A machine learning approach to self-consistent RBS data analysis and combined uncertainty evaluation

Abstract

We introduce a machine learning approach designed for the self-consistent analysis of data acquired through simultaneous Rutherford backscattering spectrometry (RBS) in multiple scattering geometries, with subsequent determination of the analysis uncertainty. Using a simulated data set, the successful self-consistent evaluation of up to six simultaneous RBS data collections was achieved by employing artificial neural networks (ANN). The results demonstrate enhanced accuracy and precision when concatenating multiple RBS spectra as input to the ANN. The precision was quantified through the combined uncertainty, encompassing the ANN random uncertainty, the ANN systematic uncertainty, and the model robustness. While applied to only RBS analysis, this approach holds promise for advancing the analysis of data collected by next-generation RBS equipment and Total ion beam analysis, offering a robust and efficient framework for future research.