing as an upright cyclist is in the range of 0.35 – 0.5 m² we’ll take that as a good assumption and one we can further base the results off.
  
  
• The Brompton is equipped with Schwalbe Marathon Plus tires. Bicycle Rolling Resistance approximates these tires to be in the rolling resistance (Crr) range of ~0.0076 at 60 psi from their benchmark laboratory testing. However, with the 16-inch wheel size creates a wider and shorter road footprint compared to the tested 700cc so must scaled upwards. This gives us roughly a Crr of ~0.008 at 85 psi. I personally prefer a firmer tire, and this is the recommended pressure for city commuting.
  
  
• GCN also assumes a 95.5% drivetrain efficiency for its 2-speed comparison. I could dive more into the mechanical details, but for this test, I am focused mostly on the aerodynamics.
  
  
• Race analysis: Going back through my Garmin data, which unfortunately did not include power, I compared my race results from the A – Qualifying Lap, and B – Final Lap. A was a mixed gender group, so while I tried my best to jump on a wheel and draft, I was subsequently dropped and rode out the qualifiers mostly solo. That being said, removing the warmup lap, my average speed hung around the 30km/h, with a maximum of 41km/h. The finals were female only, and a tactical paceline ensued. I averaged 31km/h, maxxed out into almost 45km/h with an equally maximum heart rate of 198bpm (no comment). As far as aerodynamics goes, those are speeds where we can start to see some real effects of aerodynamic improvements, so I would also need to make sure my testing speeds could reflect this.

Results

3 hours of testing complete, and several strange looks and questions from the public, I finally concluded on the aerodynamic performance of my Brompton and was pleasantly surprised. It turns out, despite my reservations, this bike is engineered in line to cycling aerodynamic literature, and in hindsight, makes perfect sense that it does. There is no defiance of physics laws here.

Data

Streamlines recorded my test protocol as follows: seven runs, twenty-eight traverses. A traverse is one pass through a 693 m gated section that drops 8.8 m from one gate to the other; a run is a set of out and back.

Configurations A and B are the same position: an upright ‘commuter’ position. I split them because of my position stability reservation. A as my main baseline and B acting as a position repeatability check; three consecutive runs of that same position, followed by C, the aero tuck. Then the last run of the day goes back to A which acted as my baseline repeat to recheck my stability and determine any data drifts.

Note: A1 below is the pooled baselines. For Streamlines, this is best practice for data comparison and post processing. this makes comparison easier and the system works better on the calibrations. Below the A runs are parsed apart for data presentation, but CdA delta is compared to the pooled results.

Every traverse, in the order the configurations were ridden. Circles are eastbound, triangles westbound; bars are the mean and its standard error. The two A groups are the same configuration two and a half hours apart, and B is that position again.

Performance statistics

Five of the seven runs are the same upright position — the opening A, the three B runs and the closing A, twenty traverses spread across three hours. In the order they were ridden they came out at 0.524, 0.511, 0.512, 0.520 and 0.546 m². Mean 0.523, standard deviation 0.014, and a spread of 0.036 m². The results produced a coefficient of variation of 2.7%, which happens to be almost the lap-to-lap scatter of the session. (0.0146 m², 2.7%).

The baseline repeats help confirm the repeatability of the data. The first baseline, at 11:09, read 0.524 m². The last run of the day, the same position at 13:44 after the tuck was done, read 0.546. A drift of 0.022 m², about 4%, over two and a half hours.

The aero tuck takes 0.223 m² off the baseline — 0.313 against 0.535, a 42% reduction, worth 187 W at 40 km/h. Measured against all twenty upright traverses pooled, it is 0.210 m² and 40%. Those two answers agreeing to within 0.013 m² where the difference carries a Welch p of 0.0009 and a 95% confidence interval of 0.165 to 0.255 m².

Why the tuck saves so much

Any athlete would be delighted with a 0.223 m² reduction in CdA for a positional change. It is a seemingly unrealistic reduction of an almost 42% decrease, it does make sense when it comes down to the physics and aerodynamics.

