Hits:
Indexed by:Journal Papers
Date of Publication:2019-11-01
Journal:JOURNAL OF MATERIALS IN CIVIL ENGINEERING
Included Journals:SCIE
Volume:31
Issue:11
ISSN No.:0899-1561
Key Words:Mechanistic-Empirical Pavement Design Guide (MEPDG); Asphalt concrete; Modulus; Falling weight deflectometer (FWD); Machine learning; Regularized regression; Long-Term Pavement Performance (LTPP)
Abstract:The modulus of the existing asphalt concrete (AC) layer, back-calculated from nondestructive pavement tests, is a crucial input for an accurate overlay design in the pavement mechanistic-empirical (ME) design system. However, nondestructive testing (NDT) data for this purpose are not always available for network-level rehabilitation analyses. To address this issue, this paper proposes a regularized regression method to accurately estimate the moduli with data readily available from pavement management systems, including distress, structural information, and climatic conditions. The data from the Long-Term Pavement Performance (LTPP) database were used for model training. Prediction performance comparisons among three regularization regression methods (ridge, elastic net, and lasso) and the ordinary least-squares regression were conducted. The results showed that the elastic net regression outperformed the other three methods in terms of predictability and interpretability. The mean squared errors of the regularization regression methods were found to be considerably lower than that of the ordinary least-squares regression. The moduli estimated by the regularization methods were very close to the back-calculated ones from the LTPP database, which demonstrated the feasibility of estimating the moduli of existing pavement when in paucity of NDT data. After applying the estimated moduli in the pavement ME design system, the predicted alligator cracking was closer to the measured data than those without these data.