Prediction of Marshall Test Results for Dense Glasphalt Mixtures Using Artificial Neural Networks

Abstract

Asphalt mixture comprising waste glass as an aggregate is referred to as “glasphalt”. Limited studies have been oriented to investigate the Marshall test results of dense-graded glasphalt mixes considering a wide range of variables. This study aims to utilize the artificial neural networks (ANNs) to develop predictive models for Marshall stability and Marshall flow of dense glasphalt mixes based on a large experimental database collected from the literature. Eight independent variables covering the material and mix properties were utilized as inputs in the models. The proposed models resulted in an experimental-to-predicted ratio of 1.00 and 1.00, coefficient of variation of 8.6% and 8.7%, RMSE of 1.63 kN and 0.54 mm, and R-squared of 93.6% and 85.7% for the glasphalt stability and flow models, respectively. Comprehensive parametric analyses have been conducted to further validate the models by investigating the sensitivity of their parameters to the predicted stability and flow values. The analyses revealed some desirable design values that could be considered for a better performance of dense glasphalt mixes. The results indicate that 4% is the desired design air void content of glasphalt mixes. High stability value can be achieved for glasphalt mixes containing a crushed aggregate of 12.5 mm maximum size and 50% glass cullet of 4.75 mm maximum size. Lower viscosity asphalt binder would provide uniformly compacted mixes. Furthermore, glasphalt flow increases as the maximum size of ingredient particles, the penetration grade of asphalt cement, asphalt cement content, and VMA% increase.