Prediction of Pavement Base-Course Mixture Performance Using an Artificial Neural Network Approach

The availability of virgin materials for road pavement construction is gradually decreasing over the years. For this reason, the effort of many researchers worldwide is focused on alternative sustainable solutions capable of achieving suitable mechanical performance and, at the same time, reducing the consumption of virgin raw materials. Furthermore, the development of data-driven models capable of predicting the performance of these mixtures would further contribute to increasing sustainability by reducing the need for new experimental tests to be carried out to investigate their performance under different testing conditions. The main goal of the present study consists in training, validating and testing an artificial neural network model based on a dataset related to pavement base-course mixtures realized with construction and demolition waste coupled with reclaimed asphalt pavement partially substituting virgin raw materials. Both mechanical and volumetric performance expressed in terms of indirect tensile strength at 25 °C and air voids, respectively, were successfully predicted based on the main compositional variables. Pearson correlation coefficients equal to 0.9377 and 0.9409 were respectively obtained for the former and the latter predictions. This preliminary feasibility study confirmed that reliable mechanical and volumetric values related to these sustainable mixtures could be confidently predicted using soft-computing techniques, thus allowing these results to be used during the conventional mix-design operations significantly lowering the effort needed from laboratory technicians to determine the same results experimentally.