Neural network algorithms for rigid airfield pavement responses

This study evaluates Artificial Neural Network (ANN) training algorithms, architecture, and training data to identify optimum ANN models capable of predicting finite-element solutions of critical pavement responses for concrete pavement under a complex airplane gear load. The best architectures and best training algorithm are determined for 1,000 and 4,789 entry datasets by evaluating prediction accuracy for independent data sets not utilized for model development, with results showing that increasing the size of datasets improves the capability for generalization. It is also concluded that results for different ANN models using different architecture and training algorithms becomes more consistent as database size increases.