Prediction of the Fatigue Life of Asphalt Mixtures using Artificial Neural Networks

Prediction of the fatigue life of asphaltic concrete is required for design and management of pavements. The prediction of the fatigue life of asphaltic concrete using Artificial Neural Network has been investigated in this research. Sufficient experimental data containing fatigue test conducted on asphaltic concrete over a wide range of conditions was needed for modeling. Therefore, reviewing the literature, the results of the fatigue tests on Kansas State asphalt mixtures were found to be appropriate for this research. The parameters of asphalt viscosity, strain level, stiffness of mixture, asphalt content, air voids content and gradation were selected as input variables, while the fatigue life was set as the only output variable. Progressive multilayer perceptron Artificial Neural Network modeling has been used for prediction of the fatigue life, with back propagation training algorithm and numerical optimization technique of Levenberg-Marquardt. Modeling was performed using the neural network tool box and programming in MATLAB, and the results have been compared. It is shown that the programming, can better predict the fatigue life than directly utilizing the tool box. It is also found that Artificial Neural Network can predict the fatigue life of the asphaltic concrete than the regression equation.