Malm, Nils Oscar Wilhelm
(2023)

*Predicting the antenna properties of helicon plasma thrusters using machine learning techniques.*
[Laurea magistrale], Università di Bologna, Corso di Studio in

Aerospace engineering [LM-DM270] - Forli', Documento full-text non disponibile

Il full-text non è disponibile per scelta dell'autore.
(

Contatta l'autore)

## Abstract

When designing helicon plasma thrusters, the properties of the radio frequency antenna that is used to deposit power into the plasma are paramount to the performance of the thruster. To calculate these properties a program called Adamant can be used. Adamant is able to simulate the currents developing on the antenna and these can then be used to calculate the antenna impedance, an important parameter for calculating the power deposition and for the design of the matching network between the generator and the antenna. Since running Adamant simulations takes a long time and requires a non-negligible amount of computer resources this project has attempted to find a machine learning model, trained on data generated by Adamant, that can be used instead of Adamant for future small design evaluations. The performance target was to have less than 5\% error when evaluating them point-to-point. Six different machine learning models from MATLAB were implemented and evaluated: decision trees, ensembles, support vector machines, Gaussian process regressions, generalized additive models and artificial neural networks. The model hyperparameters were optimised using Bayesian optimisation and evaluated using both nested k-fold cross-validation and other metrics. The result is that the artificial neural network performed the best when taking both error magnitudes and generalisation ability into account, with a maximum error of 3.98\% and a mean of 0.53\%, while the GPR came close second. Future work should include more training data, and could possibly expand the machine learning model's capabilities by also looking at different antenna shapes.

Abstract

When designing helicon plasma thrusters, the properties of the radio frequency antenna that is used to deposit power into the plasma are paramount to the performance of the thruster. To calculate these properties a program called Adamant can be used. Adamant is able to simulate the currents developing on the antenna and these can then be used to calculate the antenna impedance, an important parameter for calculating the power deposition and for the design of the matching network between the generator and the antenna. Since running Adamant simulations takes a long time and requires a non-negligible amount of computer resources this project has attempted to find a machine learning model, trained on data generated by Adamant, that can be used instead of Adamant for future small design evaluations. The performance target was to have less than 5\% error when evaluating them point-to-point. Six different machine learning models from MATLAB were implemented and evaluated: decision trees, ensembles, support vector machines, Gaussian process regressions, generalized additive models and artificial neural networks. The model hyperparameters were optimised using Bayesian optimisation and evaluated using both nested k-fold cross-validation and other metrics. The result is that the artificial neural network performed the best when taking both error magnitudes and generalisation ability into account, with a maximum error of 3.98\% and a mean of 0.53\%, while the GPR came close second. Future work should include more training data, and could possibly expand the machine learning model's capabilities by also looking at different antenna shapes.

Tipologia del documento

Tesi di laurea
(Laurea magistrale)

Autore della tesi

Malm, Nils Oscar Wilhelm

Relatore della tesi

Scuola

Corso di studio

Indirizzo

CURRICULUM SPACE

Ordinamento Cds

DM270

Parole chiave

Machine Learning, Helicon plasma thruster, Adamant, cross validation, Bayesian hyperparameter optimisation

Data di discussione della Tesi

13 Luglio 2023

URI

## Altri metadati

Tipologia del documento

Tesi di laurea
(NON SPECIFICATO)

Autore della tesi

Malm, Nils Oscar Wilhelm

Relatore della tesi

Scuola

Corso di studio

Indirizzo

CURRICULUM SPACE

Ordinamento Cds

DM270

Parole chiave

Machine Learning, Helicon plasma thruster, Adamant, cross validation, Bayesian hyperparameter optimisation

Data di discussione della Tesi

13 Luglio 2023

URI

Gestione del documento: