Speakers
Description
Magnetic confinement is widely recognized as one of the most promising strategies for achieving controlled thermonuclear fusion on Earth. In toroidal devices, the precise regulation of plasma equilibrium, including plasma current, position and shape, is essential to guarantee safe, stable and high-performance operation. In this framework, RFX-mod2 (Reversed Field eXperiment), the upgraded configuration of RFX-mod, represents a highly flexible experimental platform capable of operating in both reversed field pinch and tokamak configurations. The ongoing hardware refurbishment provides an opportunity for redesigning the magnetic control system. This work presents an integrated framework for plasma equilibrium control in tokamak operation of RFX-mod2, combining electromagnetic modeling, constrained optimal control, and real-time boundary reconstruction. A linear control-oriented plasma model is first computed using the CREATE-L code, explicitly accounting for the complex circuital interconnections between the poloidal field (PF) coils and the power supplies. A major challenge arises from the one-quadrant a.c.–d.c. converters driving the PF coils, which impose strict positivity constraints on the commanded voltages and significantly restrict the admissible control actions. To address these issues, the well-known Model Predictive Control (MPC) approach has been formulated for simultaneous plasma current and shape control. Unlike conventional two-layer architectures, the proposed MPC provides a unified optimization-based framework that explicitly embeds the electromagnetic model and actuator limitations within the control law. In particular, the controller incorporates shape control objectives by directly regulating plasma boundary descriptors through coordinated voltage references. This enables accurate tracking of desired equilibrium configurations while rigorously satisfying voltage constraints. Extensive simulation studies based on the RFX-mod2 model, including plasmaless discharge scenarios derived from experimental data of the previous RFX-mod device, demonstrate equilibrium control, effective constraint handling and robust multivariable performance. Furthermore, the computational feasibility of the proposed controller is assessed within the MARTe2 real-time framework envisaged for RFX-mod2 operations, confirming its suitability for practical implementation. To complete the control architecture, a physics-informed neural network approach for real-time plasma boundary reconstruction is also integrated in the control loop. The CARONTE algorithm leverages an Extreme Learning Machine to solve the homogeneous Grad–Shafranov equation using magnetic sensor measurements, enabling real-time training and therefore adapting the network to the evolving plasma equilibrium. Compared with other Neural Network-base reconstruction approaches, the CARONTE eliminates the need for extensive offline training datasets.
Overall, the integration of electromagnetic modeling, optimal constrained predictive control with explicit shape regulation and physics-informed data-driven reconstruction provides a coherent and computationally viable framework for advanced plasma equilibrium management in magnetic confinement experiments in the RFX-mod2 device.