Stochastic modeling of porous sound absorbing materials using homogenization method and machine learning

Abstract

This research introduces a novel approach to investigate porous sound absorbing materials using stochastic modeling techniques. The study employs the homogenization method and machine learning to predict the acoustic properties of porous materials. The homogenization method simplifies the analysis of microstructural properties in porous materials, while machine learning is a data analysis technique that utilizes algorithms to learn from data and make predictions. By integrating these methods, this study quantitatively assesses the impact of microscopic structural randomness on sound absorption properties. The study assumes that microstructural parameters are random variables and calculates the sound absorption coefficient through homogenization method. This dataset serves as input for machine learning. A Gaussian process regression model is developed using the dataset to evaluate the distribution of the sound absorption coefficient via the Gauss-Hermite quadrature method. The study implements Bayesian optimization, utilizing the variance of the resulting probability distribution as the acquisition function, to adaptively update the dataset. By iteratively performing these processes, the probability distribution of the sound absorption coefficient can be predicted efficiently and with high accuracy.