Rational Design of Polymer Composites With Desired Dielectric Properties by Random Forest Classification

Data-driven approaches have been widely used in various fields, and there is a growing interest in using machine learning (ML) to discover potential materials to save trial-and-error time and cost. In this work, an ML framework based on random forest (RF) classification method and limited data is built to explore the relationship between dielectric properties and composites microstructure, and composites with desired dielectric properties are designed rationally. Through feeding physical, geometric, and distributional parameters of nanofillers as fingerprints into the RF algorithm, the linkage between the dielectric properties, including breakdown strength, permittivity, energy density, and the features, is established. The resulting ML model is used to predict the dielectric properties of 100 000 of randomly generated candidate composites, and some suggestions for microstructure design of high energy density composites are presented. The results indicate that the tradeoff between permittivity and breakdown strength can be achieved through the interplay of physical and distributional characteristics, which provides a route to effectively enhance the energy density, and the coordination between nanofiller and matrix is also crucial to increase the energy density of composites, among which a significant contrast in the bandgap between the filler and matrix is more preferred.