Fiorenza, Federico (2026) Advanced magnetic control and reconstruction in fusion devices. [Tesi di dottorato]

Documento PDF fiorenza_federico_38.pdf Download (28MB)

Abstract

The ITER tokamak, currently under construction, is expected to be the first magnetic confinement device capable of producing a net surplus of fusion energy. While awaiting its completion, several satellite projects have been launched to support ITER’s mission. Among them are the EU–Japan joint project JT-60SA—presently the world’s largest operating fusion device—and the Divertor Tokamak Test (DTT) facility in Italy, which will investigate heat exhaust solutions for next-generation reactors. These satellite tokamaks play a crucial role in developing and validating technologies and control strategies for ITER. In this framework, the initial commissioning phases of tokamak devices pose significant challenges for plasma magnetic control, particularly due to the temporary absence of critical subsystems, such as in-vessel stabilization coils and the full suite of magnetic diagnostics. This thesis tackles the challenges of early-stage tokamak operations while also proposing advanced control and magnetic reconstruction algo- rithms for the effective operation of next-generation tokamaks. In partic- ular, the contributions are the following: • validation of ITER’s magnetic reconstruction tools under reduced sensor availability and the development of a fast boundary recon- struction algorithm; • the design of advanced control strategies for plasma stabilization in early operational scenarios or in the presence of strict actuator constraints. On the magnetic reconstruction side, this work leverages data from the JT-60SA Integrated Commissioning campaign to preliminarily val- idate ITER’s real-time magnetic reconstruction algorithms, testing the algorithms’ reconstruction capability with reduced diagnostics. In fact, ii during the JT-60SA IC, only a limited number of magnetic sensors were available. Furthermore, an alternative fast boundary reconstruction algo- rithm is proposed, based on Extreme Learning Machines (ELMs)—a class of neural networks characterized by fast training and inference times. Its performance is firstly evaluated in the same reduced sensors scenario and then is benchmarked against the XLOC algorithm, that has been used for decades at JET to reconstruct in real-time the plasma boundary, with DTT serving as a case study. On the control side, the thesis provides a twofold contribution. First, it introduces an ITER-like, model-based magnetic control technique tailored for early operation scenarios lacking in-vessel actuators. This architecture enables the use of external superconducting coils for multiple tasks, in- cluding vertical stabilization, and its effectiveness is demonstrated through nonlinear simulations of JT-60SA plasmas. Second, the thesis develops a Model Predictive Control (MPC)–based scheme for the upgraded RFX-mod2 device. Following the removal of the Inconel vessel and the introduction of a conductive cage, a new electro- magnetic model is derived, and consequently an MPC control system is designed. Importantly, the proposed approach explicitly accounts for the hardware constraints on the actuators, such as the use of the already exist- ing one-quadrant converters, thereby ensuring reliable tracking of desired plasma shape and current in RFX-mod2 operations. Together, these developments contribute to the creation of flexible and resilient magnetic reconstruction and control strategies, crucial for the safe and efficient operation of next-generation tokamak devices.

Downloads

Actions (login required)

Modifica documento