Relating aggregate friction properties to asphalt pavement friction loss through laboratory testing, statistical analysis, and machine learning insights
Ahmed S. El-Ashwah, Magdy Abdelrahman
Abstract
This study applied statistical and machine learning approaches to relate aggregate friction properties and shape characteristics to asphalt mixture friction loss, focusing on gap- and dense-graded (DG) mixtures. Aggregate morphology and friction properties were quantified through an aggregate imaging measurement system (AIMS), a dynamic friction tester (DFT), and a British pendulum tester (BPT) over varying Micro-Deval test (MDT) durations. Meanwhile, DFT and circular track metre (CTMeter) assessed the friction and texture properties of asphalt mixtures in conjunction with the three-wheel polishing device (TWPD), simulating traffic-induced polishing. Random Forest Analysis (RFA) emphasized the significance of aggregate terminal friction properties in pavement frictional performance. Loss of aggregate friction properties (%), measured by DFT at 20 km/hr (DFT20), was the most significant material property for evaluating pavement friction loss, accounting for approximately 20% of the total effect, followed by the terminal texture index. Two analytical models were proposed: the first correlated the friction properties of asphalt mixtures to the corresponding aggregate sources using DFT20 values. Meanwhile, the second model integrated mixture gradation and aggregate morphology, achieving an overall coefficient of determination (R2) of 0.93. Finally, the study provides a preliminary approach for screening aggregate quality and mix design to achieve cost-effective, optimal frictional performance.