Inverse Design and Optimization of Vibroacoustic Responses of Beam Structures Using Tandem Neural Networks

Abstract This article presents a study of the inverse design of vibroacoustic responses of beam structures for vibration and noise control. The aim is to develop an efficient method for designing structural shapes that achieve desired vibroacoustic behaviors. To this end, we propose a tandem neural network architecture capable of directly mapping desired vibroacoustic response to the optimal geometry of non-uniform beams. Unlike traditional approaches, our method enables rapid design by leveraging tandem neural networks. We explicitly incorporate physical constraints relevant to shape optimization into the loss function of the tandem neural network. This ensures that the generated designs are not only computationally feasible but also physically realizable and practical for engineering applications. The proposed method is validated through several case studies, demonstrating its ability to generate shapes with precise tuning of natural frequencies, suppression of vibrations, or realization of specific vibroacoustic phenomena such as acoustic black hole-like responses. This study provides valuable insights for the development of innovative solutions to complex vibroacoustic design problems.