Article Dans Une Revue Journal of Building Engineering Année : 2024

Machine learning-aided prediction of Shrinkage in Modern Concrete: Focus on Mix Proportions and SCMs

Résumé

Due to the emergence of low-carbon concrete, a deeper understanding of the role of supplementary cementitious materials (SCMs) and environmental conditions on the shrinkage properties and models capable of estimating the shrinkage amplitude are desirable. This study offers one of the first comprehensive applications of interpretable machine learning to predict the total shrinkage amplitude of modern concrete formulations. The models provide valuable insights into key factors affecting shrinkage and the role of SCMs, thus paving the way for future research on more efficient prediction models. To this end, a comprehensive database has been developed, extending the Bažant shrinkage dataset to include recent results on total shrinkage measurement of concrete with SCMs and high performance and ultra-high performance concrete. Several state-of-the-art ensemble machine learning models (XGBoost, Light Gradient Boosting, Random Forest and Extra Trees) are then optimised and used to predict the evolution of total shrinkage over time based on mix formulation, specimen geometry and test conditions. The Extra Trees model showed the highest accuracy (R²=0.957). The identification of the most influential inputs was performed, and individual prediction analysis provided quantitative information on their relative influence. The main influencing parameters were found to be the sand/aggregate ratio and test parameters such as volume to surface ratio, curing time and relative humidity, but the model highlighted the beneficial influence of reduced cement content and the positive effect of additives such as fly ash. The dataset created is made available to enable further investigations on shrinkage mechanisms, particularly concerning SCMs.
Fichier sous embargo
Fichier sous embargo
0 5 30
Année Mois Jours
Avant la publication
dimanche 3 août 2025
Fichier sous embargo
dimanche 3 août 2025
Connectez-vous pour demander l'accès au fichier

Dates et versions

hal-04922653 , version 1 (31-01-2025)

Identifiants

Citer

Benoît Hilloulin, Rejoice Umunnakwe. Machine learning-aided prediction of Shrinkage in Modern Concrete: Focus on Mix Proportions and SCMs. Journal of Building Engineering, 2024, 98, pp.111410. ⟨10.1016/j.jobe.2024.111410⟩. ⟨hal-04922653⟩
0 Consultations
0 Téléchargements

Altmetric

Partager

More