| description abstract | Asphalt concrete stripping is a critical distress that occurs underneath the pavement surface due to the accumulation of moisture with no visual indication at the surface. Past studies have suggested the feasibility of using nondestructive methods, such as a ground penetrating radar and portable seismic property analyzer, to identify stripping-affected pavements. However, these devices are yet to reach widespread implementation potential due to some limitations in their operation and measurement system. In the present study, the deflection measurements of a traffic speed deflection device (TSDD), namely, the Rolling Wheel Deflectometer (RWD), were evaluated to identify pavement sections that may suffer from stripping and moisture damage. Statistical and artificial neural network (ANN) models, which used RWD measured deflections, pavement characteristics, and performance data as inputs, were developed to predict the probability of stripping damage in the tested sections. A logistic model that considers the linearity between the predictor and response variables was developed and validated with reasonable accuracy in estimating stripping probability. The nonlinear relationship between the input and output variables was simulated using a generalized additive model (GAM) and was found to have improved stripping prediction accuracy. To further address the complex relationship between the predictor and response variable, an ANN-based pattern recognition system was developed; this approach was more accurate than the previous models and is recommended as a cross-validation tool for the results of the logistic and GAM models. A regression-based classification tree was developed based on the available dataset that is easy to interpret and is convenient for highway agencies for preliminary stripping evaluation. The analyses and methods presented in the study may be used for stripping damage detection and as an additional benefit of TSDD testing. | |
