Machine learning and microstructure design of polymer nanocomposites for energy storage application

Abstract

Film dielectric capacitors have been widely used in high-power electronic equipment. The design of microstructure and the choice of fillers play an important role in nanocomposites' energy storage density. Machine learning methods can classify and summarise the limited data and then explore the promising composite structure. In this work, a dataset has been established, which contained a large amount of data on the maximum energy storage density of nanocomposites. Though using processed visual image information to express the internal information of composite, the prediction accuracy of the prediction models built by three machine learning algorithms increase from 84.1% to 91.9%, 80.9% to 68.9%, 70.6% to 81.6%, respectively. By calculating the branch weight in the random forest prediction model, the influence degree of different descriptors on the energy storage performance of nanocomposites is analysed. A total of 10 groups of composites with different structure and filler amount were prepared in the laboratory, which were used to verify the reliability of prediction models. Finally, the effective filler's structure is explored by three prediction models and some suggestions for the interface design of filler are given.