Long-Term Performance Analysis of Demolition Waste Blends in Pavement Bases Using Experimental and Machine Learning Techniques

The use of reclaimed asphalt pavement (RAP) for the production of asphalt mixtures has become a common practice in the pavement industry. However, there have been limited studies on the long-term performance of RAP in the unbound layers of pavements. The main objective of this study was to characterize the resilient modulus (Mr) and long-term permanent deformation responses of RAP blends with recycled concrete aggregate (RCA) in the unbound base and subbase courses of pavements. To this end, up to 70% RAP by dry mass was mixed with RCA to characterize the resilient modulus, long-term permanent deformation, and shear strength responses of the blends through an extensive experimental program. The behavior of RAP/RCA blends was classified according to their permanent deformation responses and the shakedown concept, which indicated that adding up to 30% RAP could provide desirable performance in the unbound pavement layers. The support vector regression (SVR) approach was utilized as a machine learning technique for predicting the Mr and permanent deformation behavior of RAP/RCA blends. Three different kernel types, including linear (linear_svr), radial basis function (rbf_svr), and polynomial (poly_svr), were utilized in developing the machine learning models. The performance of the optimal SVR models for Mr and PS was compared with that of a random forest regression model to evaluate their predictive performance further. Further analyses were provided to validate the developed models. This study aims to promote an increased proportion of recycled aggregates in the design and construction of pavements justified through extensive laboratory characterizations backed up with robust machine learning modeling.