Comparison of feed-forward, cascade-forward, and Elman algorithms models for determination of the elastic modulus of pavement layers

This research presents the ability of three different types of artificial neural networks (ANN) algorithms to predict the elastic modulus of flexible asphalt pavements subject to dynamic traffic load utilizing a falling weight deflectometer (FWD). (The feed-forward, cascade-forward, and Elman) back-propagation network types have been developed with different numbers of neurons in hidden layers to define the optimal ANN model using the Matlab software. The developed ANN models were used to predict the elastic modulus values for 30 new pavement sections that were not used in the process of training, validation, or testing to ensure its suitability and the efficiency of each model. The best model from each algorithm type of developed models was chosen by using evaluation metrics and their results were compared with each other and the real obtained data to determine the most successful model. There were very minor differences in the results between the expected and the actual data. The results indicated that, among these algorithms, the feed-forward model has a better performance compared to the other two ANN types. The research confirms the possibility of ANN models to predict pavement layers elastic modulus with high speed and accuracy for using it in Pavement performance evaluation.