International Roughness Index Prediction of Flexible Pavements Using Neural Networks

Abstract

International roughness index (IRI) is a widely-accepted parameter that indicates pavement performance and ride quality. This study develops a prediction model for IRI using artificial neural networks (ANN) for flexible pavements located in wet-freeze, dry-freeze, wet no-freeze and dry no-freeze climate zones. The long-term pavement performance (LTPP) database is used for obtaining climate and traffic data. Annual average temperature, freezing index, maximum humidity, minimum humidity, precipitation, average daily traffic, and average daily truck traffic are considered as input parameters for predicting IRI. The proposed ANN model is trained with 50% of the available climate and traffic data and the remaining 50% of the data are used for testing the model. The comparison of LTPP recorded data and ANN predicted data is validated by calculating root mean square error (RMSE). The 7-9-9-1 ANN model with a hyperbolic tangent sigmoid transfer function generated the lowest RMSE of 0.01. The 7-9-9-1 ANN model is further tuned for robustness and consistency with several synthetic data sets and 70%, 15%, and 15% of the synthetic data sets are used to train, test, and validate, respectively, the ANN model. The ANN model predicts the IRI with reasonable accuracy and the lowest RMSE 0.027 in measured.