A New Backcalculation Procedure Based on Dispersion Analysis of FWD Time-History Deflections and Surface Wave Measurements Using Artificial Neural Networks

Abstract

A new layer moduli backcalculation procedure using transient measurements from both a falling weight deflectometer (FWD) test and a surface wave test is presented in this paper. This algorithm employs numerical solutions of a multi-layered half-space based on Hankel transforms as a forward model and Artificial Neural Networks (ANNs) for the inversion process. Two ANNs are used in series; a depth to a stiff layer is predicted first and used as input for the second ANN to be used for the prediction of layer moduli. Falling weight deflectometer (FWD) and surface wave tests were performed on experimental pavement sections in North Carolina with different asphalt mixture types and surface conditions. Dispersion analysis was performed on FWD transient deflections and surface wave test measurements using the Short Kernel Method. Conventional backcalculation was also performed on FWD deflections using the MODULUS 5.0 program. It was concluded that the dispersion-based backcalculation method is sensitive to changes in upper layer condition and results in less variable sub-surface layer moduli and more accurate prediction of depth to a stiff layer than the conventional backcalculation does using FWD peak deflections.