Prediction of negative hydrogen ion density in permanent magnet-based helicon ion source (HELEN) using deep learning techniques

Abstract

In the present work, a deep-learning model is developed for a permanent magnet-based helicon plasma source. Non-invasive cavity ring-down spectroscopy (CRDS) characterizes the HELEN ion source as a negative hydrogen ion source. This paper discusses different deep learning techniques for modelling the ion source and subsequently predicts the ion source density. Experiments were conducted measuring the plasma density for different ranges of hydrogen gas pressure, magnetic field and RF power. Consequently, experimental data trains the deep learning model. The performance of various deep learning models has been assessed by the root mean squared error and the coefficient of determination values. The deep learning techniques also develop a correlation between the electron temperature and plasma densities. It reasonably mimics the behaviour of the HELEN ion source and can classify the helicon plasma generation at a high RF power range (800-850 W). Also, the influence of other input parameters such as gas pressure and the magnetic field is assessed using the correlation matrix.