Inverse Design Method for Helmholtz Resonant Acoustic Absorbers Based on Physics-Informed Neural Networks
Year: 2026
Type: conference-paper
Abstract: Efficient inverse design of low-frequency acoustic absorption structures is important for underwater acoustic applications. Traditional methods often rely on repeated forward calculations or purely data-driven trial-and-error processes, which are time-consuming and lack physical reliability. To overcome this problem, this paper proposes an inverse design framework for Helmholtz resonant acoustic absorber by integrating parametric modeling, an analytical acoustic model, an autoencoder, and a physics-informed neural network (PINN). A dataset of structural parameters and absorption coefficient curves is first constructed under geometric constraints. Then, an autoencoder is used to extract low-dimensional latent features from high-dimensional absorption curves. Based on these features, the PINN establishes the inverse mapping from target absorption curves to structural parameters by embedding the analytical acoustic model into the training process. Results show that the proposed method achieves accurate curve reconstruction and outperforms a conventional multilayer perceptron (MLP). Compared with MLP, the proposed PINN reduces the geometric parameter MAE, RMSE, and relative error by 26.28%, 19.60%, and 17.70%, respectively, while the sound absorption curve MAE, RMSE, and average maximum error are further reduced by 14.04%, 6.32%, and 36.18%, respectively. Finite element simulations further verify the feasibility of the predicted structures.
Authors: Wei Zhang, Shikai Jing
Institution: Beijing Institute of Technology
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204183655