Neural Network–Based Thickness Determination Model to Improve Backcalculation of Layer Moduli without Coring

Abstract

Traditionally, deflection data from a falling weight deflectometer (FWD) test and thickness data from pavement coring are used to backcalculate layer moduli. In this study, instead of coring, a neural network (NN) model is developed to determine layer thickness from FWD time-deflection histories. Using the NN predicted thicknesses, layer moduli are backcalculated using a commercially available backcalculation software. For validation, backcalculated moduli are compared with the laboratory determined moduli. Results show that backcalculated moduli are nearly equal to the laboratory moduli. Thus, the inclusion of NN-based thickness data has the potential to replace pavement coring, which is very expensive, and/or to enable a backcalculation method to run with a reasonable assumption of thickness whenever coring information is not available.