Our Aero Design Process

A few people have understandably asked us, “How can three nerds make a faster road bike than the rest of the industry?” And we get it-- extraordinary claims require extraordinary justification. Today I want to talk about the aerodynamic side of the answer to that question (in part because due to my ignorance. Stephen might - or should - judge me if I try to talk about the other stuff). 

 

What is it we do differently?  Aerodynamics isn’t a new science - some people way smarter than me published the equations we use for most aero simulation by the mid-1800s, back when steam engines were the new hotness. CFD was invented back when the word “computer” meant people who were good at math. So we’re not doing anything crazy here. Simply put, we do two things, and we do them well:

1. We use the right tools for the job. 
2. We follow the science - design happens as a result of data; not feelings, budgets, deadlines, or marketing. 

So (in detail) what does that mean exactly?

 

Using the Right Tool for the Job

Ask any aerospace engineer how they design an airplane, and not a single one will tell you they start with full-fat CFD (computational fluid dynamics), like some of the bike industry does. Why not? Because experience has shown there’s much better tools to begin with. 

 

Just as you wouldn’t start a woodworking job with 3000-grit sandpaper, you shouldn’t start aerodynamic design with full Navier-Stokes CFD. You start with a saw or a coarse file, get nearer the final result, and then move to more and more time-consuming methods as you get closer and closer to your desired result. It’s the same with aerodynamics. 

 

So what’s the rough tool for aero? There’s plenty of options. At the roughest, there’s back-of-the-napkin estimation. If I want to reduce bike drag by X, what CdA do I need to achieve, and what coefficient of drag (Cd) and/or area (A) combination would I need to get there? Simple stuff like this can tell us as engineers how aggressive we need to get to achieve what we want to - this is how we decided, for example, that the RAKTT could use a less proprietary stem attachment method than the competition and still beat them in performance. 

 

Next up comes the physics-based computer simulations. The Navier-Stokes equations (what many use) are great, but they’re expensive. There are many simplified sets of equations that cover some of the physics of aerodynamics, and with experience and careful understanding of their limitations, you can do really solid early design using them. I won’t nerd out too hard here, but if you’re interested, some of the more useful equation sets for bike-relevant fluid flow include (in order of roughly increasing cost) vortex lattice, potential, full potential, Euler, Reynolds-averaged Navier Stokes (RANS, what most folks use), unsteady RANS, hybrid RANS-LES (large eddy simulation), LES, and DNS (direct numerical simulation). All but that last one make some amount of assumptions about the nature of the flow’s physics, giving up some of the real world physics to reduce cost, all while providing useful information. No model is perfect, but some are useful. 

 

Until now, we’ve just been cranking away on increasingly hefty computers. The next step up in complexity comes by doing the world’s best fluid flow simulation: letting the actual fluids do the simulation for us. There’s various levels of complexity here too.  Most cycling wind tunnels are low in cost, but the tradeoff is that they have very high turbulence levels, and are therefore less like the real world’s fluids. To achieve low turbulence, a wind tunnel uses screens in front of the test section to smooth the air, but those screens slow the air too, and cost more energy as a result. A smaller, high turbulence tunnel can run on hundreds of dollars of power per hour, while the bigger, very low turbulence tunnels can cost hundreds of thousands. Sometimes, the turbulence levels matter, and sometimes they don’t. The magic is in knowing when and by how much the turbulence levels matter so that you can apply the results of a given tunnel appropriately. As an interesting aside, turbulence levels matter a LOT for cycling clothing, and our use of high turbulence tunnels in the cycling industry is why I think we have a long way to go until peak clothing aerodynamics. 

 

One of the many wind tunnels I wish we could afford in cycling - check out the tiny car on far right. Image credit Oleg Alexandrov, under the Creative Commons Attribution-Share Alike 3.0 Unported license.

 

Finally, as you might guess, there’s actual real-world testing. It’s often expensive, because you need a functional bike and specific instrumentation, and then because the real world is chaotic and unpredictable, you often have to run many tests to rule out variables like wind variation, elevation changes, rider positioning, etc. We’re talking about methods beyond the Chung method here. That’s great for getting overall CdA, but with the right instrumentation and the kind of testing I’m talking about, you can characterize the flow over the entire bike and literally see aerodynamics in the real world, which is more challenging but also far more valuable. To my knowledge, Stromm is the only company in the bike industry that has done that.

