A Heterogeneous Markov Chain Model to Predict Pavement Deterioration and Optimize Repair Activities

Abstract

In an era where system needs exceed available funding across all infrastructure components, planners and decision makers need tools to make informed decisions about the value of assets, such as accurate models to predict pavement condition over time. This prediction is essential in pavement management systems (PMS) because it provides information that allows forecasting repair demands and optimizing life-cycle costs. Markov Chains have been proposed in the past as a tool for forecasting pavement performance and deterioration. A weakness of the traditional Markov Chain is that the conventional transition matrix has limited ability to account for site-specific variability. To address this problem, this dissertation proposes a combination of ordinal logistic regression and Markov Chains. The logistic regression models were found to bring two major improvements to the Markov model. First, the enhanced Markov transition probability matrix allows for site specific predictions because the models use specific characteristics of each pavement section. Second, the enhanced matrix offers additional benefits by allowing the comparison of several factors and the analysis of how each of them influences the pavement performance and deterioration, as well as providing an understanding of the interaction between these several external factors, such as district location, repair history, functional class, base thickness, speed limit and pavement thickness. Numerical examples are provided to demonstrate how the Markov Transition Matrix can be used to model future pavement deterioration. The final Markov probability matrix was also used to determine an optimal sequence of pavement repair activities.