Abstract

The Federal Aviation Administration (FAA) initiated the Extended Airport Pavement Life (EAPL) program to evaluate and enhance the long-term performance of airport pavements, aiming to extend their service life beyond the 20-year current standard considered in pavement thickness design. The FAA collected extensive pavement performance data - including surface groove geometry, longitudinal profile roughness, and surface distresses - from 22 major U.S. airports. The data included both flexible and rigid pavements. Data analysis shows that pavements designed to fail structurally in 20 years often remain structurally intact but exhibit functional failure sooner than the intended design life. It was found that with effective routine and preventive maintenance, pavements can remain serviceable well past their original design life. To quantify this extended serviceability, the FAA introduced the serviceability level (SL) index, a combined measure of structural integrity and functional condition that indicates a pavement’s suitability for aircraft operations. Supplementary data from each airport was gathered by the FAA to support the analysis, including material characterizations, pavement cores, maintenance histories, runway usage, and weather data. All data have been consolidated in a dedicated database, PA40.

This study documents the analysis of the PA40 data to calculate key pavement condition indexes contributing to overall serviceability and to explore their relationships with factors like weather, traffic, material properties, and pavement structure. Machine learning (ML) models were developed to predict the SL index for both flexible and rigid airport pavements. The SL model is structured as a classification ML task, aiming to classify the runway pavement as either serviceable or unserviceable, based on a set of pavement condition indexes including pavement condition (structural and non-structural), runway roughness, and groove condition. In addition, individual ML models were created to predict each of these indexes as part of the SL model, incorporating predictors such as pavement age, environmental factors, and traffic.

The ML models were developed in Python and integrated into a Visual Basic (VB) library in anticipation of future use in standard FAA pavement design and management programs, including FAARFIELD.