Use of machine learning for a helium line intensity ratio method in Magnum-PSI. (October 2022)

• Record Type: Journal Article
• Authors: Kajita, Shin; Iwai, Sho; Tanaka, Hirohiko; Nishijima, Daisuke; Fujii, Keisuke; van der Meiden, Hennie; Ohno, Noriyasu
• Is Part Of: Nuclear materials and energy. Volume 33(2022)
• Publication Date: 2022-10
• Subjects: Plasma -- Helium -- Optical emission spectroscopy -- Neural network -- Machine learning; Nuclear energy -- Periodicals; Nuclear fuels -- Periodicals; Nuclear reactors -- Materials -- Periodicals; Radioactive substances -- Periodicals; 621.4833
• DOI: 10.1016/j.nme.2022.101281
• Languages: English
• ISSNs: 2352-1791

Abstract

Abstract: Optical emission spectroscopy (OES) of helium (He) line intensities has been used to measure the electron density, n e, and temperature, T e, in various plasma devices. In this study, a neural network with five hidden layers is introduced to model the relation between the OES data and n e / T e from laser Thomson scattering in the linear plasma device Magnum-PSI and compared to multiple regression analysis. It is shown that the neural network reduces the residual errors of prediction values ( n e and T e ) less than half those of the multiple regression analysis in the ranges of 2 × 10 18 < n e < 8 × 1 0 20 m −3 and 0 . 1 < T e < 4 eV . We checked two different data splitting methods for training and validation data, i.e., with and without considering the unit of discharge. A comparison of the splitting methods suggests that the residual error will decrease to ∼ 10% even for a new discharge data when accumulating a sufficient data set. Highlights: The relation between the helium line emission and n e / T e from laser Thomson scattering in the linear plasma device Magnum-PSI are modeled. A neural network (NN) with five hidden layers is introduced to model and compared to the multiple regression (MR) analysis. The NN reduces the residual errors of prediction values less than half those of MR. The residual error can be decreased to ∼ 10% by accumulating sufficient data set.