Nothing readily available on the consumer level, but the folks behind the STAC Zero trainer have a service that is similar to what you are describing.

https://www.staczero.com/vwt/

---

Strava

These are separate problems, with independent solutions.

If you can find a way to scan an object, it should be fairly easy to export a mesh of arbitrary quality. However, it will need a lot of editing to be nice enough for reasonable CFD work.

Blender is free and can do this job pretty well, but it will take some time depending on the quality you are after.

For actual fluid analysis, you can create a mesh using OpenFOAM’s (also free and open source) snappyhexmesh, providing the stl you’ve cleaned up and closed as a “watertight“ object. Once meshed, you could also try Stanford’s SU2 cfd code with some mesh conversion scripts.

It’s a long road :) but not too difficult depending on your background.

I tried the first part of that with the Kinect a few years ago. It didn't give very good results in terms of resolution and kinematic correctness. But it may be much better with the newer generations of these devices.

I have since decided that putting effort into measuring stuff while riding in the real world has a better quality-for-effort payoff than the CFD route. For now.

I'm a cfd engineer. If I imagine to work on an arbitrary scan only with open source software it seems like a nightmare for me. I would hire a bachelor student for that job...

Much easier and insightful is an outdoor ride with powermeter in aero position on a track/road fairly flat, nearly no wind and nearly constant power output (ideal conditions if possible). Repeat 3-5 times to find a good average value. Then analyse that ride with golden cheetah, average and you will find a very confidential cda value. And you need maybe one day for the hole stuff...

It’s true that nothing in the open source world matches the speed and convenience of software like ANSA or Geomagic, but Blender can compete as a subdiv-based, reverse-engineering tool.

Also, open source photogrammetry is pretty competitive with commercial codes in terms of quality, this was done in alicevision/meshroom and blender, using only a camera.

There is a big difference in image quality between picture 1 and 2, how do you explain that?

---

- Antony Costes -
PhD in Biomechanics / Professional Triathlete (9 pro wins)

"If you cannot measure it, you cannot improve it."
Lord Kelvin

The top image is a subdiv reverse-engineered surface based on the raw photogrammetry data in the second image.

I’ve worked as an an aerodynamics lead, (cfd and wind tunnel testing), where we had access to state-of-the art scanning systems (ATOS and similar), and frankly they are never good enough to use directly. All scans serve mainly as references for reverse engineering. That is to say, even if one had 200k+ to invest in metrology, you’d still end up reverse engineering most if not all of the object for quality results.

So, the nominal uncertainty in the scans may differ from about +/- 0.005” in the best scans (of large objects), and one might expect about +/- 0.05” in a carefully made photogrammetry scan... but the latter is free, and even 0.005” levels of noise are not acceptable for most production or cfd purposes, so practically, I’d say the free stuff is pretty competitive to commercial scanning systems in the 10’s of thousands or lower.

interesting, thanks for some of the links/info.
I do some dabbling with Blender (and F360)

Might give an attempt to fumble around with some of this

---

My Blog - http://leegoocrap.blogspot.com

I did lots of point cloud to mesh conversion for FEA in my previous job as a biomedical engineer and in my masters thesis. Not CFD, but a mesh is a mesh. We used everything from Artec to CT scanners.

It’s actually doable in solidworks scan to 3D module if one has a sufficiently high fidelity point cloud. This isn’t the best way and it needs ALOT of work, but doable. Obviously stuff like geomagics or hypermesh would be ideal but those are expensive.Not tried the Blender route.

Just to add that anyone processing a large enough mesh with enough fidelity for a bike +rider system to draw any meaningful conclusions is going to need some serious computer hardware....