ABSTRACT
Objective decisions related to the management of road networks are based on important measures of pavement performance. The International Roughness Index (IRI) is a critical indicator of pavement performance, and it is considered the standard for pavement roughness. A reliable pavement performance prediction model is needed to predict future pavement conditions and identify maintenance and rehabilitation (M&R) needs. This study intends to develop pavement roughness models using the Artificial Neural Networks (ANNs) approach for composite (asphalt overlay on concrete) pavements using the Long-Term Performance Pavement (LTPP) program database for the Wet-Freeze climate region. A total of 186 pavement sections with 1,930 data points were analyzed. Five models were developed using different independent variables (i.e., Initial IRIRight, Age, Seasons, Asphalt Thickness, Concrete Thickness, Subbase Thickness, Subbase Type, Construction Number (CN), Cumulative Equivalent Single Axle Load (CESAL), Air Temperature, Freeze Index, Freeze-Thaw, and Precipitation) and one dependent variable (i.e., IRIRight). The best-performing model was selected based on the lowest average square error (ASE), lowest mean absolute relative error (MARE), and highest coefficient of determination (R²). Results showed that the developed models had satisfactory results with a good fit of observed and predicted data. Therefore, local and state agencies can use the developed ANN roughness models as a tool for better condition assessment and effective M&R scheduling. Furthermore, the use of available climatological and historical traffic data to predict IRI changes will also eliminate time-consuming data collection and processing, accordingly, reducing costs.
