Enhancing the Analysis of Rheological Behavior in Clinker-Aided Cementitious Systems Through Large Language...

Area of Science:

• Materials Science
• Chemical Engineering
• Data Science

Background:

• Understanding cement and polycarboxylate ether (PCE) admixture compatibility is crucial for concrete performance.
• Grinding aids (GAs) influence cement properties and their interaction with PCE admixtures requires detailed investigation.

Purpose of the Study:

• To investigate parameters affecting cement and PCE admixture compatibility using various grinding aids (GAs).
• To evaluate the impact of Large Language Models (LLMs) for synthetic data augmentation on predictive modeling of cement rheology.

Main Methods:

• Prepared 29 cement types with seven GAs at four dosages and 87 paste mixtures with three PCE dosages.
• Evaluated rheological behavior (dynamic yield stress and viscosity) using the Herschel-Bulkley model.
• Modeled data using Convolutional Neural Networks (CNN), Random Forest (RF), and Neural Classification and Regression Tree (NCART), enhanced with LLM-generated synthetic data.

Main Results:

• NCART demonstrated the best baseline performance for viscosity and dynamic yield stress (DYS).
• LLM-based synthetic data augmentation significantly improved the predictive accuracy of all tested models (CNN, RF, NCART).
• Cements produced with GAs showed higher DYS and viscosity than the control, attributed to finer particle size distribution.

Conclusions:

• LLM-based synthetic data augmentation shows significant potential for enhancing predictive models in cement admixture compatibility studies.
• The findings underscore the importance of GAs in modifying cement properties and their interaction with PCE admixtures.
• This research pioneers the use of LLMs for synthetic data augmentation in materials science modeling.