Predicting microstructure of a polymer nanocomposite using machine learning

Polymer nanocomposites (PNCs) offer a broad range of thermophysical properties that are linked to their compositions. However, it is challenging to establish a universal composition-property relation of PNCs due to their enormous composition and chemical space. Here we address this problem and develop a new method to model the composition-microstructure relation of a PNC through a machine learning pipeline, named as nanoNET. The inputs to the model are five composition parameters viz., NP size, polymer chain length, NP-polymer interaction strength, NP-NP interaction, and NP concentration of a PNC. The output is radial distribution function (RDF) of NPs in the polymer matrix. We conduct molecular dynamics simulations of a model PNC within a carefully selected range of compositional parameters, which span a wide region of the phase space, including experimentally known phases, and utilize the data to establish and validate the nanoNET. Within this framework, the grayscale images of NPs RDF in a polymer matrix are encoded to a latent space using a convolutional neural network (CNN) autoencoder. Subsequently, a random forest regressor establishes a correlation between the composition of the PNC and the latent space representation of its NPs' RDF. The nanoNET predicts NPs distribution in many unknown PNCs very accurately. This method is very generic and can accelerate the design, discovery, and fundamental understanding of composition-microstructure relations of PNCs and other molecular systems.