Acoustic metamaterial design framework using deep learning and generative modeling

Abstract

This talk presents our findings and research performed on the application of deep learning and generative modeling in acoustic metamaterial design. Specifically, we will discuss our research findings published in recent papers on the application of deep learning algorithms, generative neural networks, reinforcement learning models, and global optimization for the inverse design of 2-D and 3-D acoustic metamaterial structures. The examples will be shown for the implementation of neural networks models for the inverse design of 3-D pentamode structures resulting in a low shear modulus, high bulk modulus, and an impedance matched with water. The generative 2-D GLO-Nets and the reinforcement learning models producing broadband low scattering effects for 2-D planar configurations of scatterers under plane wave incidence will be presented. The current challenges encountered during the application of deep learning methods in scaled metamaterial design will be discussed.