Speaker
Description
Disruptions are one of the major problems for the control of thermonuclear plasmas. The fast termination of the discharge can indeed induce huge heat loads on the plasma-facing components and high electromechanical forces on the structures of the devices. In the commercial reactors, disruptions will have to be completely avoided as much as possible, since even a single disruption could compromise their integrity.
In present Tokamaks, disruptions are unavoidable. Their occurrence is particularly likely in the baseline at low safety factors (around q95=3), the reference scenario in ITER. On DIII-D, the disruptivity at this safety factor is around 60%. On JET, in some high current low q95 campaigns, the disruptivity rate also reached 60%, even for a low radiated fraction. The next generation of devices will have to operate with radiation above 90% of the input plus alpha particle power, and with fully detached divertors, conditions that have been proven experimentally to increase significantly the disruptivity rate.
This contribution reports the deployment of new AI based analysis methods on thousands of JET discharges. The investigated regimes cover the isotopic compositions from hydrogen to full tritium and include the last major D-T campaigns. A new approach to proximity detection permits to determine both the probability of and the time interval remaining before an incoming disruption. The developed techniques combine physics and data-driven methodology, implement adaptive and from scratch learning, and real real-time compatible. The proposed control logic results in no missed alarms, 65% of avoided and prevented disruptions and almost no false alarms or wrong actions. Consequently, the obtained results indicate that physics-based prediction and feedback schemes can be developed, to deploy realistic strategies of disruption avoidance and prevention, meeting the requirements of the next generation of devices.
