Development of algorithms to support the plasma control in disruptive scenario in DEMO relevant machines

The lessons learned from disruptions in current tokamaks play a crucial role in the EUDEMO Research & Design (R&D) strategies. The data sharing among tokamak experiments is crucial in advancing our understanding and mitigation capabilities of disruptions in future fusion devices. In this context, this thesis aims to support the EU-DEMO R&D activities with inter-machine studies on the several disruption implications. The chapter 3 present a machine learning algorithm application for the real-time automatic tracking of the Multifaceted Asymmetric Radiation From the Edge (MARFE) evolution at AUG, which is a precursor of H-mode density limit disruptions disruption. This study represent the rst step of a crossmachine algorithm, which will consider also JET and WEST data, to be scaled for the MARFE detection in EU-DEMO and ITER. In the chapter 2 an inter-machine database is presented. The database collects EU-DEMO relevant plasma perturbations causing both Vertical Displacement Events, Major Disruption in Single Null and Quasi-Double Null congurations both from JET and AUG. These experimental perturbations, properly scaled to EU-DEMO, are the starting point for the predictive analyses to foreseen the plasma position and the EM loads during VDEs. Disrupted experiments with tungsten (W) accumulation in the plasma both from AUG and JET have been collected in the database to study the eect of W accumulation in the core on the plasma performance and to quantify the mechanisms that determine the W concentration in the plasma. In addition, ux pumping eligible experiments have been collected from hybrid scenarios JET experiments. The hybrid scenario is a good candidate for ITER and EU-DEMO scenarios thanks to its robustness and high performances. The chapter 4 present the procedure conducted to characterize the inverse boundary reconstruction errors due to the white noise eect on in-vessel pick-up coils. Finally, in the conclusions, the results discussed in three chapter are summarized and next steps of the work are presented.