Prediction of Pavement Performance Using Multistate Survival Models

Abstract

An accurate assessment of actual pavement condition and its future performance is essential in developing effective pavement management systems. The main drawbacks in the current developed deterioration models are that they do not examine pavement deterioration transition time and transition probabilities from one state condition to the next. Omitting transition time or transition probabilities between pavement condition states while developing pavement management models results in deterministic models that do not consider the probabilistic nature of the pavement deterioration process. This study overcomes the drawbacks in current developed deterioration models and proposes a methodology that can provide new insights into the development of long-term probabilistic pavement management tools. For this purpose, data were collected on pavement roughness, macrotexture, and age for major roads in Jordan. In addition, the vehicle speed while measuring pavement condition and macrotexture was recorded and used in the analyses. The collected data were analyzed using both fully parametric and semiparametric multistate survival approaches to identify the factors that affect pavement performance deterioration and to predict the transition probabilities between pavement condition states throughout pavement useful life. The performance of the developed models was evaluated using common statistical metrics developed in literature. The results demonstrated that the described methodology could complement conventional asset management approaches and provide descriptive and formal additional insights into road system management.