Joint application of the finite element method and artificial neural network in the identification of parameters of layered pavement

In this paper, an inverse analysis of the mechanical parameters of existing layered pavements was performed. An artificial neural network (ANN) was used to approximate the response of the numerical pavement model to the input parameters. Two methods were presented. In the first method, the ANN directly maps the inverse relationship between the pavement surface deflection and mechanical parameters. In the second method, the approximated model was used in the classical back-calculation of pavements by a procedure of minimising the differences between the model response (now a neural network) and field measurements obtained from the falling weight deflectometer (FWD) test. Only one parameter of each pavement layer, the Young’s modulus, was identified. It was found that the identification is not unambiguous. This means a given pavement deflection can be observed with different sets of layer stiffness moduli. However, the average stiffness of the layers is always identified with high accuracy.