Reliability-based and back analysis of rutting in unbound granular pavements using MCMC simulations

Unbound granular pavements are highly sensitive to variability in material properties, construction quality, environmental conditions, and traffic loading, resulting in significant uncertainty in rutting performance. Deterministic pavement designs based on single-point inputs and empirical safety factors often fail to quantify this uncertainty, potentially leading to overly conservative designs or premature failure. This study proposes a probabilistic back-analysis framework for rutting assessment that integrates Bayesian inference with Markov chain Monte Carlo (MCMC) simulation within a mechanistic–empirical framework. Field compaction data are used to define prior distributions for dry density and degree of saturation, with distribution types selected based on the Bayesian Information Criterion (BIC). Bayesian updating is then applied to infer posterior distributions of design parameters while accounting for parameter uncertainty and model error. Posterior sampling is performed using the Metropolis–Hastings algorithm, and the effects of proposal scaling and sample size on sampling efficiency are evaluated. The framework is further extended to establish reliability-based compaction targets directly within the compaction plane for the first time. Comparison with the First-Order Reliability Method (FORM) shows similar failure probability trends, while the proposed approach requires substantially less computational effort. The methodology provides a practical and rigorous framework for reliability-based pavement design and quality control.