 

Following the Science

At Stromm, we use most of these methods. We start with the rough tools, refine our designs a little, move up in complexity, and then refine a little more -- stepping up and down in complexity as needed until we get to a design we want to build and ride. We do as many cycles of this as needed to get the bike we want. If a bike isn’t performing the way we want, we try again until it does. Deadlines don’t rule our design process, only results. Maybe most importantly though, we learn as we go. Each time we step up in complexity of our tools, we learn a little bit about where the simpler tools worked well and where they didn’t, and we go back and apply that learning to better use the simpler tools next time. The RAKTT is our first public road bike, but it’s actually the fourth road bike we’ve built into a rideable machine, and each of those past three bikes allowed us to hone our tool usage until it was good enough to be very, VERY confident in. We can get higher accuracy out of our lower fidelity tools than many can achieve using much more expensive Navier-Stokes simulations.

 

On most of those bikes, we performed on-road aerodynamic testing. Each helped us figure out exactly where on the bike, and in what wind environments, our wind tunnel and computer modeling gave us the right answer (or not), and by how much. As far as I’m aware, there are no other cycling companies that do this, but it’s incredibly helpful because it helps us check our work. It allows us to re-tune all our models where needed to get the best accuracy predictions at the lowest possible cost. And this iterative learning process is how we can be aerospace-level confident in our performance claims.

 

You’ll notice I didn’t mention budgets, marketing, or deadlines much here. It took us a full decade to make a road bike we were happy enough to sell. I haven’t counted the number of design iterations for each tube, or the number of simulations we’ve run, but it’s A LOT. We’ve thrown out plenty of designs we thought were really cool and revolutionary, because despite looking impressive they just didn’t work as well. Would designs that looked really radically different have made the marketing easier? Maybe, but the bike would have been slower or less usable. And while we don’t use tools that are needlessly expensive, we don’t shrink from spending what’s necessary to get the job done. The majority of our company’s spending is on R&D of some kind. The science comes first, and we figure out how to make the logistics work once we’re happy we’ve got a great design that we ourselves would want to ride. 

 

Overall, using the right tools for the job and following the science allows us to get more performance out of our R&D process. We spent a great deal of time and effort refining this process, but the best way to make the fastest road bike in the world is to first make sure you have the most effective design tools and process. 

 

So What About RAKTT?

Hopefully those details are as cool and interesting to you as they are to us, but what about the specifics of how all that applied to our RAKTT road bike? 

 

For the RAKTT, we used our lower-fidelity tools to make several promising concepts. Each concept had a markedly different tube layout, and all were modular to allow us to try different combinations. Once we were happy with each within the accuracy our low-fidelity tools could tell us, we spot-checked with higher fidelity tools where needed to make sure we were on the right track. We wound up taking all of those concepts to the wind tunnel, which we do early in the design process so that we can have time to apply what we learn in the tunnel. Then we revamped the winning concepts to improve them where possible and combine them into a cohesive frame design. 

 

Overall, we were very happy with the process. We achieved aerospace-level accuracy comparing between our low-fidelity tools and the wind tunnel. That’s a big accomplishment, even for major defense contractors, much less for three nerds in a garage, so please forgive us for tooting our own horn here. Another interesting tidbit - the combination of concepts we thought would be fastest wasn’t. It was a little more proprietary but had no meaningful drag savings. That surprised us a bit, but it’s good news-- we will happily make a less proprietary bike that is just as fast as our crazier ideas. 

 

Our low-fidelity design process is incredibly proprietary - it’s the secret sauce that allows us to compete with the bigger players in the aero road market. I can’t share much, but we spend most of our design effort here. As an example, we spent serious effort on 70+ tube shapes for the fork alone, which came out of hundreds of lesser candidates, and thousands of prior options before that. And again, we’ve got our low-fidelity-to-reality corrections down well enough that what we predicted was always within 5% of the real world, and often closer to within 1-2%. 

 

Just of few of the thousands of tube shapes tested in the interest of going faster.  

