Neural Network–Based Multiple-Slab Response Models for Top-Down Cracking Mode in Airfield Pavement Design

Abstract

The Federal Aviation Administration (FAA) has recognized for some time that its current rigid pavement design model, involving a single slab loaded at one edge by a single aircraft gear, is inadequate with respect to top-down cracking. Thus, one of the major observed failure modes for rigid pavements is poorly accounted for in the FAA Rigid and Flexible Iterative Elastic Layer Design (FAARFIELD) design software. A research version of the FAARFIELD design software has been developed in which the single-slab three-dimensional finite-element (3D-FE) response model is replaced by a four-slab 3D-FE model with initial temperature curling to produce reasonable thickness designs accounting for top-down cracking behavior. However, the long and unpredictable run times associated with the four-slab model and curled slabs make routine design with this model impractical. Artificial intelligence (AI)-based alternatives such as artificial neural networks (ANNs) have great potential to produce accurate stress predictions in a fraction of the time. ANNs could be practical replacements for a full 3D-FE computation that requires long computation times. In the development of ANN models, both individual input parameters and dimensional analysis have been considered, and accuracy of predictions from both methods was compared. ANN models for only mechanical and simultaneous mechanical and thermal loading cases were developed using individual input parameters and dimensional analysis. It was observed that very high accuracies were achieved in predicting pavement responses for all cases investigated.