Inverse design of ventilated acoustic resonators using a sound transmission loss-encoded variational autoencoder

Ventilated acoustic resonators (VARs) for simultaneous sound attenuation and ventilation have presented a unique challenge in acoustics. Traditional methods for designing VARs are limited by their reliance on human intuition and extensive computational resources. This study proposes a novel sound transmission loss-encoded variational autoencoder (STL-VAE) for the inverse design of ventilated acoustic resonators (VARs). The STL-VAE model overcomes these limitations by encoding the target sound transmission loss (STL) into a latent space, enabling the generation of VAR designs that achieve broadband sound attenuation. STL-VAE significantly reduces the mean squared error (MSE) between the target STL and the generated VAR designs, outperforming the best designs from the training dataset by over 100fold. The proposed method offers a highly efficient and accurate approach for designing complex acoustic metamaterials with applications in urban and industrial noise mitigation.