Acoustic structure inverse design and optimization using deep learning

Abstract

From ancient to modern times, acoustic structures have been employed to manage the spread of acoustic waves. Nevertheless, designing these structures traditionally remains a laborious and computationally intensive iterative process. Recognizing that complex acoustic systems can be effectively analyzed using the lumped-parameter method, we introduce a deep learning model that learns the correlation between the equivalent electrical parameters and the acoustic properties of these structures. As an illustration, we consider the design of multi-order Helmholtz resonators, showing experimentally that our model can predict structures with high precision that closely align with the specified design criteria. Furthermore, our model can seek multiple solutions in conjunction with dimensionality reduction algorithms and support evolutionary algorithms in optimization tasks. Compared to traditional numerical methods, our approach offers greater efficiency, flexibility, and universality. The designed acoustic structures hold broad potential for applications including speech enhancement, sound absorption, and insulation.