Use of Artificial Neural Network Model to Predict Rut in Asphalt Concrete Pavements

Pavement rutting negatively impacts vehicular travel by posing a safety hazard while water fills the rut and hydroplaning occurs. The vehicle can run off from the pavement, or in a worst-case scenario, it can turn over due to hydroplaning and cause serious bodily injury or fatal crashes to passengers. The ability to predict pavement rutting through an Artificial Neural Network (ANN) can ensure rutting is fixed before severe damage to vehicles can occur and save lives. Climate and rutting data for 15 roads in the US Midwest region were collected from the Long-Term Pavement Performance (LTPP) database to model an ANN that can accurately predict pavement rutting. The LTPP data includes Annual Average Precipitation, Average Temperature, Average Freeze Index, Average Humidity, Annual Average Daily Truck Traffic (AADTT), 18-Kip ESAL, and Time. These inputs were used to train an ANN to match the measured rutting output data. An optimal ANN model was determined through trial and error based on LTPP data. A model-data ratio of 70% training data, 15% testing, and 15% validation was used to create the model. In one pavement section, the ANN model predicted growth of rut between 2012 and 2020 with R² 0.9817. For all 15 tested pavement sections, the accuracy of the ANN model is R² 0.9974. This ANN model was able to predict rutting in pavement sections with high accuracy, which results in this method being a viable way to predict pavement performance and the need for repairs.