Neural Network Prediction of Active and Passive Imaging Neutral Particle Analyzer Signals in the DIII-D Tokamak

First results are reported on the prediction of the ‘active’ and ‘passive’ images of imaging neutral particle analyzers(INPA) [1], based on neural networks trained on the database of DIII-D tokamak INPA data. ‘Active’ signal refers to the INPA-collected fast neutrals from charge-exchange reactions of confined fast ions and active high-energy beam neutrals. ‘Passive’ signal refers to the reactions with background cold neutrals near the plasma boundary. Passive images interrogate the phase space of energy and pitch at a fixed radius, but active images interrogate the phase space of energy and radii at a nearly fixed pitch. Patterns of the passive images are reproducible as they are determined by fast ion orbits, related to the specific neutral beams [2]. The neural network may accurately separate passive and active signals and reliably predict their images, where active images are often distorted by Alfvénic instabilities [3]. This is crucial to understand fast ion transport across the phase space [3]. The model also helps estimate edge cold neutrals via passive images.

[1] X.D. Du et al., Nucl. Fusion 58, 082006 (2018).

[2] D. Lin et al., Nucl. Fusion 60, 112008 (2020).

[3] X.D. Du et al., Phys. Rev. Lett. 127, 235002 (2021).