Deep learning-based inverse design of micro-perforated plate and Fibonacci-cavity coupling structure for low-frequency broadband sound absorption

Abstract

In this work, a low-frequency broadband acoustic metamaterial absorber is designed based on a micro-perforated plate and a Fibonacci-cavity coupling structure, where the Fibonacci sequence arrangement of cavities facilitates the coupling of multiple resonance peaks and enables high-efficiency sound absorption. To effectively couple different units, a deep learning method is used to automatically optimize structural parameters. By constructing a multilayer perceptron, a highly accurate correlation between structural parameters and the sound absorption coefficient, with over 97 % accuracy, has been established. The results show that a well-trained forward neural network can effectively replace the intricate physical processes involved. We propose a method using a tandem neural network to achieve the inverse design of an acoustic absorber, which enables efficient and fast directional design compared to traditional empirical approaches and optimization algorithms. The achieved sound absorption efficiency is 81.8 % within the low-frequency spectrum of 300–800 Hz using only two resonance units, and the feasibility of the method is experimentally verified. The deep learning design method proposed in this study can be applied to various types of metamaterials, thereby significantly improving design efficiency.