Structure Predictor: A Machine Learning Algorithm to Reliably Fit Thin Film Neutron Reflectivity Curves

Overview

Neutron reflectometry is a technique for studying the structure of thin films. Thin films are layers of materials that simulate systems with only the processes a scientist is interested in. Interactions between materials, such as new layer growth on a batt ery or the strength of a protective coating, can be studied at the nanometer scale with a technique called neutron reflectometry. Data gathered from neutron reflectometry can be used to plot a neutron reflectivity (NR) curve from which the number of layers and the corresponding thickness, roughness, and scattering length density of each layer can be extracted and evaluated to understand how the materials interact. Currently, NR curves are manually fit with multiple parameters, which is time consuming. Machi ne learning algorithms offer classification capabilities that may be more time efficient. A Python package, Structure Predictor, was created using k -Nearest Neighbors and convolutional neural network algorithms which provides three methods for predicting the number of layers and parameter values of a NR curve. The accuracies of the three methods range from 66.6% to 73.5% indicating Structure Predictor to be a reliable supplement, or even alternative, to manual NR curve fitting.

Competition history

• JSHS 2024 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.