Intelligent Prediction of Asphalt Concrete Air Voids during Service Life Using Cubist and GBRT-Ensemble Learning Approaches Hybridized with an Equilibrium Optimizer Algorithm

Abstract

There are four critical factors that affect air voids (VA) of asphalt concrete over time: traffic loads and repetitions, environmental regimes, compaction, and asphalt mix composition. Because of the high as-construct VA content of the material, it is expected that voids will reduce over time, causing rutting during initial traffic periods. Eventually, the material will undergo shear flow when it reaches its most dense state with optimum aggregate interlock or refusal VA content. Furthermore, to accurately model the performance of an asphalt mixture, the VA must be predicted over time. This study aims to implement two ensemble learning methods namely Cubist and gradient boosting regression tree (GBRT) hybridized with equilibrium optimizer algorithm (EOA) to predict the VA percentage of asphalt concrete during the service life of flexible pavements. For this purpose, 324 data records of VA were collected from the literature. The variables selected as inputs were original as-constructed air voids, VAorig (%); mean annual air temperature, MAAT (°F); original viscosity at 77°F (25°C) , ηorig,77 (Mega-Poises); and time (months). GBRT-EOA and Cubist-EOA were found to be superior to other classical single ML approaches [for instance, artificial neural network (ANN), support vector machine (SVM), and Gaussian process regression (GPR) and M5Tree] and existing nonlinear regression models. In the training phase, the GBRT-EOA had R2, root mean square error (RMSE), mean absolute error (MAE), and mean of absolute percent error (MAPE) values of 99.8%, 0.136%, 0.093%, and 1.324%, respectively, while these values changed to 95.1%, 0.701%, 0.53%, and 7.825% in the testing phase. Also, only less than 5% of the records were predicted using this model with more than 20% deviation from the observed values. As determined by the sensitivity analysis, time (months) is the most significant of the four input variables, while MAAT (°F) is the least one. A parametric study showed that regardless of the MAAT, the ηorig,77 of 1.0 Mega-Poises, and the VAorig above 6% can be ideal for improving the pavement service life in terms of the VA of the asphalt concrete layer during service life.