Using Artificial Neural Network Methods for Modeling Moisture Susceptibility of Asphalt Mixture Modified by Nano TiO2

Abstract

The aim of the current research was to evaluate, model, and predict the water sensitivity of a mixture modified by nano TiO2 using the artificial neural network (ANN) and multiple linear regression (MLR) models. Also, the effect of nano TiO2 on the performance of asphalt binder was evaluated. The indirect tensile strength (ITS) test was used to measure the performance of different mixtures through ITS, tensile strength ratio (TSR), and fracture energy ratio (FER) parameters. To measure the effect of nano TiO2 on the performance of mixtures, five percentages of nano TiO2 (0%, 2%, 4%, 6%, and 8%) were added to the asphalt binder. In the current research, steel slag was used as a substitution of coarse aggregate not only to use waste materials but also to omit environmental emissions. Results revealed that using nano TiO2 decreased the penetration grade and ductility and increased the softening point of asphalt binders. Results revealed that addition of nano TiO2 decreased the temperature susceptibility of binder. Regarding the viscosity test results, applying nano TiO2 increased the viscosity of control asphalt binder. Moreover, the viscosity values of asphalt binders increased as the nano TiO2 percentages raised. The viscosity-temperature susceptibility (VTS) test results showed that TiO2 reduced the VTS of the control asphalt binder and lower-VTS asphalt binders had a better resistance. The results also indicated that with substitution of aggregate ITS, TSR, fracture energy (FE), and FER of asphalt mixtures reduced. Based on results, addition of nano TiO2 increased the ITS and FE of mixtures in wet and dry condition and also the TSR and FER of mixtures increased. Moreover, based on the results, by assessing and comparing the experimental and predicted values, an appropriate accuracy in the estimation of values by the ANN and MLR models was determined. Also, the ANN model had better accuracy than MLR in predicting results in all cases.