Machine Learning Driven Fluidity and Rheological Properties Prediction of Fresh Cement-Based Materials

Area of Science:

• Materials Science
• Civil Engineering
• Computational Science

Background:

• Controlling cement-based material workability is crucial but time-consuming.
• Artificial intelligence (AI) offers potential for efficient mix design optimization.
• Predicting early workability indicators aids in rapid material development.

Purpose of the Study:

• To apply AI and machine learning (ML) for predicting cement-based material workability.
• To develop optimized hybrid ML models for enhanced prediction accuracy.
• To analyze the influence of mix components on workability using SHAP and PDP.

Main Methods:

• Experimental testing to generate a dataset of 233 cement-based material samples.
• Utilizing ML algorithms, including optimized hybrid models like Particle Swarm Optimization (PSO)-based CatBoost and XGBoost.
• Employing Shapley Additive Explanations (SHAP) and Partial Dependence Plot (PDP) for variable influence analysis.

Main Results:

• Accurate predictive models for fluidity, dynamic yield stress, and plastic viscosity were established.
• Optimized hybrid models demonstrated improved prediction capabilities for workability indicators.
• Key input variables influencing workability were identified through SHAP and PDP analyses.

Conclusions:

• AI and ML significantly enhance the efficiency of cement-based material mix design.
• Optimized hybrid models provide a robust framework for predicting workability.
• This study offers a valuable reference for rapid and accurate control of cement-based material workability.