 

Our high-fidelity design process is much more run of the mill. We use the OpenFOAM open source CFD solver and ecosystem. If we’re so confident in our low-fidelity, you might ask, why do we need this? Well, our low-fidelity can only be as accurate as they are through calibration with the real world. This higher fidelity CFD gives us early warning signs of when we might be stepping outside the design space where we’re confident in that calibration, which can help save wind tunnel time and tell us where we should best spend that wind tunnel time. Tunnel time is very close to the absolute truth, but it is expensive. High fidelity CFD tells us where we require that higher cost to calibrate our other tools. 

 

For wind tunnel testing, we did several sorts of tests:

1. Small change comparisons: Here we’re after 0.5-watt differences in shaping.  We just test the bike itself to get the low uncertainty we need to make the smallest of shape changes. Larger changes happen via swapping 3D-printed sections of the bike, while the smallest changes happen via custom heat-moulded clay additions in real time. Ever wonder how our bike can get great wheel-frame interaction without being as closely spaced as others?  Small clay changes in the wind tunnel is how!

2. Big-picture benchmarking: Here we’re seeing how we stack up against competitors and prior designs. We test several variations here:

   1. Like-to-like against competitors - we use the most similar components possible, even when that means giving our bike a disadvantage. While we could claim the more aero-friendly components that we sell the bike with as an advantage, we don’t. Why? Because those advantages are due to the components, not RAKTT itself. We do these tests without a rider for accuracy, and with a stationary, cranks-horizontal lower body dummy (similar to the one used by Tour Magazine) for real-world applicability. Stationary legs in that position have been shown by researchers smarter than myself to be an excellent average of the total effect of moving legs, without all the data uncertainty added by that motion. 

   2. Component choice: We test different component choices to understand how they affect the benchmarking above, and to make sure any public claims we make about our bike hold up under any reasonable use case.  Some of our customers might want to take their RAKTT on hillier terrain - do shallower wheels make our fork no longer as aero? We want to test that so we can design around it. Some of our customers will run without bottles, use aero bottles, or round bottles - how do each affect our delta to competitors? These tests tell us that answer. 

Out of all the wind tunnel testing that we do, the test data we release publicly is our best attempt to give people the most accurate, real-world “what performance can I expect to get no matter what bike I’m moving from and no matter what components I use?” answer. While that means there are many situations in which the RAKTT will be more than 10 watts faster than others, we don’t want to advertise any more than we can confidently promise. Overall, that means our public data is based on a test without bottles, with a stationary lower-leg dummy, 170 mm cranks, and 40 mm deep wheels, because that’s our worst-case scenario for the delta between our bike and others. You typically carry two round bottles? Great, so do I! We’ll both likely see an even greater performance benefit to RAKTT, because our bottle shielding is excellent. You have deeper wheels? Great choice, me too - they will, at the very worst, make no difference to that 10-watt/40 kph performance delta. You’re going to switch to the shorter cranks the RAKTT will come stock? Great, you’ll get even more performance than we advertise.

We don’t want a single customer to get less performance improvement than what we’re promising them, so that’s the data we release -- even when there’s data that might make us look even better. 

 

Would other data make us look better?  Sure - but who cares, if it doesn't make you faster in the real world?

 

As an aside, we want to make riders faster-- not shame other manufacturers doing their best to make great bikes. Which is why we won’t be releasing which bike(s) we tested against. We’ll happily stand by our performance claim against any of the fastest bikes tested by Tour Magazine as of now. 

 

Finally, as we always do with our bikes, we’ll do on-road testing to learn for next time.  How we do this is also very proprietary, but it helps us calibrate for that final difference between the wind tunnel and real-world air conditions. We’ve already learned all we needed to design the RAKTT, but on-road testing with RAKTT will be vital for some future projects we have in mind. 

 

I hope this helps answer some questions we’ve gotten on our process. Feel free to keep asking if there’s more you’re curious about! We can’t share everything, because then others could do what we do (and we like being able to buy food), but we’re big proponents of transparency in aerodynamics, and we want to encourage our customers and other companies to ask and answer these sorts of questions. Truth in aerodynamics makes us all faster, and we’re here for it.

 

Thanks for reading!