Data-driven approach for investigating and predicting rutting depth of asphalt concrete containing reclaimed asphalt pavement

Abstract

The aim of this paper is to investigate and predict the rutting depth of asphalt concrete containing Reclaimed Asphalt Pavement (RAP) content by the data-driven approach with aid of 6 six tree algorithm including Random Forest (RF), Gradient Boosting (GB), Adaptive Boosting (AdaB), Extreme Gradient Boosting (XGB), Light Gradient Boosting (LightGB) and Categorial Gradient Boosting (CatB) using default hyperparameters. A database containing 396 data samples derived from Hamburg Wheel Tracking Test (HWTT) and consisting of 11 input variables such as Corrected Optimum Asphalt Content (COAC), Reclaimed Asphalt Pavement, Nominal Maximum Aggregate Size (NMAS), Flashpoint, m-Value, Mass loss, Specific gravity, Viscosity, Creep stiffness, Temperature, and Number of wheels and 1 output Rutting depth are created. 10 times of 10-Fold CV and four metrics such as R2, RMSE, MAE and MAPE are used to verify the robustness of ML models. XGB model has highest performance with the mean value of R2, RMSE, MAE and MAPE to be equal to 0.9792, 1.0767 mm, 0.7085 mm, 0.1034% respectively. Comparing with other factors like the number of load cycles (number of wheels) and mix design considerations, temperature has a dominant impact on the rutting depth of asphalt pavement (type of binder, NMAS, RAP content). The rutting depth value rises with increasing temperature and COAC. NMAS has a negligible impact on asphalt pavement's resistance to rutting. To increase the rutting resistance when building the asphalt pavement, the chosen asphalt binder has high Viscosity and Flashpoint values. The local interpretation clearly demonstrates each factor's contribution to the rutting depth value in each unique scenario. Engineers can design the rutting resistance of asphalt pavement in advance thanks to the mapping produced by the Partial Dependence Plot 2D under grid contour.