Étude et prédiction des propriétés de matériaux cimentaires à l'état frais pour la fabrication additive à l'aide de réseaux de neurones

Abstract

Additive manufacturing has gained the interest of the construction actors during these past few years. Potential cost reduction and process automation provide a new way of thinking about the manufacturing methods. 3D printing faces new challenges related to cementitious materials, especially with extrusion techniques. The mixture has to be fluid enough to be extruded properly but also stiff enough to maintain its shape under its own weight. Moreover, the printed element has to support the next layers placed upon it. This second condition is related to the yield stress of the material. In order to address these issues, admixtures can be used in formulations. In this context, this study focuses on the properties of cementitious materials in their fresh state. To do so, a two-level full-factorial design with four factors was implemented. The factors were the proportions of accelerator, viscosity modifying agent, nanoclays and CSH seeds used in the mixes under study. First, mortars were tested in conventional experiments such as the slump test and the flow test. In order to simulate the conditions of the layer overlay during 3D printing, a stability test was implemented. Linear regressions between these different experiments make it possible to establish that the slump test is well correlated to the stability test. The fact that the slump test is easy to realize makes it a very useful tool to investigate cementitious materials fresh properties. Cement pastes were then studied using a rheometer and a mini-cone to perform the mini-slump test. It was shown that the mini-slump test correlates very well with the yield stress of the paste. Artificial neural networks were trained to be able to predict the yield stress and the mini-slump as a function of the proportions of the different admixtures incorporated in the mix. Therefore, the influence of each admixture was highlighted. The accelerator under study decreases the yield stress while the other three tend to increase it. The CSH seeds admixture seem to have the greatest capacity to increase the stiffness of the paste. In addition, using a printability criterion on the yield stress, it was possible to conclude on the proportions of the admixtures ensuring a good stability of the mix during printing. Finally, two hundred and fifty mixtures were simulated, their respective yield stress and mini-slump were computed thanks to the artificial neural networks in order to conclude on the effectiveness of these regressions. Artificial neural networks are proving to be powerful tools for determining the characteristics studied.

Date	29 Aug 2019
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Claudiane Ouellet-Plamondon (Supervisor)