Neural Network Modeling for the Rotational Viscosity of Reacted and Activated Rubber-Modified Binders

Abstract

Abstract Crumb rubber surface activation and pretreatment are considered one of the promising newly introduced methods for asphalt rubber production. Reacted and activated rubber (RAR) is an elastomeric asphalt extender produced by the hot blending and activation of crumb rubber with asphalt and activated mineral binder stabilizer. Besides RAR’s ability to enhance the performance of asphaltic mixtures, its dry granulate industrial form enables its addition directly into the mixture utilizing the pugmill or dryer drum with very minimal to no modification required on the plant level. This study aims to develop an artificial neural network (ANN) viscosity prediction model for extracting a standalone viscosity prediction equation. Three different performance graded (PG) asphalt binders modified by 10 dosages of RAR were tested and evaluated under this study. Sixty-six samples that generated more than 3,000 viscosity data points were utilized in ANN modeling. The developed ANN model as well as the extracted standalone viscosity prediction equation had a high value of the coefficient of determination and were statistically valid. Both have the ability to predict the RAR-modified binder viscosity as a function of binder grade, temperature, testing shearing rates, and RAR content.