Nonlinear Backcalculations of Inverted Pavements Using Hybrid Artificial Neural Network and Colliding Body Optimization Algorithm

Falling Weight Deflectometer (FWD) is one of the common methods for determining the bearing capacity of pavement. The inverted pavement system developed in 1970 in South Africa. This type of pavement is constructed as a sandwich structure, so that a crashed stone base layer is placed between two layers with high stiffness (asphalt concrete layer and cement treated base layer). The main purpose of this research is developing a method for predicting the moduli of the inverted pavements based on the measured deflection bowl using the FWD. Since, the nonlinear modeling of aggregate base is very important in case of inverted pavements, to stablish a dataset, about 38,000 inverted pavement sections have been analyzed by the non-linear finite element program, MICHPAVE. Then, using a hybrid artificial neural network and the colliding body optimization algorithms (ANN-CBO) method, a procedure was developed for backcalculation of inverted pavements assuming nonlinear behavior of aggregate base. The results of this study indicate that the results of MICHPAVE program are very close to the results of the KENLAYER program as well as field data. This study also showed that the artificial neural network is able to predicted the surface deflection bowl of the inverted pavements accurately (R2> 99.99%). In addition, it was found that the hybrid ANN-CBO has superior accuracy and speed in comparison with the hybrid ANN-GA for nonlinear backcalculation of inverted pavements.