What “AI-designed” means
There are several layers, and they are often mixed up:
- Generative geometry. The engine is described as code and rules, so a new requirement produces a complete new geometry automatically.
- Fast physics models. Neural networks trained on simulations, such as physics-informed networks or neural operators, predict flow and heat in milliseconds instead of hours.
- Search. Optimisers explore thousands of variants against thrust, mass and wall temperature.
What it cannot do
A model is only as good as the physics and data behind it. An AI-proposed design must be checked with established solvers, and finally on a test stand. Language models in particular should plan and write code, not estimate physical numbers.
Where it helps most
Where the design space is too large for engineers to explore by hand: cooling-channel layouts that change along the engine, injector patterns, and, in electric propulsion, magnetic field shapes for plasma thrusters.