Development of prediction models for moisture susceptibility of asphalt mixture containing combined SBR, waste CR and ASA using support vector regression and artificial neural network methods

Abstract

Precise estimation of the moisture susceptibility of mixtures is a difficult approach because of the complicated characteristics of used materials under numerous environmental and traffic situations. As the virgin binder has low performance to different traffic and environmental conditions, utilization of additives are proposed. Current research discovers the possible utilization of artificial neural network (ANN) in forecasting the moisture susceptibility of asphalt mixtures modified by different additives. Over 100 samples were fabricated with three types of anti-stripping agents (named A, B, and C) with various contents (0.2%, 0.4%, and 0.6% by weight of binder), one percentage of CR (7% by weight of binder), one percentage of SBR (2% by weight of binder), and one type of AC-85/100 penetration grade bitumen, and tested through different tests such as; Texas boiling test, Tensile Strength Ratio (TSR), Fracture Energy Ratio (FER), and Resilient Modulus Ratio (RMR). Also, some physical and rheological properties of modified binders were investigated. The fracture energy (FE), indirect tensile strength (ITS) and resilient modulus (Mr) of mixtures improved by incorporation of CR and SBR. Also, have positive impact in enhancing the properties of mixtures. Based on results, ASA (B) has the best impact on enhancing the moisture susceptibility of mixtures. Moreover, some prediction models were proposed to compare with experimental methods. Support vector regression (SVR) and artificial neural network (ANN) models were designed for the prediction of TSR, RMR, and FER values. The results showed that ANN had better performance than SVR in all cases.