Using artificial neural networks to predict the 28-day compressive strength of roller-compacted concrete pavements containing RAP aggregates

Abstract

This paper presents an Artificial Neural Network (ANN) model to predict the 28-day compressive strength of roller-compacted concrete pavement (RCCP) mixes containing reclaimed asphalt pavement (RAP) aggregates. Three supervised learning ANN algorithms, namely, Levenberg–Marquardt (LM), Bayesian Regularisation (BR), and Scaled-Conjugate Gradient (SCG) were used to predict the 28-day compressive strength of the RAP-RCCP mixes. It was observed that constructed ANN models showed good prediction between the input parameters and the output value. In fact, the BR-ANN model of 13–9–1 (13 neurons in the input layer, 9 hidden neurons, and 1 output layer) architecture showed an excellent prediction amongst the supervised learning ANN models with the coefficient of determination of 0.985 in the testing phase. Using 2-hidden layers (13–9–9–1 architecture) was found to have the highest precision with the regression values for the training, validation, and testing being 0.975, 0.994, and 0.961, respectively. In fact, the constructed model predicted the 28-day compressive strength with a very low MSE of 0.99 MPa. Hence, it can be concluded that the use of ANN could be used to predict the compressive strength of RCCP mixes containing varying types of RAP. Moreover, a sensitivity analysis was also carried out to study the effect of the influencing and contribution input parameters to the output prediction.