Bayesian Analysis of Heterogeneity in Modeling of Pavement Fatigue Cracking

Abstract

Fatigue cracking models are important for the design, analysis, and management of pavement structures. There are two types of cracking models available in the literature. The first type of model is structured to predict time to failure (given a specific failure criterion, e.g., 25% of the total area shows cracking). The other type of model is intended to directly predict the cracking area as a function of a set of covariates. For the first type of model, previous researchers have suggested a probabilistic approach such as survival modeling (or duration modeling) to take various uncertainties into consideration. However, to the authors’ knowledge, no current study considers the effect of heterogeneity when applying survival models to pavement fatigue cracking. In the context of deterioration modeling, heterogeneity can be defined as the performance difference across different individuals. Ignoring unobserved heterogeneity may lead to biased and inconsistent parameter estimates. The current paper proposes applying the frailty model to address this problem by adding a random term to the hazard function. The proposed model is demonstrated using the long-term pavement performance (LTPP) database.