Use of Random Forest to Predict Intermediate Temperature SCB J c Parameter of Long-Term Aged Asphalt Mixtures

Abstract

With the growing use of reclaimed asphalt pavement (RAP), recycled asphalt shingles (RAS), and other innovative additives in asphalt mixtures, there are concerns about the accuracy of Superpave volumetric-based mixture design. Performance tests in mix design procedures are needed to ensure desirable pavement performance. The Louisiana Department of Transportation and Development (LADOTD) has adopted the semi-circular bend (SCB) critical strain energy release rate parameter, J_c, to evaluate the cracking resistance of asphalt mixtures. To address practical shortcomings, this study aims to develop a predictive model to estimate SCB J_c at intermediate temperature of long-term aged (LTA) asphalt specimens from volumetric properties of plant-produced asphalt mixtures, and rheological and chemical properties of unaged asphalt binder by using a machine-learning model, random forest. Asphalt mixture SCB test and asphalt binder rheological and chemical properties tests were conducted. Stepwise correlation analysis was used to determine the most influential variables to SCB J_c. Random forest model was optimized using grid search with hyperparameter combinations. Results show that the developed random forest model was able to predict the SCB J_c fracture parameter of asphalt mixtures and a good agreement was observed between predicted and measured SCB J_c values. Variable importance scores based on mean decrease in impurity (MDI) showed that the ΔT_c had the most influence on SCB J_c prediction accuracy, and other inputs such as coarse aggregate type, LTA days, Fourier transform infrared spectroscopy test carbonyl index, and linear amplitude sweep test A_Las also had significant effects on SCB J_c.