AI-Driven Acoustic Metamaterials for Pixel-Accurate Sound Insulation Control

Acoustic metamaterials have emerged as a promising platform for efficient and flexible low-frequency sound insu- lation, overcoming the limitations imposed by the mass law governing conventional materials. While metamaterials achieve extraordinary low-frequency sound insulation via local anti-resonances from membranes of their meta-units, their broadband performance is inherently constrained by the narrow-band nature of resonances. Although tailoring the distribution of attached masses offers a pathway to modulate these membrane modes’ spectral features, the complexity of such configurations renders analytical solutions intractable. Here, we propose a deep learning frame-work that bridges this gap by encoding intricate mass distributions as pixelated images (mass-loaded and mass-free regions) and establishing a direct mapping between these images and the resulting transmission loss (TL) spectra. This approach facilitates inverse design of broadband sound-insulating metamaterials for a target TL spectrum and enables rapid performance prediction for arbitrary mass configurations. By synergizing artificial intelligence with the complicate mode engineering of acoustic metamaterials, our work establishes a data-driven paradigm for advanced wave manipulation, opening avenues for next-generation noise control technologies.