In the aero tuck I rode 7.11 m/s on 93 W at the pedals. Upright I rode 6.28 m/s on 102 W. Faster, on less power. To hold the tuck’s speed sitting up would have taken 158 W. Air speed and power are both measured directly, so before any post processing, we already have some definitive performance numbers. CdA is just the aero power divided by the aero power a 1 m² body would absorb at the same air speed, so the raw ratios also highlight the reduction. The aero power went down by 11 W while the reference went up by 48%, because I was going faster: 

The magnitude of the decrease comes down to two separate mechanisms. The first is frontal area. Projected torso height scales with the sine of torso angle. Upright is somewhere around 75°; a forearm tuck is around 35 to 40°. That is a factor of 0.58 to 0.67 on the biggest contributor to frontal area, as well as considering the tuck also pulls the arms inside the shoulder line and tucks the head down.

The second is pure aerodynamics. A cyclist’s drag coefficient is not a constant and it depends on the Reynolds number where, for simplicity, is speed multiplied by the length of a body. The reference length that matters here is the torso: roughly 0.3 m across when you are sitting up, roughly 0.6 m once it is inclined. Tucking doubles it. On top of that, I was riding faster in the tuck, 7.1 m/s against 6.2. Combining these factors gives a Reynolds number result which creates a phenomenon of a low drag region. The product of this phenomenon makes 0.22 m² possible. See the Appendix for a more detailed analysis of Reynolds Number.

A decrease like this would be rare on a road bike, but this is not a road bike test, so the following must also be considering in explaining such a large CdA change:

• The starting point is much higher: My road bike wind tunnel upright figure is 0.358 m². The Brompton upright is 0.535 — 0.177 m² worse. The Brompton handlebars are high and close, so my torso sits near vertical in a way it never does on a road bike, even in a normal upright position.
  
  

• Geometry interaction: The Brompton is low and the bars are reached by a long stem, so my torso sits in open air. On a road bike, dropping into a tuck puts the torso partly behind bars, stem and head tube that are already disturbing the flow. Here there is less bike in the way, my aerodynamics lands much more on my own position.

• Nothing defines the end of the movement. No pads, no extensions, no reach limit. I could place myself in a position as far as flexibility allowed, which is further than a time trial fit would normally put me. The same freedom that made the position hard to repeat is what made it so low.

What would this mean in a race situation?

Assumptions first, because this is a model and not a measurement: the Lee Valley VeloPark road circuit at 1.62 km a lap, eight laps, 12.96 km, my race average 30–31 km/h. Rider and bike 70.2 kg, Crr 0.008 as reported, air density 1.18 kg/m³, 97% drivetrain efficiency. The Streamlines model reproduces the session report’s 40 km/h figures at 511 W upright, 324 W tucked.

At a constant 30 km/h the upright position needs 236 W at the pedals; the tuck needs 158 W. A saving of 78 W, or 33%. Aerodynamic drag is 80% of the total resistance even at that speed: 189 W of the 236, against 47 W of rolling resistance.

Held at the same 236 W instead of the same speed, the tuck rides 35.3 km/h. Over eight laps that is 25:55 against 22:00, 235 seconds, very nearly four minutes, or 29 seconds a lap.

This race prediction also comes with its own caveats:

• It assumes I can hold the tuck for twenty-two minutes. I held it for one run — four traverses, about seven minutes of riding. In the actual race, I really began to suffer lap by lap in such an aggressive position.
  
  
• The model is flat and constant speed. The circuit climbs about 5 m a lap, and the model does not account for any accelerations or cornering.
  
  
• Drafting is not modelled: Sitting in a paceline has significant impacts on aerodynamics. While I raced mostly solo for the qualifier, I was in a peloton for the final. Two very different aerodynamic scenarios.

As a final performance note, over the eight laps the tuck saves roughly 122 kJ of work. For a race settled by a sprint out of the last corner, that is arguably the more useful number: not the time on the clock, but what is left in the legs when you get there. As I am not a sprinter even on a traditional bike, I felt every lap in the final meters of the sprint, so I do believe having a more efficient race strategy overall could have saved me here.

