Data Analysis Procedures for Long-Term Pavement Performance Prediction

Abstract

The results of 3 years of research aimed at investigating data analysis methods used in the development of pavement performance relationships are reported. The research was part of the U.K. collaborative program linked to the U.S. Strategic Highway Research Program (SHRP), in particular the Long-Term Pavement Performance (LTPP) experiment. The development of pavement performance models usually concludes with the application of regression techniques to determine coefficients for model parameters. It is important to identify the model forms and the engineering or mechanistic principles to be used in the data analyses in the initial stages and then to censor any obvious anomalies in the data. This was applied to data on pavement rutting measured by the Transport Research Laboratory over a 20-year period in the United Kingdom. Engineering knowledge of rutting progression suggests a cubic model form, with the quadratic component representing typical performance in early pavement life. An attempt was made to derive a rutting model that took into account material properties, layer thickness, and aggregate types. The pavement structural number concept was applied as a proxy for pavement strength for the different pavement structures used in the test sites. The results of the analyses confirmed that material properties, layer thickness, and their combined effects influence rutting, but in ways that vary greatly. No simple model form was found to adequately predict rutting for a variety of pavement types, even with general categorical model forms.