Machine Learning Augmented Computational Reverse-Engineering Analysis for Scattering Experiments of Assembled Mixtures of Nanoparticles

In the field of materials science and engineering, structural characterization is a critical step needed to link the materials’ structure to its macroscopic properties. Small angle scattering (SAS) is a useful technique to characterize structure at multiple length scales. The output of SAS experiments is the averaged scattering intensity at various wave vectors, <i>I</i>(<i>q</i>) vs. <i>q,</i> whose interpretation often relies on fitting with analytical models. The selection of analytical models can be a limitation for analyzing the scattering profile if and when appropriate analytical models exist for the material’s structure of interest. In this talk, we will present a computational method we have developed called Computational Reverse Engineering Analysis of Scattering Experiments (CREASE) and show its application for analysis of spherical nanoparticle mixtures’ assembled structure. We test the strengths and limitations of CREASE by using a variety of <i>in silico</i> <i>I</i>(<i>q</i>) obtained from simulations of binary nanoparticle assemblies and nanoparticle solutions with varying mixture composition/concentration, nanoparticle size distribution, and degree of mixing/aggregation. We will also highlight the machine learning (ML) model used in CREASE that links features of the nanoparticle solutions (e.g., concentration, nanoparticles’ tendency to aggregate) to computed scattering profile; this ML model is applicable to different nanoparticle sizes without the need for additional data to retrain the model to be specific to the size of interest. Finally, we show how the nanoparticle structure reconstructed from scattering using CREASE can serve as input to optical modeling and achieve a computed reflectance spectrum that matches ones from the experimental systems.