| Abstract: | Traffic noise emission has long been a pervasive environmental and ecological problem, especially in the metropolitan cities with large-scale traffic network and high population density. Low noise road surface (LNRS) has been actively developed and applied as an effective measure to maintain the quieter environment of mobility service system. However, when LNRS is applied for noise abatement, the relationship between the acoustic performance and degradation of pavement has not been fully understood yet. To this end, this study aims to model the acoustic longevity of asphalt pavement as a function of the thickness, binder content, maximum aggregate size, and air void content of the pavement surface, as well as vehicle speed based on the long-term tyre-road noise data collected from 270 asphalt pavement sections in Hong Kong. Two machine learning techniques, namely artificial neural networks (ANN) and support vector machines (SVM), were employed and compared. It was found that both ANN and SVM could successfully model the pavement acoustic performance with acceptable model performance metrics. A case study showed that the ANN model was more aligned with the aging mechanisms of porous road surface, but the SVM model showed better training performance. The predicted acoustic deterioration rates of the porous surface case varied from −0.1 to 0.28 dB(A)/month rather than keeping a constant linear increasing trend, depending on pavement ageing periods and vehicle speed levels. The two-dimension sensitivity analysis (2D-SA) revealed the relative importance of pavement age and vehicle speed in controlling the acoustic performance. |
