Hits:

Indexed by:期刊论文

Date of Publication:2019-06-01

Journal:JOURNAL OF TRANSPORTATION ENGINEERING PART B-PAVEMENTS

Included Journals:SCIE、EI

Volume:145

Issue:2

ISSN No.:2573-5438

Key Words:Mechanistic-empirical pavement design guide (MEPDG); Alligator cracking (AC); Gradient boost machine; Machine learning; Extreme boosting machine (XGBoost)

Abstract:This study developed a gradient boosted model (GBM) to enhance the fatigue cracking predictive performance of transfer functions in the mechanistic-empirical pavement design guide (MEPDG). Two transfer functions, respectively, for the alligator cracking (AC) and longitudinal cracking (LC), were considered. The extreme boosting machine (XGBoost) package in R programming language based on the GBM algorithm was employed to develop the model. The data collected from a report of the National Cooperative Highway Research Program (NCHRP) Project 01-37A were used for training the GBM, which are the same data originally used to establish the national transfer functions of the MEPDG. The inputs included damage indices (DI) computed by the MEDPG software, pavement thickness, materials related parameters such as asphalt mixture gradation and resilient modulus of subgrade, climatic conditions, and annual average daily truck traffic (AADTT). The experiment used 93 out of 461 and 81 out of 414 observations as the testing sets for the AC and LC, respectively. The results indicated that the predictive performance of the presented GBM significantly outperformed that of the national transfer functions. For the AC, the testing R2 between measured and predicted values increased from 0.104 to 0.671, whereas it rose from 0.0455 to 0.784 for the LC. Compared with the corresponding transfer functions in MEPDG, the precision of the GBM was also improved, in which the standard errors decreased from 6.2% to 4.35% for the AC and from 1,242.25ft/mi to 52.11ft/mi for the LC.