Sound absorption metamaterials designed by machine learning and their applications

Abstract

Acoustic metamaterials are artificial materials composed of acoustic functional elements along the spatial order, which have extraordinary acoustic properties that are not possessed by homogeneous materials. Since the conception of phononic crystal in the 1990s, acoustic metamaterials have gradually matured, which prospects industrial application. In our report, we present an acoustic metamaterial design method based on deep learning for sound absorption. The design method is firstly set up with the array of Fabry-Perot resonators, then adapted for the array of Helmholtz resonators. The method uses a tandem inverse design model based on the convolution neural network(CNN), which is capable of on-demand design. It can generate acoustic metamaterial design of high absorption in targeted frequency range. Three designs of acoustic metamaterials for different purposes are presented. Firstly, sound absorption metamaterials with selective bandwidth absorption, are used to control tonal noise from electric transformers. Secondly, metamaterials with broadband absorption are used in vehicle cabin noise control. Lastly, full-band near-perfect absorption materials are created to construct acoustic anechoic chambers.