Overview

• HEAT-ML is a deep-learning surrogate trained on roughly 1,000 HEAT simulations to predict magnetic shadow masks in fusion vessels.
• The AI model cuts computation times for shadow-mask simulations from about 30 minutes per run to just a few milliseconds.
• Demonstrated on a 15-tile section of SPARC’s exhaust system, the tool currently applies only to that geometry and is available as an optional HEAT setting.
• The research team is pursuing generalization across diverse tokamak components and broader validation to extend the model’s applicability.
• By embedding HEAT-ML into design workflows and future operational decision support, developers aim to help SPARC reach net energy gain by 2027.