PropulsionLab · Neural surrogate
Turbojet performance, predicted in-browser.
A small neural network, trained from scratch on thousands of solver evaluations, approximates the turbojet cycle. Move the sliders for instant predictions with no server round-trip, then verify any point against the exact physics.
Loading model…Design & flight inputs
Predicted performance
,
Model card · held-out accuracy
The cycle is a smooth, deterministic function, so high accuracy here is expected, this demonstrates the full pipeline (space-filling sampling → from-scratch backprop → held-out validation → in-browser inference), not a hard benchmark. Predictions are trustworthy only inside the sampled ranges; the exact solver remains the source of truth.