Going down a rabbit hole - Parametric foil design and printable molds in Onshape

Jb69 · October 4, 2026, 1:27pm

Fair point on the volume. I’ll keep this one short.

Agreed, the joint is the hard part, and it’s where most of my last week went. Short answer for now: the semi socket. The fuselage’s nose, or a flat bar on a home build, drops into a pocket molded into a hub on top of the wing, with carbon cheeks either side. The cheeks take the twist and the side load, so the bolts only have to hold it down. The fit is set in the mold, 0.1 to 0.2 mm a side, rather than by sanding. The next thing I’m adding is bolting it from the wing side, up through solid laminate into the fuselage instead of into inserts, as a lot of production wings do.

The longer answer is that I’m not trying to pick one. In a parametric tool the joint is a dropdown and a few dimensions, so the variations are cheap. What’s useful is having options that fit what people already own. Right now that’s a flat mount, the semi socket, a full square socket for a bar and a round tube socket, plus a production-style mount pad on the stab. Next is a joint that mates with Gong’s V3 fuselage, so a molded wing bolts straight onto a fuselage that’s already common. That one waits on me buying a V3 and measuring it.

Whichever joint, the tool works out the loads in a hard carve (3 g by default) and checks the laminate against them:

root bending and the carbon caps
core shear
how much carbon is left under each bolt head
the fuselage in torsion

What it can’t tell me is play and fatigue over a season, and you’re right that that’s where the bigger foils struggle. That takes someone building one and riding it, which is the next step.

Jan12 · October 5, 2026, 1:46pm

I have been trying to get the stall speed of the AFS Silk 1050 and it is really hard to do.

Stall simply does not happen.

What I have figured out for my weight (95kg complete) is, 17.3 km/h is too slow. But I do not stall, but simply loose height over water until touchdown.

With constant pull I could probably set an AoA to stay up, but this would be a nonsense number with absurd drag.

So I think the minimum usable speed is around 18.5 km/h.

Jb69 · October 5, 2026, 2:43pm

Thanks Jan, that’s the most useful number in the thread: a real minimum speed on a wing I’ve modelled, with a weight.

“Stall simply does not happen” is what the Silk’s tubercles are supposed to do. When I went through the published tubercle tests (Miklosovic’s humpback flipper model, Johari, Hansen, Bolzon), the one result they all share is the shape of the stall, not the numbers before it. The stall starts in the troughs between the bumps. Each bump sheds a pair of vortices that keeps the flow attached behind it, so the stall stays confined to the troughs instead of spreading along the span. The wing doesn’t drop; it runs out of lift gradually while the drag climbs. Losing height until touchdown, and being able to hold it with a constant pull at absurd drag, is that same thing seen from the board. The tool can’t calculate it: a lifting line and a 2D polar don’t contain those vortices, so it reports the tubercles’ effect separately, as a range.

Where your numbers bite is how early it starts. I ran the Silk 1050 in the twin at 95 kg (with my guessed sections, since AFS don’t publish theirs):

Speed	Wing CL needed	Share of the section’s CL max	Board angle	On the water
13.2 to 13.6 km/h	about 1.2	100 %: the model’s stall		never reached
17.3 km/h	0.74	57 to 61 %	5.3 to 5.9°	too slow, sinks
18.5 km/h	0.65	50 to 54 %	4.3 to 4.8°	minimum usable

The ranges run from Ncrit 9 to Ncrit 1. So where you’re already sinking, the model thinks the wing has 40 % of its lift in hand. On this wing the soft stall sets in at about half the CL max a smooth section would reach. That’s the troughs, plus a real finish, the real sections, and a rider who isn’t a fixed weight, none of which the model has.

That gives me what I asked for back in post 3. For a tubercled wing like this one, the minimum usable speed is about 1.4 times the tool’s stall speed. Put another way, it’s where the take-off loading reaches 50 to 55 %, not the 85 % I’d been using as the limit. For an 85 kg rider on the same wing that’s about 17.5 km/h, where the tool quotes a 13.1 km/h stall. Your 18 km/h guess back in post 2, for a 1000 cm² wing at the same weight, lands in the same place. I’ll make your two numbers a check the model has to pass: at 95 kg on the Silk 1050, 17.3 has to come out below the usable limit and 18.5 above it. And I’ll stop quoting a stall speed for tubercled wings, and give the minimum usable speed instead.

Two questions, if you don’t mind. How were you riding it for this: winging, towed, pumping, or with a motor? It changes what “constant pull” means. And did anything log speed with power, current or heart rate? The model has the foil’s drag almost flat between those two speeds: about 40 N at 18.5 and 42 N at 17.3 for the wing and stab at 95 kg, under 60 N with a mast and fuselage. So if whatever was pushing you had to work much harder at 17.3 than at 18.5, that jump is the soft stall measured as drag. It’s exactly the part of the curve the model can’t see yet.

Jb69 · October 6, 2026, 2:33pm

A side branch of the rabbit hole: scanning a foil back into the tool.

The itch: nobody publishes a wing’s section. The best you can do for a production wing is “a NACA something at a guess”. So I’ve been building a scanner that goes the other way round from the molds: photograph a wing, get the tool’s own parameters out, and from there tweak it, re-mold it, or just know what you’re riding.

How it works so far:

An iPhone app that takes about 300 photos by itself as you walk round the wing, nothing to tap, nothing in your photo library. Depth sensors turned out too coarse for a section (LiDAR is a centimetre, the face camera a millimetre or two), so it’s photogrammetry.
Printed AprilTag targets round the wing for scale and reference: a plate of tags, two measured bars, a cradle for the fuselage, and a printed NACA test section in every shot so each scan reports its own error.
The Macbook rebuilds the 3D model, splits it into wing, fuselage and stab, slices it at stations and fits the tool’s section and planform laws, with an uncertainty on each number. It refuses what it can’t fit rather than guessing.
A separate printed jig and the phone’s own tilt sensor for incidence, twist and decalage, which the photos can’t do to a tenth of a degree.

The check so far: a wing built in the tool, every parameter known, exported and run through the fitting with phone-photo noise added. It came back with the camber to 0.04 % of chord, the thickness to 0.05 %, incidence to 0.08°, the winglet’s angle to 0.2° and even the tubercles’ pitch and height.