ABSTRACT
The United States Federal Aviation Administration (FAA) has initiated research to develop robust methodologies for airfield extended-life pavement design. Pavement life in this context refers to the functional life that is not considered directly in the FAA thickness design models. Under the extended life study, data were gathered for a selection of runways at large- and medium-hub U.S. airports, including construction, performance, material property, traffic, and environmental data. The FAA has proposed a performance index, called Serviceability Level (SL), which is a comprehensive index quantifying the suitability of a pavement for use by aircraft. A previous FAA study used the collected data to develop prediction models for pavement performance indexes for functional conditions including potential to develop foreign object damage (FOD), roughness and loss of friction. That study proposed a framework for development of the SL index based on logistic regression models that incorporated various performance indicators, including pavement condition index (PCI), roughness index and friction index. In the initial SL model, pavement age was considered the only predictor. This paper presents the results of using machine learning to improve the performance index anti-SCI, which is a component of SL index, by taking into account the contribution of environmental variables as predictors. The developed model is based on an autoregressive approach and used random forest as the learning algorithm. This study is an initial step toward enhancement of the SL model.
