Improving Pavements with Long-Term Pavement Performance: Products for Today and Tomorrow. Papers from the 2003-2004 International Contest on Long-Term Pavement Performance Data Analysis

摘要

This paper presents an analysis between the International Roughness Index (IRI) and the standard deviation of longitudinal roughness, as well as a neural network study developed to predict the critical level of roughness. Measured longitudinal profiles available in the Long-Term Pavement Performance (LTPP) program database were used. A total of 207 pavement sections in 42 States of the United States were used to do this analysis. Using suitable software, the International Roughness Index (IRI) and the standard deviation of longitudinal roughness values were computed for every longitudinal pavement profile measured. Afterwards, these values were used in regression analysis and a high correlation was found between them (R2=0.93). Neural network analysis correlated the IRI-computed values with the type of sub grade soil, pavement structure (layer thickness), climate, and traffic data of 157 pavement sections. The neural network could forecast the IRI with an extremely high correlation factor (R2=0.99). Besides, the neural network provided a sensitivity analysis indicating the relative contribution of factors related to the structural number (49 percent), climate (31 percent), and traffic (20 percent). Multivariate linear and nonlinear statistic regressions were also performed to predict IRI, but no correlation was found.