Prediction of rutting deterioration in flexible pavements using artificial neural network and genetic algorithm

Rutting is one of the major deteriorations of asphalt pavement significantly impact on road safety and service quality. Forecasting models are necessary to prevent and control the damage caused by this deterioration in the pavement management system. In this study, using the artificial neural network algorithm, models have been developed to predict the amount of rutting deterioration using the long-term pavement performance (LTPP) database. These models have been developed for wet freeze, dry freeze, and dry no freeze climates. Since proper accuracy and simplicity are the most important features of a prediction model, using the NSGA ІІ-MLP multi-objective optimization method, the more important variables in predicting rutting deterioration are identified and selected as the model input. Then, using traffic, climatic and structural variables selected from the genetic algorithm, rutting deterioration prediction models were developed. The coefficient of determination and the mean squared error for the model made in the wet freeze zone and the common model of dry freeze and dry no freeze zones are equal to 0. 96, 2.05, 0.94 and 3.45, respectively. Also, by performing sensitivity analysis, the effect of input data of each model on rutting deterioration was determined. Among the variables that have the greatest impact on rutting deterioration are the cumulative maximum and minimum daily temperature difference per year, pavement age, asphalt layer thickness, annual equivalent single axle loads, and bitumen penetration.