Abstract
As JET is developing and testing operational scenarios for higher fusion performance, an increase in pulse disruptivity is being observed. On a deeper analysis, we find that several radiative phenomena play an active role in determining the outcome of the pulse. The analysis is enabled by the use of real-time tomography based on the bolometer diagnostic. Even though plasma tomography is an inverse problem, we use machine learning to train a forward model that provides the radiation profile directly, based on a single matrix multiplication step. This model is used to investigate radiative phenomena including sawtooth crashes, ELMs and MARFE, and their relationship to the radiated power in different regions of interest. In particular, we use real-time tomography to monitor the core region, and to throw an alarm whenever core radiation exceeds a certain threshold. Our results suggest that this measure alone can anticipate a significant fraction of disruptions in the JET baseline scenario.
Deep Learning for the Analysis of Disruption Precursors Based on Plasma Tomography
