Predicting Pavement Roughness Using Deep Learning Algorithms

Abstract

Transportation agencies should measure pavement performance to appropriately strategize road preservation, maintenance, and rehabilitation activities. The international roughness index (IRI), which is a means to quantify pavement roughness, is a primary performance indicator. Many attempts have been made to correlate pavement roughness to other pavement performance parameters. Most existing correlations, however, are based on traditional statistical regression, which requires a hypothesis for the data. In this study, a novel approach was developed to predict asphalt concrete (AC) pavement IRI, utilizing datasets extracted from the Long-Term Pavement Performance (LTPP) database. IRI prediction is categorized by two models: (i) IRI progression over the pavement’s service life without maintenance/rehabilitation and (ii) the drop in IRI after maintenance. The first model utilizes the recurrent neural network algorithm, which deals with time-series data. Therefore, historical traffic data, environmental information, and distress (rutting, fatigue cracking, and transverse cracking) measurements were extracted from the LTPP database. A long short-term memory network was used to solve the vanishing gradient problem. Finally, an optimal model was achieved by setting the sequence length to 2 years. The second model utilizes an artificial neural network algorithm to correlate the impacting factors to the IRI value after maintenance. The impacting factors include maintenance activities; initial (new construction), milled, and overlaid AC thicknesses; as well as IRI value before maintenance activities. Combining the two models allows for the prediction of IRI values over AC pavement’s service life.