AI cuts turbine cooling air 7.8% with just $14 of compute on NASA C3X vane
hbouammar · x · 2026-11-14
Inspired by Elon Musk's comments on gas turbines, the author shares an AI optimization project on the NASA C3X turbine vane (tested in the 80s, with 10 internal cooling tubes).
The question: how little cooling air can be used without the hottest metal exceeding NASA's own design temperature? The best design, verified in full simulation, uses 7.8% less cooling air while running 1.1 K cooler at the hottest point than NASA's layout. All 5 optimization runs found savings, and total compute cost was about $14 — his argument that "AI can be cheap if you do it right." He's seeking domain experts to collaborate.
More from Research
- Walden talk: base models improving fast, mid-training data commoditizing — ZeYanjie · 2026-11-14
- Pharma's two blind spots on AI labs' bid to cure disease, argued in viral thread — bookwormengr · 2026-11-14
- A clear video deriving cosine similarity from dot products, both algebraic and geometric — johnolafenwa · 2026-11-14
- Why 'Mathematics for Machine Learning' keeps getting recommended for refreshing ML math — Abhishekcur · 2026-11-14
- IBM trains small open models to ignore hidden injection attacks, cutting success to near zero — mawcopolow · 2026-11-14
- Delaunay-based nearest-neighbor search delivers 57x faster ambient occlusion than Three.js GTAO — ssh4net · 2026-11-14