Analysis of hydrogen isotopes retention in thermonuclear reactors with LIBS supported by machine learning
Year: 2022
Type: article
Abstract: Retention of hydrogen isotopes in thermonuclear reactors is a safety and maintenance issue, thus requires a contactless measurement method of its level. Although calibration free LIBS proved its usefulness on calibrated samples and even in tokamaks, its performance may be deteriorated in next-step fusion devices because of its sensitivity on the plasma parameters as electron temperature and density which are more challenging to be precisely measured than in laboratory conditions. Moreover, both Beryllium and Tungsten have spectral lines in the vicinity of hydrogen alpha line, which may introduce additional uncertainties. Because of these problems the uncertainty of the temperature estimation at the level of several percent leads to serious uncertainties in estimation of elemental composition, even higher than + − 20 at.%. To deal with problems with machine learning (ML) algorithms are proposed. To obtain satisfactory amount of data for training the models, simulated spectra may be used. The simulation allows obtaining synthetic spectra for different concentration of W, Be and D/T for a range of electron temperature and density. At the present stage of the research the data are used for development of classification and regression ML algorithms as logistic and linear regression, SVM, LDA, ANN, which are supported or not with dimensionality reduction by PCA. The results show differences in performance of various models and outline further works on the development of systems operating on real spectroscopic data.
Authors: P. Gąsior, W. Gromelski, M. Kastek, A. Kwaśnik
Institution: Institute of Plasma Physics and Laser Microfusion
Language: English
Subfield: Spectroscopy
Field: Chemistry
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,100
OpenAlex ID: W4309531270