An artificial-intelligence based approach for predicting structural damages of paved-road systems under superloads

When studies on mechanistic-based pavement analysis and design are actively conducted, highly-accurate and meaningful data related to structural damages caused by traffic loads have accumulated. Although widely-used pavement design programs such as the American Association of State Highway and Transportation Officials (AASHTO)Ware Pavement Mechanistic-Empirical (M-E) Design software supports M-E pavement design for comprehensive pavement structures and traffic loadings, it is still unable to predict extrapolated mechanistic responses when pavements are subjected to superloads having non-standardized loading configurations not yet included in the software. In this study, artificial neural-network (ANN)-based surrogate models were developed and optimized to provide high accuracy in predicting critical pavement responses related to representative structural damages of jointed plain-concrete pavements (JPCPs) and flexible pavements when subjected to a single pass of various superload types, thereby extensively broadening the scope of constrained mappings in terms of loading variables. Sensitivity analysis on pavement structural and loading variables was performed using the ANN models developed in this study to identify the significance level of each explanatory variable in generating target-pavement responses.