First try at AI for Design & CAD with CFD Simulation for Aerodynamics

#AI + Simulation: a 1-hour aerodynamics workflow (Gemini → Copilot3D → Stallion 3D)

Sometimes the fastest way to learn is to let a messy workflow happen… then measure what physics says.

Quick summary

1) I asked Gemini for the “best airfoil shape” → it generated something that looks like a supercritical airfoil.
2) Microsoft Copilot 3D turned that into an STL → surprisingly it became a biplane-ish configuration.
3) Stallion 3D solved the CAD at M = 0.825 within the hour.

Figure: (1) Gemini “best airfoil shape” prompt result, (2) Copilot3D STL interpretation, (3) Stallion 3D pressure visualization + Cp plot.

Step 1 — Asking AI for “the best airfoil shape”

I used a deliberately vague prompt: “draw a picture of the best airfoil shape”. AI doesn’t know your mission requirements (Re, Mach, thickness constraints, lift target, stall margin, structure, manufacturing, etc.), so the output is always going to be a guess—but it’s still interesting what it “reaches for” when asked.

In this case, Gemini returned an airfoil that looks supercritical-ish: thicker mid-chord, flatter upper surface, and a sharper-ish trailing region. Is it “best”? No. But it’s a recognizable design intent: manage transonic pressure gradients and reduce wave drag.

Step 2 — Converting the concept into geometry (and getting a surprise)

Next, Copilot3D generated an STL from the concept image. Here’s the fun part: it didn’t produce a clean monoplane wing. It produced something closer to a biplane / joined-surface interpretation.

This is a good reminder that “image → CAD” isn’t a deterministic pipeline yet. The tool is inferring 3D structure from ambiguous cues—so you can get creative geometry even if you didn’t ask for it. That’s not a failure. It’s a feature (as long as you validate the aerodynamics).

Step 3 — Let physics vote (Stallion 3D at M = 0.825)

Once the STL exists, you can stop debating what the shape “means” and just run it. I brought the CAD into Stallion 3D and solved at Mach 0.825. From there, the workflow becomes familiar: surfaces, pressure/Cp trends, and whatever integrated outputs you care about (lift, drag, moments).

The point isn’t that the AI created a production-ready aircraft. The point is that you can now move from an AI sketch to a solvable geometry to CFD-based insight fast enough to iterate.

What this workflow is (and isn’t)

• It is: a rapid way to generate “candidate geometry” when you’re brainstorming.
• It is: a quick filter—physics can reject bad ideas early, before they waste days.
• It isn’t: an optimizer, a certification path, or a substitute for requirements-driven design.
• It isn’t: proof that AI “understands aerodynamics.” It’s proof that AI can accelerate the setup—and CFD can validate the result.

Try your own workflow

If you want to run this experiment yourself, keep it simple:

1. Ask an AI for a concept (airfoil, wing, fairing, inlet—anything).
2. Convert to STL (expect surprises).
3. Run a quick CFD sweep (one condition is enough to learn something).
4. Decide what to keep, what to change, and repeat.

Try your workflow ➡️ https://www.hanleyinnovations.com/stallion3d.html 

(If you’d like, send me your STL + flight condition and I’ll tell you what I’d look at first: Cp trends, shocks, separation risk, and the integrated forces/moments.)