Development of ANN-GA program for backcalculation of pavement moduli under FWD testing with viscoelastic and nonlinear parameters

Abstract

In this study, an Artificial Neural Network (ANN)-based backcalculating program combined with a Genetic Algorithm (GA) optimization algorithm was developed for backcalculation of flexible pavement layer moduli from Falling Weight Deflectometer (FWD) test. Axisymmetric finite element (FE) models were developed considering dynamic loading of FWD drops and viscoelastic and nonlinear material parameters of pavement layers. The FE models were used to generate the synthetic database that covers variations in material parameters, pavement structures, temperatures, and loading levels. The ANN-GA program was trained and verified using the synthetic database. The accuracy of backcalculation was evaluated with measured data from Long-Term Pavement Performance field testing sections. The ANN-GA program was found having acceptable accuracy through the verification and validation processes. The input variables of the ANN-GA program are available from FWD test including the deflections at different offsets, shape indicators of hysteresis loop (force-displacement curve), layer thicknesses, loading magnitudes, and air and surface temperatures. The ANN-GA possesses advantages over traditional iteration-based backcalculating program such as the elimination of seed moduli and consideration of complex material properties. More importantly, the backcalculated pavement layer parameters can be directly used for Mechanistic-Empirical design of pavement overlays.