Modeling of moisture damage in carbon nano tube modified asphalt using hybrid of artificial neural network and other computational intelligence approaches

This paper explores and analyses the application of a number of Artificial Neural Networks for modeling adhesion force in CNT modified asphalt binder. A number of different learning algorithms was applied to train the ANN. An AFM was used to collect adhesion force in CNT modified asphalts at nano-scale level. A total of 270 data samples were collected by using an AFM in this study. All samples were tested under dry and wet conditions. Base binder, 4% SB and SBS, 5% SB and SBS (total 5 types) of asphalt binders were used and CNT modification was attempted on the polymer modified asphalt binders. All the binders were mixed with 2 types of CNT (Single Wall and Multi Wall). Each of the SWNT and MWNT were also used in 3 different percentages. A Particle Swarm Optimization (PSO) was employed to identify an optimal architecture of ANN and three other approaches namely, Backpropagation algorithm (BP), PSO and Simulated Annealing (SA) were explored to adapt the connecting weight of ANN. The experimental results show that the hybrid of PSO-ANN can better model the adhesion force in CNT modified asphalt binder compared to the other individual and hybrid ANNs (e.g., BP-ANN and SA-ANN).