Bayesian Analysis of Pavement Maintenance Failure Probability with Markov Chain Monte Carlo Simulation

Abstract

This study presented a Bayesian logistic model to evaluate the failure probability of asphalt pavement preventive treatments. The Markov Chain Monte Carlo (MCMC) simulation using Metropolis-Hasting sampling was adopted for the Bayesian analysis. Pavement performance data and other related information, including traffic level, climate and pavement structure, were collected from the long-term pavement performance experiments for the analysis. Four preventive maintenance treatment methods, including asphalt overlay, chip seal, fog seal and slurry seal, were compared. Both a logistic model and a Bayesian logistic model with MCMC simulation were developed. Compared with the logistic model, the Bayesian logistic model can greatly reduce the uncertainty of parameter estimates. In addition, by setting the previous distribution of the parameters, the estimates can be in accordance with practical experience or previous research after Bayesian analysis. Therefore, some abnormal estimates can be corrected. Both models suggest that the pretreatment pavement condition is the most significant factor for the failure of maintenance treatments. Generally, severe climate, traffic, or poor structural capacity increased the failure probability of pavement treatments. As for the four treatments, fog seal and slurry seal performed significantly poorer than asphalt overlay and chip seal.