Development of Prediction Models for Complex Shear Modulus and Phase Angle of Asphalt Mastic Modified with Styrene-Butadiene-Styrene

Predicting the viscoelastic properties of asphalt mastic is very important in pavement engineering. Prediction of viscoelastic properties of mastic using machine learning methods resulting in closed form formulation is not considered in the technical literature up to now. This study aims to develop prediction models for complex shear modulus (G*) and phase angle (δ) of modified asphalt mastic with Styrene-Butadiene-Styrene (SBS) at low and medium temperatures. Three different amounts of SBS (2, 4, and 6%) are considered for bitumen modification and four different volume filling rates (10, 18, 25, and 35%) are considered for making asphalt mastic samples. Dynamic Shear Rheometer (DSR) test was performed in frequency sweep mode at 21 loading frequencies from 0.1 to 100 Hz and seven temperatures of -22, -16, -10, 0, 10, 16, and 22 ° C. This test was used to measure the G* and δ of samples of asphalt, modified asphalt, and asphalt mastic. Multi-gene genetic programming has been used to develop the G* and δ asphalt mastic prediction model based on the additive's dosage, loading frequency, temperature, filler volume filling rate, G* and δ of the base asphalt. Finally, two separate prediction models for G* and δ are developed, with a R2 value of 0.96 and 0.98, respectively. The results show that multi-gene genetic programming can accurately predict the viscoelastic behavior of asphalt mastic. After examining the performance of the models, it was shown that the viscoelastic properties of the asphalt have the greatest impact on the prediction of output variables by performing sensitivity analysis on the prediction models..