Prediction of fast neutron spectra with a spherical TEPC using a machine-learning algorithm

Abstract

For many years, new unfolding methods based on machine learning have been studied and developed to improve the prediction capabilities of fluence spectra in neutron fields. These methods are mainly applied to traditional devices used for specific measurements like Bonner spheres, activation detectors or liquid scintillators. In this paper, we attempt to develop a new method based on the unfolding of the fluence spectrum from the micro-dosimetric spectrum measured by a tissue-equivalent proportional counter (TEPC). This type of counter is commonly used for neutron kerma measurements and quality factor assessments, but has never been employed as a neutron spectrometer. This work focuses on the fast neutron region, which is an extremely relevant subject in various fields of nuclear energy. We have tested different machine-learning models to define an unfolding algorithm that can be used to reconstruct the energy distribution of the fluence for spectra of various origins, ranging between 50 keV and 20 MeV.