Machine Learning‐Based Analysis and Prediction of the Influence of Meso‐Void Characteristics on Sound Absorption Coefficient of Porous Asphalt Concrete

Résumé fourni par la source

ABSTRACT Porous asphalt concrete (PAC) is widely recognized as an effective noise‐reducing pavement material due to its interconnected void structure. However, the quantitative relationship between meso‐scale void characteristics and the sound absorption coefficient is not well understood, which limits the precision of mixture design for optimal acoustic performance. This study establishes a comprehensive database of PAC specimens, featuring eight meso‐void characteristic parameters: computed void ratio, reconstructed void ratio, average equivalent diameter, average void volume, average void surface area, number of voids, total void surface area, and total void volume. A systematic machine learning framework was developed, integrating correlation analysis, multicollinearity diagnosis, stepwise regression, grey relational analysis, and multi‐model comparison, to quantify the influence of void features on the average sound absorption coefficient (500–1600 Hz) and to create a high‐precision prediction model. Five models, single‐factor linear regression, multiple linear regression, random forest (RF), support vector regression (SVR), and Extra Trees (ET), were trained on 139 samples and evaluated against an independent validation set of 22 samples. The results indicate that the computed void ratio and reconstructed void ratio are the dominant factors, with Pearson correlation coefficients of 0.934 and 0.923, respectively. The ET model demonstrated the best predictive performance, achieving an R 2 of 0.9964 on the independent validation set, significantly outperforming SVR and multiple linear regression. Feature importance analysis revealed that the two void‐ratio parameters accounted for 76.08% of the total ET feature importance. Sensitivity analysis confirmed that a ±10% perturbation in the computed void ratio induced the largest fluctuation in the predicted sound absorption coefficient. The proposed methodology and optimal ET model provide a reliable tool for PAC mixture design aimed at achieving targeted noise‐reduction performance.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.