A statistical based framework for predicting field cracking performance of asphalt pavements: Application to top-down cracking prediction

Most existing cracking performance models of asphalt pavements, such as top-down cracking models, are mechanistic or mechanistic-empirical based. These models usually focus on a specific type of cracking mechanism. The prediction quality of these models can also vary when field cracking conditions are complicated and clear identification of the distress type is difficult. Literature suggests that a statistical based method can account for variability and a large number of influencing factors, and could be a promising alternative. This paper aims to introduce a statistical based framework for performance prediction using top-down cracking as an example. Such a framework can be modified and implemented by local agencies for a variety of cracking distresses based on specific needs and requirements. Detailed steps of the statistical framework are presented through the development of top-down cracking models. Results indicate that the framework works effectively by integrating pavement performance concepts with several statistical methods including Partial Least Squares (PLS) Regression, Binary Logistic Regression, and Leave one out cross validation (LOOCV). Using the developed statistical based framework, critical factors that may affect the initiation and propagation of top-down cracking are identified, and their sensitivities to the field top-down cracking are further discussed.