Overview

• A public-private collaboration of Commonwealth Fusion Systems, PPPL and ORNL transformed the open-source HEAT toolkit into HEAT-ML, an AI surrogate for mapping magnetic shadows.
• HEAT-ML uses a neural network trained on about 1,000 HEAT simulations to shrink field-line tracing times from roughly 30 minutes to a few milliseconds.
• Demonstrated on 15 tiles in SPARC’s exhaust system, HEAT-ML operates as an optional AI mode within HEAT tailored to that specific geometry.
• Magnetic shadow maps identify surfaces shielded from direct plasma contact, guiding durable component design and preventing heat-induced damage.
• Next efforts focus on validating and extending the model to other reactor geometries and embedding it into real-time operational control ahead of SPARC’s 2027 net-gain objective.