Sound localization via deep learning, generative modeling, and global optimization

Abstract

An acoustic lens is capable of focusing incident plane waves at the focal point. This talk will discuss a new approach to design metamaterials with a focusing effect using effective and innovative methods by using machine learning. Specifically, the physics simulations use multiple scattering theory and machine learning techniques such as deep learning and generative modeling. The 2-D-Global Optimization Networks (2-D-GLOnets) model [1] developed initially for acoustic cloak design is adapted and generalized to design and optimize the acoustic lens. We supply the absolute pressure amplitude and gradient into the deep learning algorithms to discover the optimal scatterer positions that maximize the absolute pressure at the focal point in the confined region. We examine and evaluate the performance of the generative network, searching for an optimal configuration of scatterers over a range of parameters to produce desired features, such as broadband focusing and localization effects. The model will show examples of planar configurations of cylindrical structures producing focusing effect. [1] L. Zhuo and F. Amirkulova, “Design of acoustic cloak using generative modeling and gradient-based optimization,” in INTER-NOISE and NOISE-CON Congress and Conference Proceedings (Institute of Noise Control Engineering, 2021), Vol. 3.