Tokamak fusion reactor experiments aim to achieve self-heating plasma, generating enough energy from fusion reactions to sustain itself. A key challenge is plasma equilibrium reconstruction, which uses diagnostic signals to determine the plasma’s shape, pressure, fluxes, and current profiles inside the reactor.
We are seeking a student to apply machine learning (ML) methods for real-time equilibrium reconstruction using data from the Alcator C-Mod tokamak.
Responsibilities include:
• Designing and fine-tuning ML algorithms (e.g., neural operator networks (CNNs, FNO, KANs),
Bayesian models), applied to tokamak equilibrium reconstruction.
• Training these algorithms on real Alcator C-Mod diagnostic data.
• Evaluating performance using robust metrics.
• Refining model outputs to improve the predictability and reliability for next gen tokamak like SPARC.
Throughout the project, the student will also learn how plasma properties are measured and why accurately predicting plasma shapes and profiles is vital for successful tokamak operation.
The result of this UROP will be an end-to-end ML pipeline that boosts the accuracy, speed, and interpretability of plasma equilibria – a great opportunity for anyone passionate about machine learning and cutting-edge fusion research.
Our Group website : https://disruptions.mit.edu/
Head to MIT's UROP site to apply for this opportunity.