Machine Learning for Neutral-Beam Modeling
Surrogates approximate neutral-beam deposition and current drive, giving fast estimates of heating and driven current for scenario design and control.
Neutral beams as an actuator
A neutral beam injects fast neutral atoms that ionize and deposit energy and momentum in the plasma, providing heating, current drive, and fueling. Predicting where and how much a beam deposits requires modeling its penetration and the resulting fast-ion population, which is computationally involved.
What the surrogate learns
- Deposition profile: where beam power and particles are absorbed
- Driven current: the current the beam sustains
- Fast-ion pressure: contribution to plasma pressure and stability
- Shine-through: fraction passing through without absorption
Trained on beam-deposition code outputs across plasma conditions and beam settings, a surrogate returns these quantities instantly. This lets scenario studies and controllers account for beam effects without launching the full deposition calculation each time.
Control use
Because beams are a primary actuator for shaping temperature and current profiles, a fast deposition surrogate is valuable inside control and reinforcement-learning loops. It lets the controller predict the effect of a beam command before issuing it, improving profile tracking.
Cautions
Deposition depends sensitively on density and geometry, so the surrogate must cover the operating range and handle edge cases like high-density shine-through limits. Fast-ion behavior can also drive instabilities not captured by a deposition-only model, so surrogates are paired with stability checks.
Heating and current-drive planning is part of scenario design for concepts like the Hyperion breeder. In Kronos design work beam surrogates are developed against simulations ahead of construction; they model an actuator on a plasma that does not yet exist, and their outputs are computational estimates.