Dynamic pore water pressure in asphalt mixtures under steady state vibration: Response and machine-learning-assisted prediction

The pore water pressure response of asphalt mixtures under steady-state vibration is crucial for describing systematically dynamic behavior of asphalt pavements. However, this dynamic response characteristic is not clear, and the related traditional prediction model of dynamic pore water pressure is subjective and complex. This study aims to clarify the response characteristics of the dynamic pore water pressure in asphalt mixtures under steady-state vibration and establish a nonparametric prediction model based on the self-developed measurement system and three Machine Learning (ML) algorithms. The result indicates that the pore water pressure increases with higher loading pressure, frequency, and saturation degree, with fully saturated specimens showing the strongest response. SMA-13 specimens have greater pressure peaks and valleys than AC-13 due to better void connectivity, affecting water transmission. The dynamic response will be affected only when the confining pressure is less than the pulse pressure. The peak and valley values of pore water pressure increase linearly with time, and the variation of peak values is more obvious. All established ML models demonstrate strong prediction performance, Gradient Boosting Machine (GBM) model achieves the highest accuracy, minimal prediction error, and superior generalization ability when the regression tree depth is 6, and the minimum sample number in the leaf nodes is 4. This study describes comprehensively dynamic behavior of asphalt pavement from a new aspect and provides a new perspective for predicting the pore water pressure of asphalt mixture within the context of data-driven science.