AI-based design technique for developing highly broadband passive acoustic metamaterials for target frequencies
S. Akram, Asif Israr
Abstract
This study presents an AI-based design technique for developing passive acoustic metamaterials for desired frequencies with enhanced acoustic performance. The aim is to enhance sound transmission loss and broaden bandwidth at the target frequency by optimizing the design within a limited thickness while ensuring ease of fabrication. A raw hybrid design based on Helmholtz and space-coiled configurations is identified through finite element analysis and literature. The proposed design is an interconnected network of bent necks and cavities, providing a range of acoustic properties depending on the relative geometric parameters. This study explores the design space for desired acoustic characteristics, which is often limited by computational costs and restricted creative input, by employing NSGA-II alongside a machine learning model trained on data from FEA of design variations based on the raw hybrid model. For a fixed thickness, the design parameters of the Acoustic Metamaterial can be predicted on the Pareto front, achieving optimal sound transmission loss and bandwidth for the target frequency. The results are validated through an impedance tube based on ASTM E2611 with 3D printed samples of designs. The methodology can help produce practically viable low frequency AMMs that are light weight, thin, and easy to fabricate for commercial applications.