Session: CIE–31: Graduate Student Poster Symposium

Paper Number: 142254

142254 - Deep Generative Model-Based Inverse Design of Broadband Acoustic Metamaterials via Latent Space Exploration 

This study introduces a ventilated acoustic resonator (VAR), a type of acoustic metamaterials (AM) that enables simultaneous noise reduction and ventilation. However, the inverse design of VAR with analytical methods is a challenging task due to the high degree of nonlinearity. Furthermore, existing deep learning-based inverse design methods for AMs mainly rely on parameter-based approaches, which limits the design flexibility, or pixel image-based methods which deteriorate functional shape, the shape essential for attenuation performance. To address these challenges, we propose a latent space exploration strategy based on genetic algorithm (GA) optimization for the inverse design of broadband VAR under target peak frequency. A conditional variational autoencoder (CVAE) was employed to generate dimension-reduced latent space encoding semantic information on geometric features of VAR and perform GA optimization on the latent space to explore non-parametric VAR structures with broadband sound attenuation performance. As a result, atypical VAR having a functional shape but cannot be defined by design parameters could be confirmed and these non-parametric VARs demonstrated a considerably wide bandwidth compared to the VAR having the widest bandwidth in the training dataset under target peak frequency, which expands the limit of the sound attenuation performance of the VAR. Our proposed method presents a novel approach to the optimization of atypical mechanical structures with complex mechanisms.

Presenting Author: Min Woo Cho Pusan national university

Presenting Author Biography: Min Woo Cho is currently pursuing a M.S. under the supervision of Prof. Sang Min Park, at Pusan National University, South Korea. He started his academic studies in the field of mechanical engineering and received a B.S. degree from Pusan National University, South Korea in 2023. His research interest lie in deep learning-based design and optimization methods for complex mechanical structures and their engineering applications.

Authors:

Min Woo Cho Pusan national university
Kyungjun Song Pusan national university
Sang Min Park Pusan national university