Investigation of complex modulus of asphalt mastic by artificial neural networks

Abstract

This study investigates the complex modulus of asphalt mastic by using artificial neural networks (ANNs). The complex modulus of asphalt mastic samples are determined by using dynamic shear rheometer (DSR). Seven filler-asphalt ratios (F/A) are considered in this study: 0.0, 0.4, 0.6, 0.9, 1.2, 1.5, 1.8 by the weight of asphalt binder. In ANN model, the asphalt mastic temperature, frequency and F/A are the parameters for the input layer where as the complex modulus is the parameter for the output layer. The variants of the algorithm, such as the Levenberg-Marquardt (LM), scaled conjugate gradient (SCG) and Pola-Ribiere conjugate gradient (CGP) algorithms are used in this study. A tangent sigmoid transfer function is used for both hidden layer and the output layer. The statistical indicators, such as the root-mean squared error (<i style="mso-bidi-font-style:normal">RMSE</i>), the coefficient of multiple determination (<i style="mso-bidi-font-style:normal">R</i>²) and the coefficient of variation (<i style="mso-bidi-font-style:normal">COV</i>) are utilized to compare the predicted and measured values for model validation. Results indicate that the LM algorithm appears to be the most optimal topology. It is also demonstrated that neural networks is an excellent method that can reduce the time consumed and can be used as an important tool in evaluating complex modulus of asphalt mastic.