Litcius/Paper detail

Towards automated analysis for neutron reflectivity

D. I. Mironov, James H. Durant, Rebecca Mackenzie, J. F. K. Cooper

2021Machine Learning Science and Technology24 citationsDOIOpen Access PDF

Abstract

Abstract We describe a neural network-based tool for the automatic estimation of thin film thicknesses and scattering length densities from neutron reflectivity curves. The neural network sits within a data pipeline, that takes raw data from a neutron reflectometer, and outputs data and parameter estimates into a fitting program for end user analysis. Our tool deals with simple cases, predicting the number of layers and layer parameters up to three layers on a bulk substrate. This provides good accuracy in parameter estimation, while covering a large portion of the use case. By automating steps in data analysis that only require semi-expert knowledge, we lower the barrier to on-experiment data analysis, allowing better utility to be made from large scale facility experiments. Transfer learning showed that our tool works for x-ray reflectivity, and all code is freely available on GitHub (neutron-net 2020, available at: https://github.com/xmironov/neutron-net ) (Accessed: 25 June 2020).

Topics & Concepts

Computer sciencePipeline (software)Artificial neural networkNeutronReflectivityData miningOpticsArtificial intelligencePhysicsNuclear physicsProgramming languageNuclear Physics and ApplicationsMachine Learning in Materials ScienceHydrocarbon exploration and reservoir analysis