Investigating impacts of asphalt mixture properties on pavement performance using LTPP data through random forests

Numerous laboratory studies have demonstrated that the properties of hot-mix asphalt (HMA) play a crucial role in the performance of the HMA. However, few studies have directly correlated HMA properties to field pavement performance. The present study addressed this issue with data obtained from the database of the Long-Term Pavement Performance (LTPP) program through the relative importance score inherent in the random forests. The data from 78 sections with a relatively complete record of mixture properties were used to develop the random forests models and determine the relative importance of exploratory variables. A total of seventeen variables representing the physical properties of the HMA and the pavement were incorporated, including the aggregate gradation, mixture volumetric parameters such as bulk specific gravity and air voids, asphalt binder properties such as viscosity and content, the section service age, and the thickness of the pavement. The pavement performance considered was represented by three types of cracking (alligator cracking, wheel-path longitudinal cracking, and transverse cracking), the rutting, and the roughness (IRI). The results showed that the gradation of aggregates has close association with the alligator cracking; the viscosity and stiffness of asphalt binder were strongly correlated with the longitudinal cracking; and the density and indirect tensile (IDT) strength of the mixture significantly affected the transverse cracking. The air voids and binder stiffness played a critical role in the rutting performance. The percentage passing the No. 200 sieve was found a determinant of the evolution of the IRI. (C) 2019 Elsevier Ltd. All rights reserved.