Work

Sound absorption of resonant composite metastructure based on machine learning reverse-assisted design

Year: 2026

Type: article

Abstract: To realize machine learning reverse-assisted design for resonant sound-absorbing structures, this study focuses on resonant composite metastructures, theoretically deriving the sound-absorption coefficients within the range of 10,000 Hz and conducting an in-depth discussion on the sound-absorption mechanism. Leveraging Latin hypercube sampling, a database of 100,000 randomly generated first peaks was constructed for machine learning purposes. Two approaches, a one-stage deep neural network and a two-stage deep neural network, were employed to inversely predict the structural dimension parameters of eight resonant sound-absorbing structures for any given 24 sets of desired first peaks. The results indicated that the two-stage deep neural network, incorporating a forward neural network, achieved relative errors of only 4.2% and 2.25% in predicting the frequency values and sound-absorption coefficients of the first peaks, respectively. The accuracy of machine learning predictions was validated by testing two samples, forecasted by a two-stage deep neural network, through acoustic impedance tube experiments, with errors less than 1.5% compared to the expected targets. This finding offers significant guidance for the implementation of machine learning reverse-assisted design for resonant sound-absorbing structures.

Source: Applied Acoustics

DOI: 10.1016/j.apacoust.2026.111242