Application of machine learning technology to optimize the structure of two-dimensional metamaterials based on the spectrum of natural frequencies.

A technique based on a neural network was developed, which has shown a high efficiency of searching for the best structure of meta-atoms for modular design of acoustic metamaterials on the model system. A neural network with a three-layer structure required less than 1% of computational time in comparison with a direct numerical analysis of all possible variants of metamaterials in order to find a metamaterial with the required acoustic properties. The calculations were carried out using finite element method and CAE Fidesys software. [ABSTRACT FROM AUTHOR]