Performance prediction and inverse design of cylindrical plate-type acoustic metamaterials based on deep learning

Abstract

Acoustic metamaterials are artificial structures that possess distinctive acoustic characteristics, allowing for modulation effects that are challenging to achieve in the natural world. Nevertheless, the design of acoustic metamaterials is a challenging process due to the intricate relationship between their structural parameters and nonlinear performance. In view of the limitations of conventional design methodologies, which rely on the a priori knowledge of experts and are often hindered by prolonged computation times and the necessity for iterative trials to achieve design objectives, this paper introduces a deep learning-based method for performance prediction and inverse design of Cylindrical Plate-type Acoustic Metamaterials (CPAMs). The creation of a dataset is initiated by generating a large number of samples using a parametric model, with bandgap characteristics calculated through the finite element method. A forward-design deep learning model is then developed, predicting the upper and lower bandgap limits based on input structural parameters. Additionally, an inverse design model is constructed, enabling rapid generation of structural parameters based on desired acoustic performance. The results of the inverse design are validated through simulation and experimentation, confirming the accuracy and reliability of the model. This study demonstrates the potential of deep learning in efficiently designing complex acoustic metamaterials, offering a promising solution for CPAMs development.