The mechanistic-empirical design guide (MEPDG) recommends two flexible pavement analysis methods to estimate the response of pavement systems under actual wheel loads. Multilayer elastic theory is the simplest method to characterize the behavior of flexible pavements. However, it does not take the nonlinear behavior of unbound materials into account. Alternatively, nonlinear finite element (FE) analyses are recommended for determining pavement responses (i.e., stresses, strains and displacements) more realistically. However, FE analyses require significantly more computation time. To evaluate the benefits of using a nonlinear response and minimize the additional effort to implement the FE method, a practical approach for transforming the results from a linear elastic analysis was developed in this study. A database of typical pavement sections was analyzed to establish the correlation between the results from the FE nonlinear analysis and linear models. Artificial neural network (ANN) technique was then used to estimate adjustment factors that can convert responses estimated from multilayer linear analysis to corresponding nonlinear responses. The inputs to the ANN model are the responses (critical stresses, strains) calculated from the linear elastic analyses and the base and subgrade nonlinear stiffness parameters. The outputs of the ANN model are adjustment factors that can be simply multiplied by the linear-elastic responses the way the calibration factors are implemented in the MEPDG software.
展开▼
