Determination of pavement internal condition from a non-destructive field test is a persistent topic for its practical necessity and difficulty. It is essentially an inverse problem calibrating pavement material and structural properties from pavement responses. Considering the intrinsic complexity of asphalt pavement materials (e.g., time and temperature dependencies of asphalt mixture and stress dependency of unbound granular materials), this problem has become a typical high-dimensional optimization problem with a large and diverse set of calibrated parameters. This study investigated the feasibility of artificial intelligence-based finite element model updating in addressing this problem, and focused on the accuracy as well as stability of the backcalculated results. For a comprehensive evaluation of this method, the effects of its components such as the surrogate model representing the pavement system, the applied pavement response, the optimization algorithm and the backcalculation scheme were characterized. Finally, we found that the sensitivity of applied pavement responses to thebackcalculated pavement condition, the number of applied pavement responses and the balance between the backcalculated pavement condition and the applied test were of significant importance to achieving accurate and stable backcalculation results. Corresponding modifications were recommended to be conducted in future research for improving the performance of the proposed backcalculation method.
This article is part of the theme issue ‘Artificial intelligence in failure analysis of transportation infrastructure and materials’.