DEVELOPMENT OF AN ARTIFICIAL NEURAL NETWORK (ANN) FOR BACKCALCULATION OF PARAMETERS DEFINING THE MODULI OF PAVEMENT LAYERS

Abstract

Pavement evaluation is essential to assess the performance, service life and rehabilitation needs of existing pavements. In deflection testing-based evaluation of structural properties of the pavement, accuracy of the predictions depend mostly on the way backcalculation is carried out. Because of limitations of being able to handle only limited and idealistic solutions – such as a limited number of layers and linear elastic analysis – the traditional back calculation approach often does not provide reliable estimates of pavement layer properties. The purpose of this study is to build an Artificial Neural Networks (ANN) model with supervised learning utilizing the data obtained from the finite element-based software that can be used to back calculate the layer elastic properties (modulus or parameters determining the modulus) from information for four deflection basins corresponding to four different loads for a pavement structure with a four-layer system. Overall, the analysis of the estimated ANN model mapping the deflection inputs to the parameters of the layer moduli provide very encouraging results, as it provides very accurate predictions of most parameters. Moreover, the study has opened the door for further research to make the model applicable for even more field representative conditions such as varying pavement structures (i.e., different combinations of layer thicknesses), variability in load levels, and non-linearity of all unbound pavement layers.