Numerical Modeling and Artificial Neural Network for Predicting J-Integral of Top-Down Cracking in Asphalt Pavement

Abstract

Top-down cracking (TDC) is recognized as one of the major distress modes in asphalt pavements. This study aimed to determine the fracture parameter J-integral of TDC, which is a critical input to predict the crack growth rate and fatigue life of pavements for this type of distress. Previous research studies demonstrated that TDC is affected by various factors, including the complex state of high tensile or shear stresses induced by the loading at the edge of or within the tire and material properties such as the modulus gradient in the asphalt layer, moduli of the base and subgrade layers, and pavement structures. In this study, the finite element model (FEM) was adopted to simulate the propagation of TDC by considering combinations of these essential factors and to calculate the J-integral for 194,400 cases. It was shown that the modulus gradient plays an important role in determining the J-integral, and the J-integral is not uniformly distributed within the pavement depth. On the basis of the database generated from the FEM, six backpropagation artificial neural network (ANN) models—including one input layer, two hidden layers, and one output layer—were developed by using the same input variables and output variable as those for the FEM. The R2 value for each ANN model was greater than .99, which indicates the goodness of fit. After the parameters of each ANN model have been determined, the J-integral can be predicted for any combination of the design parameters without reconstruction of the FEM.