Development of Cracking Resistance Prediction Model of Long-Term Aged Asphalt Mixtures

With the increasing complexity of asphalt mixture composition, the current volumetric-based Superpave mixture design would not be sufficient to address durability concerns. To address this limitation, performance-based testing is introduced to supplement conventional volumetric mixture design in assessing the cracking resistance of asphalt mixtures. To perform cracking tests, samples need to be aged to represent the most embrittlement case. The current AASHTO standard for asphalt mixture long-term aging (LTA) is a 5-day oven-aging at 85°C. Quality control/assurance practices require samples to be long-term aged prior to a cracking test which is a time-consuming process. Therefore, it would be beneficial to predict LTA cracking resistance of asphalt mixture as a function of short-term aged cracking resistance. Asphalt mixture aging is complex, and various variables are involved including volumetric properties of asphalt mixture and chemical/rheological characteristics of asphalt binder. With the capability of artificial neural network (ANN) to address complex relationships, this study aimed to predict the fracture parameter, SCB Jc, of asphalt mixtures using ANN to fill the existing gap on the prediction of cracking performance with respect to aging level and component materials’ characteristics. A total of eight asphalt mixtures were selected for this study. Compacted asphalt mixtures were laboratory-aged at 85°C for 0-, 2-, 5-, 7-, and 10-days and then subjected to the semi-circular bending (SCB) test. Asphalt binders were extracted and recovered from the aged SCB samples for further chemical and rheological characterization. Asphalt binder rheological characterization indicated that the cracking resistance of asphalt binders decreased with an increase in the aging level. SCB test results showed that the fracture resistance of asphalt mixtures decreased with an increase in the aging level. Stepwise regression analysis was used to determine the significant parameters in the correlation with SCB Jc. Two models were developed: a research-based model and a DOT-based model. The artificial neural network using the feedforward backpropagation approach was then applied to train, validate, and test the predictive model. It was shown that the developed ANN models were able to accurately predict the fracture parameter, SCB Jc, of asphalt mixtures as compared to linear and non-linear regression models.