Neural network modeling of SBS modified bitumen produced with different methods

Abstract

Various types of polymers are added to bitumen in order to improve its properties under low and high temperatures. It is important to determine accurately the complex modulus of polymer-modified bitumen samples (PMBs) in order to make a suitable mix design. Moreover the determination of the complex modulus is important in order to evaluate the efficiency of the additives. However the manufacture processes of PMBs involve many factors. This study aims to model the complex modulus of styrene–butadiene–styrene (SBS) modified bitumen samples that were produced by different methods using artificial neural networks (ANNs). PMB samples were produced by mixing a 160/220 penetration grade base bitumen with 4% SBS Kraton D1101 copolymer at 18 different combinations of three mixing temperatures, three mixing times and two mixing rates. The complex modulus of PMBs was determined at five different test temperatures and at ten different frequencies. Therefore a total of 900 combinations were evaluated. Various different results were obtained for the same PMB produced at different conditions. In the ANN model, the mixing temperature, rate and time as well as the test temperature and frequency were the parameters for the input layer whereas the complex modulus was the parameter for the output layer. The most suitable algorithm and the number of neurons in the hidden layer were determined as Levenberg–Marguardt with 3 neurons. It was concluded that, ANNs could be used as an accurate method for the prediction of the complex modulus of PMBs, which were produced using different methods.