Models Relating Pavement Quality Measures

Abstract

A large field-measured international roughness index (IRI) data set was collected in the summer and fall of 2001 on 650 highway kilometers with sections containing full-depth hot-mixed asphalt (HMA) and HMA over portland cement; the set was used to investigate the relationships between IRI, rutting, and pavement cracking. Parametric statistical analysis, linear regression, and neural network models were used with a large database of 65,530 observations (10-m intervals). The goal was to investigate the relationship between both rutting and cracking (estimated from photolog images obtained at the same time field data were collected) and IRI. Specifically, the objective was to determine if these relationships were consistent enough to allow IRI, the more easily measured quantity, to be used as a surrogate for the others or if the effort and resources to collect all three measures were necessary in pavement management programs. The results indicate that while statistically significant relationships exist between IRI and both cracking and rutting, these relationships are not strong enough for IRI to be used as a surrogate measure for pavement condition. The weak predictive relationships are further evidenced by the lack of success in training a neural network that could significantly outperform linear regression models. It was concluded that IRI, while appropriate for measuring rideability, was not appropriate for measuring cracking or rutting.