Breakdown prediction of ion source at EAST-NBI using neural network

Abstract

Abstract The designed neural networks are trained to appraise the risk of ion sources breakdown events in the neutral beam injector (NBI) experimental device using several offline diagnostic signals as inputs. A saliency analysis proves the reasonableness of the chosen inputs, some of which are helpful to improve the network performance. The experimental tests that were carried out refer to data collected from successfully-terminated and breakdown-terminated shots performed during three years of experimental advanced superconducting tokamak NBI experiments, from 2016 to 2019. Results show that it is very possible to develop a predictor base on neural network that intervenes well in advance to avoid ion sources breakdown or mitigate its effects during the beam extraction in the high-power and long-pulse mode.