Physics-constrained machine learning for impedance-based measurement and sound absorption prediction of composite acoustic materials.

Journal: Neural networks : the official journal of the International Neural Network Society
Published Date:

Abstract

Accurate prediction of sound absorption performance in composite acoustic packages is critical for acoustic material characterization, noise control engineering, and the rational design of acoustic functional materials. However, establishing a precise mapping between material microstructure and sound absorption behavior remains a fundamental challenge in this field. Here, we propose a physics-informed machine learning framework for predicting the sound absorption performance of composite acoustic materials. An acoustic mechanistic framework incorporating the Johnson-Champoux-Allard (JCA) model and the Transfer Matrix Method (TMM) is first constructed to derive physics-based constraints. These constraints are subsequently embedded into the loss function as a hierarchical priority system, and integrated with machine learning algorithms to develop a predictive model tailored for composite acoustic packages. The proposed method is systematically validated against both the JCA-TMM mechanistic model and multiple machine learning baselines. The results indicate that the proposed method achieves both higher predictive accuracy and stronger generalization capability, with an accuracy of 95.3%. Compared with KNN, the baseline model with the lowest RMSE, the proposed model reduces RMSE by approximately 11.0% and improves prediction accuracy by 3.3 percentage points. These results suggest that the proposed framework can be extended to other composite functional material systems, providing a data-driven solution for efficient acoustic-package design and noise and vibration reduction.

Authors

Keywords

No keywords available for this article.