Post Study Thought

Brompton Racing is a special event, with so much history, nuance and defiance of what constitutes as ‘normal cycling’. It deserves its own place in cycling, with its own style of race, format, and unwritten rules, which after this study, I came to believe that aero testing may not even be part of it. The Brompton World Champion Race Rulebook maintains that no aero fairings are to be present on your racing bike, therefore my optimisation would not be relying on any sort of time trial extension of between the arm bottle mounts. This left me to my position and race attire selection which of course I could select a more efficient costume for race day. However, in the spirit of real-life applicable engineering as well as that of the heart of Brompton racing, I fear that in optimising my apparel, would lead to de-optimisation of fun where fun seems to be the most mandatory rule.

That being said, I can certainly train my position a bit better as shown here, the aero tuck does have quite a significant improvement on my race strategy. Coming into next year’s race with more than a months’ time of practice, being a bit more versed in a folding bike paceline and maybe a more ridiculous costume can all be part of this optimisation and hopefully lead to a few spots higher up the leaderboard.

I believe there is a certain je ne sais quoi about cycling on a, for lack of a better term, ‘suboptimal’ vintage bike, one that exists in pushing ancient components far beyond their limits, but certainly not past their lifetimes. I could upgrade to the latest drivetrain, swap out to carbon cranks, update my tires, but if we think about our ancestral cyclists before us and their almost opposition of optimisation; steel frames, no helmets, donning cotton jerseys as they powered through to victory with sheer willpower, brute force and determination, a cheeky mid stage cigarette, I think some things are best left to the past, more so as a testament to the robust engineering. These machines were meant to last; when AI finally destroys our electrical systems and we can no longer charge our Di2’s, what would be left in the rubble? My sturdy L line, waiting to take me from A to B, and with my head tucked low, maybe a little bit faster than originally made.

Appendix

CdA figures are the Streamlines normalised values unless the column says raw.

A1. Run by run

A2. Every traverse

Direction W is westbound, E eastbound. Every configuration has equal numbers of each, so the configuration means are not affected by the direction split. Speeds are in m/s; air speed is what the probe saw, ground speed what the wheel saw.

A3. Session settings and normalisation

A4. Reynolds Number

The Reynolds number is the ratio of inertial to viscous forces in a flow, Re = vL/ν, where v is air speed, L a reference length and ν the kinematic viscosity of air — about 1.48×10⁻⁵ m²/s at 15 °C. It governs whether the boundary layer over a curved body stays laminar and separates early, leaving a wide wake, or transitions to turbulent and clings on further round, leaving a narrow one. Across that transition the drag coefficient of a rounded bluff body (a cyclist) falls sharply. The canonical onset for a smooth circular cylinder is Re ≈ 2×10⁵; surface roughness moves it down, to as low as Re ≈ 5×10⁴ for a rough body.

Taking the torso as the reference body and its streamwise extent as L, with air speeds as reported per configuration:

• L is nominal, not measured. The torso is not a cylinder, and 0.3 m and 0.6 m are representative dimensions for an upright and an inclined torso rather than sensor readings. Reynolds number scales linearly with L, so both figures move directly with whatever reference length you pick. 
  
  
• The transition is a range, not a point. 2×10⁵ is the smooth-cylinder value. Clothing, helmet, hair and skin all trip the boundary layer earlier, and a sufficiently rough body can transition nearer 5×10⁴ which would put the upright case at or past onset and shrink the difference between the two columns.
  
  
• Viscosity moves with temperature. Across the session’s 12–22 °C, ν spans roughly 1.45 to 1.53×10⁻⁵ m²/s, about ±3% on every Reynolds number above.

• Speed and position are confounded. The tuck was ridden faster than the uprights and no configuration was deliberately swept across a range of speeds, so this session cannot separate the speed contribution from the position contribution. The figures argue that the transition sits in the right place to be part of the explanation. They do not measure how much of the 0.223 m² it accounts for, and nothing here should be read as doing so.

The normalisation assumes one slope for everything. The session applies a single linear air speed correction of +0.00159 m² per m/s to every configuration. Published field data put real CdA-against-air-speed slopes anywhere between −0.00406 and +0.0012 m² per m/s, so any configuration whose true slope sits away from the standard value carries a residual the normalisation has not removed.

Images from  @B_W_C_P