ABSTRACT
In this study Artificial Neural Networks (ANNs) have been used to predict the critical responses calculated by a NIKE3D-FAA Finite Element (FE) tool, a fully implicit three-dimensional finite element model representing a significant modification of the NIKE3D program originally developed by the Lawrence Livermore National Laboratory (LLNL) of the U.S. Department of Energy. A synthetic database has been constructed by implementing a great deal of analysis of a concrete pavement structure using NIKE3D-FAA for a reasonable range of inputs. ANNs have been trained to produce the critical responses at the top and bottom of such slabs. These responses are tensile stresses with in-plane directions, shear stress, and deflection. The concrete pavement studied in this paper is a four-layer structure pavement subjected to mechanical loading by a Boeing heavy airplane (B777-300ER) and the loading effects of a temperature gradient in the slab. The ANN models were found to successfully predict critical pavement responses in all cases, with a minimum of 0.99 for the coefficient of determination (R2). The observed results exhibits how ANN models can both accurately predict critical responses used for designing and account for failures. Because the ANN models can accomplish rapid prediction of responses it is suggested that they have significant potential for producing accurate stress predictions in a fraction of the time needed to perform full 3D-FE computation.
