Reverse Design of Absorption Performance for Typical Underwater Acoustic Coatings Based on Neural Network

Abstract

This paper presents a method for rapidly reverse designing the absorption performance of acoustic coatings, utilizing the principles of a concatenated deep neural network. It enables the swift acquisition of effective input parameters. By cascading a reverse neural network with pre-trained forward neural networks, a concatenated neural network is obtained. This network maps the absorption spectrum response to structural and material parameters, thereby resolving the nonuniqueness issue in traditional reverse design. The paper describes the detailed process of reverse designing the absorption performance of acoustic coatings and validates the correctness of the reverse design using finite element methods. A comparative analysis investigates the impact of different loss functions on result accuracy. The findings demonstrate that the proposed modified loss function algorithm significantly enhances precision compared to traditional direct reverse design. This advancement allows for the customization of acoustic coatings with specific acoustic properties, providing technical groundwork for vibration and noise reduction in underwater vehicles.