CSI (Cement Science Investigation): using machine learning to guess the OPC pastes composition from the elastic response

Tulio Honorio (1) , Sofiane Ait Hamadouche , Amélie Fau

1 LMT - Laboratoire de mécanique et technologie

Precise prediction of the elastic response is crucial to model cracking at early and late ages of cement-based materials and structures. Here, we use Machine Learning (ML) techniques to predict the elastic properties of Ordinary Portland Cement (OPC) pastes. A database with 365 observations is built on experimental studies from in the literature. We show that micromechanicsbased estimations may provide missing data in databases to be interrogated by ML, increasing the accuracy of predictions. Finally, we explore the formulation space of OPC pastes using Monte Carlo computations, which enables guessing which are the compositions that can be associated with a given elastic response. Applications of our results include the development of tailored formulations for a target elastic response and also in the forensics of existing cement pastes.

Mots clés: Elastic constants; Machine Learning; Data Science; Composition-Property Correlations

Domaines: Génie civil

Tulio Honorio, Sofiane Ait Hamadouche, Amélie Fau. CSI (Cement Science Investigation): using machine learning to guess the OPC pastes composition from the elastic response. CRC2021 : International RILEM Conference on early-age and long-term crack width analysis in RC structures, Apr 2021, Gif sur Yvette, France. ⟨10.1007/978-3-030-72921-9_17⟩. ⟨hal-03200465⟩