Airfield pavement condition prediction with machine learning models for life-cycle cost analysis

ABSTRACT

Accurate pavement condition prediction is a vital aspect of pavement management because it informs the timing, budgeting and operational impact of maintenance and repair. This study developed machine learning models for airfield asphalt pavement condition prediction and its application in life-cycle cost analysis (LCCA). The most effective models were found being Random Forest for original pavement and support vector regression for pavement overlays. A modified Recursive Feature Elimination was used for model optimisation, resulting in 20% decrease in error. For the non-overlay pavement, the aircraft group with heavy takeoff weight had distinguishable effect on predicted pavement condition index (PCI). For the overlay pavement, overlay thickness had higher relative importance than either pavement age or condition before the overlay. A longer original pavement life was found more economical, but the overlay life decreased if the pavement condition before overlay is deteriorated past an optimum point (PCI range of 65–80 depending on overlay thickness). LCCA results showed economic trade-offs between pavement condition before overlay and the subsequent overlay life.

KEYWORDS:

• Airfield pavement
• pavement condition index
• machine learning
• life-cycle cost analysis
• pavement overlay

Acknowledgements

The authors acknowledge the support of Federal Aviation Administration (FAA) for providing access to the EAPL database. The request for data need to be made to the FAA.

Disclosure statement

No potential conflict of interest was reported by the author(s).