Artificial neural network for the prediction of the fresh properties of cementitious materials

Abstract

The admixtures influence the fresh properties of cement paste, which is a key factor in the ink design for 3D printing applications. In this study, the mix contained superplasticizer and four other admixtures (calcium silicate hydrate seeds, nanoclay, viscosity-modifying agent, accelerator) based on a factorial experimental design plan. The cement paste yield stresses measured with the rheometer are compared with the mini-slump test. An empirical relationship is proposed between the dynamic yield stress and the mini-slump. The critical yield stress for printing one layer is calculated to ensure the material to maintain its shape under his own weight. Artificial Neural Networks (ANN) are trained to predict the mini-slump and the dynamic yield stress from specific admixture proportions of the mixture. The ANN defines the amount of each admixture based on the critical yield stress. Finally, the neural networks are validated through the simulation of new mixes and by a comparison of the yield stress and mini-slump from the simulation and from experiments.