Design of Acoustic Metamaterials Using Conditional Wasserstein Generative Adversarial Networks
Proceedings of the International Congress on Acoustics
In this talk, we present a method for the reduction of the total scattering cross section (TSCS) for a planar configuration of cylinders through multi-scattering, generative modeling and deep learning. We first summarize and illustrate existing approaches in the inverse design of metamaterials then discuss current algorithmic limitations and open challenges to preview possible future developments in metamaterial design. We then show a novel method [1, 2] to design broadband acoustic metamaterial configurations using deep learning and generative modeling targeting low TSCS responses. We combine a conditional Wasserstein Generative Adversarial Network (cWGAN) with a standard convolutional neural networks (CNN) to generate metacluster configurations with a specified TSCS. The proposed cWGAN model is enhanced by adding to it a coordinate convolution layer . Particularly placed scattering elements can achieve minimal TSCS producing a broadband cloaking effect at discrete sets of wavenumbers. The method is demonstrated giving examples for planar uniform configurations of rigid cylinders.
Acoustic Metamaterials, Convolutional Neural Networks, Deep Learning, Generative Modeling, Wasserstein Generative Adversarial Networks
Feruza A. Amirkulova, Peter Lai, and Diego Soto. "Design of Acoustic Metamaterials Using Conditional Wasserstein Generative Adversarial Networks" Proceedings of the International Congress on Acoustics (2022).