Development of Fatigue Cracking Prediction Models Using Long-Term Pavement Performance Database

Abstract

This study strives to develop improved fatigue cracking models using the long-term pavement performance database. The prediction accuracy of the existing models was found to be inadequate. Several modern regression techniques including generalized linear model and generalized additive model along with the assumption of Poisson distribution and quasi-likelihood estimation method were adopted for the modeling process. After many trials in eliminating insignificant and inappropriate parameters, the resulting model included several variables such as yearly KESALs), pavement age, annual precipitation, annual temperature, critical tensile strain under the asphalt-concerete surface layer, and freeze-thaw cycle for the prediction of fatigue cracking. The proposed model appeared to have substantial improvements over the existing models although their further enhancements are possible and recommended.