Pavement Roughness Modeling Using Regression And Ann Methods For LTPP Western Region

Abstract

This study aims to develop pavement roughness models using multiple linear regression equation and artificial neural network (ANN) modeling approaches. The model database uses International Roughness Index (IRI) data included in a national database for the Western region. Datasets for asphalt pavement with bound base at 32 different locations are considered in the analysis. The variables included are IRI, pavement age, design structural number, equivalent single axle load (ESAL), and also a dummy variable for construction number. The LTPP data was used to compare predicted IRI values using the improved linear regression equation with R of 0.573, with different types of ANN models. The ANN models considered are static ANN, feedback ANN, and dynamic ANN. The verifications of the improved dummy regression equation predictions for 39 data points showed only -2.9% difference compared to the mean IRI. Comparisons between the mean predicted IRI values showed that the enhanced dummy regression equation gave a better prediction compared to the ANN model with a minimum error of 37.9%. Out of three ANN models, the feedback ANN provided a better prediction compared to the static and dynamic ANNs. Further analysis showed that the predictions using training all data sets are more accurate with R ranged from 0.740 to 0.827. It is important to consider ESAL and construction number for an accurate and reliable future IRI prediction.