DIY computational fluid dynamics for aerodynamics?

Question

Has anyone experimented with using consumer 3D software to do CFD-like aerodynamics simulations?

Specifically, I was thinking:

• Setup the bike on a trainer (with rider)

• Take a series photos, and plug them photos->3D model software (something like AgiSoft PhotoScan or such as any of these free applications)

  This would hopefully give you a reasonable 3D model of the bike. Would likely require some cleanup, and possibly modelling of smaller elements if they are relevant (like spokes/chain)

• Bring the model into a 3D package with a fluid solver (e.g SideFX's Houdini)

• Simulate fluid (air) flowing around the bike model

• Somehow extract useful numbers, measurements or possible position-adjustments from the simulation

With my limited knowledge of Houdini, I manage to make something kind of interesting looking, using it's fluid-solver's velocity visualiser:

What use could all this be? For example, could this method be used to:

• see the impact various riding positions have on the flow of air (although I guess this is slightly hard to verify without a wind-tunnel)

• you could theoretically estimate CdA by seeing how much the object is pushed by the fluid. This could be verified relatively easily with a power-meter, e.g with the Chung Method

Could there be any worthwhile insights gained from this? Could it be any more effective than the more elegant method of using a power-meter?

Answer 1

I think you could take this technique pretty far, but probably not all the way (at least usefully).

I see a lot of use for the first part of your plan: taking multiple pictures and reconstructing a 3D model. If you try this for multiple postures I think it should be possible to determine an optimal position for you on your bike based on just looking at the 2D area you take up perpendicular to your riding direction (through the cutout method described in the article about the chung method you linked).

But this is where the use stops I fear. Some more insight might be gained by looking at the flow of air around your 3D model in something such as the fluid solver you link, but the accuracy is going to be too low to discern between different postures. When you go for that level of detail small things such as the ripples in your shirt or even the fabric of your shirt can have an impact, and the errors you made in the 3D reconstruction will be far larger than the differences you can make by slightly improving your posture on the bike even more than you did in the first step. Apart from this, I think that a fluid solver for movie effects is optimized for visuals first, not for scientific accuracy, so I'd be suspicious of any result it produces.

If you want to try the full fluid mechanics simulation thing you would have to use a 3D scanner or some other very accurate method to produce the 3D models of you on your bike, and then I'd still try and find an actual scientific fluid solver.

Answer 2

Has anyone experimented with using consumer 3D software to do CFD-like aerodynamics simulations? - Yes

Could there be any worthwhile insights gained from this? Yes

However, you should not view work like this as an alternative to using a powermeter or windtunnel testing. View it as a tool that can be used as an extension of sound methodical testing, to test variations that may not be possible or very difficult to test. There are so many pitfalls that scientists in the field consider it a natural part of their best practices to at least partially back up CFD results with windtunnel tests whenever possible. If you are interestid in this type of work, I recommend the Sports And Building Aerodynamics MOOC on Coursera.

Short of getting heavily involved in all of this, it is worth noting that frontal area plays a much larger role than shape of an object. As noted in the other answer, using image analysis to minimize your frontal area will likely have a much larger effort-to-reward ratio.

Answer 3

The Stac Zero Virtual Wind Tunnel has been developed and released since this question was originally asked. I saw an analysis on someone who has tested in both wind tunnels and on velodromes. The analysis I saw was in the right ballpark but a little different -- different enough that I suspect it would have had difficulty with things like skinsuits and shoe covers and other "small" differences. The analysis I saw seemed like it would be able to pick up changes in head or shoulder or arm position though I can't be sure. I think it had promise, not as a final arbiter between small things but possibly as a first cut to winnow through lots of alternatives down to a handful that you'd want to do further, careful, testing on.