Computationally efficient surrogate response models for mechanistic-empirical pavement analysis and design

This paper proposes the use of neural network- (NN-) based pavement structural analysis tools as surrogates for the flexible pavement response analysis in the new mechanistic empirical pavement design guide (MEPDG) developed for the American State Highway and Transportation Officials (AASHTO). Some of the recent successful applications of NN-based structural analysis models for predicting critical flexible pavement responses and nonlinear pavement layer moduli from falling weight deflectometer (FWD) deflection basins are highlighted. Because NNs excel at mapping in higher-order spaces, such models can go beyond the existing univariate relationships between pavement structural responses and performance (such as the subgrade strain criteria for considering flexible pavement rutting). The NN-based rapid prediction models could easily be incorporated into the newer versions of the MEPDG, which will continue to be updated. This can lead to better performance prediction and also reduce the risk of premature pavement failure