Accelerated Discovery of Optimal Ion Transport Characteristics in Nanoparticle-Based Electrolytes Using Convolutional Neural Networks

Spatial arrangement of spherical nanoparticles in nanocomposite materials can significantly influence the macroscopic behavior. However, iterative probing of all the possible nanoparticle configurations for their corresponding macroscopic properties to identify the optimal configurations is often intractable even using computer simulations. To overcome such challenges, in this work, we highlight the capability of Convolutional Neural Networks (CNNs) to serve as machine learning-based surrogate models to establish quantitative structure−property linkage in composites with monodisperse spherical particles. This is specifically demonstrated using a CNN model to quantitatively link the diffusivity of ions to the spatial arrangement of the nanoparticles in nanoparticle-based electrolytes, and its success in identifying configurations exhibiting optimal diffusivities on combining with a metaheuristic topology optimization algorithm. We also discuss the use of data-driven approaches such as Principal Component Analysis to elucidate the correlations between the simple physical descriptors of the microstructure topology and the resulting property, thus providing a physical rationale for the observed optimal configurations.