Rapid optimization of SPARC first campaign scenarios

ORAL

Abstract

Preparation for the first SPARC campaign motivates development of a workflow to design pulse trajectories that lie within the SPARC operational envelope. One critical tool is GSPulse, a novel algorithm that allows an operator to design full pulse coil current trajectories to achieve a specified equilibrium evolution while satisfying device limits on the coil currents, power supply voltages, and structural forces. GSPulse solves the free-boundary equilibrium evolution (FBEE) equations in an efficient manner by rearranging them to separate the trajectory optimization from the plasma equilibrium, resulting in a calculation much faster than typical predictive control simulations. Coupling GSPulse with the transport code RAPTOR ensures consistency with the expected evolution of thermal energy and current profiles, enabling resolution of mixed equilibrium and transport phenomena. Using a wide range of objectives and assumptions relevant to the first SPARC campaign, this workflow is used to design trajectories that: minimize ohmic flux consumption, reduce peak heat flux by sweeping strike points along divertor tiles, and ramp down the plasma current rapidly while avoiding MHD instabilities. Recent work extends the scenario optimization to include heating and fueling actuators, using gradient-based optimization with RAPTOR, as well as replacing RAPTOR with the auto-differentiable transport solver TORAX, reducing computational time to convergence.

Publication: J.T. Wai, E. Kolemen, "GSPD: An algorithm for time-dependent tokamak equilibria design", Arxiv Preprint. https://arxiv.org/abs/2306.13163

Presenters

  • Josiah T Wai

    Commonwealth Fusion Systems

Authors

  • Josiah T Wai

    Commonwealth Fusion Systems

  • Devon J Battaglia

    Commonwealth Fusion Systems

  • Christoph Hasse

    Commonwealth Fusion Systems

  • Anna A Teplukhina

    Commonwealth Fusion Systems

  • Francesco Carpanese

    Neural Concept

  • Jonathan Citrin

    Google DeepMind

  • Federico Felici

    Google DeepMind

  • Antoine Merle

    EPFL, Ecole Polytechnique Federale de Lausanne

  • Cassandre Contre

    EPFL

  • Olivier Sauter

    EPFL, SPC-EPFL, Ecole Polytechnique Federale de Lausanne

  • Egemen Kolemen

    Princeton University