Viscosity Prediction of Rubberized Asphalt–Rejuvenated Recycled Asphalt Pavement Binders Using Artificial Neural Network Approach

Abstract

The objective of this study was to develop artificial neural networks to predict the viscosity of rubberized asphalt rejuvenated reclaimed asphalt pavement binders. Eight variables were selected as input factors, namely, viscosity measuring temperature, rubber blending time, reclaimed asphalt pavement blending time, original binder blending time, rubber content, reclaimed asphalt pavement content, blending temperature for aged binder, and asphalt type. Two viscosity analysis models, backpropagation artificial neural networks and genetic algorithm modified artificial neural networks, were developed in this study. It was found that both artificial neural network models were effective in predicting the viscosity of rubberized asphalt rejuvenated reclaimed asphalt pavement binders. Through sensitivity analysis, blending temperature for aged binder, viscosity measuring temperature, original binder blending time, and reclaimed asphalt pavement blending time were found to be important variables that contributed to the binder viscosity. On the contrary, the asphalt type and rubber blending time were found to be less important. As a result, the viscosity of rubberized asphalt rejuvenated reclaimed asphalt pavement binders changed significantly with the blending temperature, blending time of the aged binder, and blending time of the original binder. Both backpropagation artificial neural networks and genetic algorithm modified artificial neural networks viscosity models were validated using data collected from prior studies, and the results were barely acceptable.