A neural network approach for predicting hardened property of geopolymer concrete

Abstract

This paper presents the application of an Artificial Neural Network (ANN) approach to predict the 28-day compression strength of Geopolymer concrete (GPC) from the input ingredients. A total of 190 test samples collected from previously published were employed for training and validating the ANN model. Additionally, a test project was also implemented to collect the experimental data for verifying the prediction ability of the ANN model. Different learning algorithms were investigated to obtain the optimal algorithm for the GPC data. Results from the study revealed that the ANN model using the "trainlm" learning algorithm provided the best prediction results. The average prediction error about 8 MPa was found for the unseen data set. Besides, the effects of changing input variables to the output of the model were also explored by conducting the sensitivity analysis. It was shown that the 28-day GPC compression strength was more sensitive to the change of coarse aggregate (CoAg) and sodium silicate (Na2SiO3) variables.