Machine Learning-Enabled Real-Time Prediction of Drying Shrinkage in Fly Ash-Modified Cementitious Materials

Area of Science:

• Civil Engineering
• Materials Science
• Data Science

Background:

• Drying shrinkage is a major durability issue in cementitious materials.
• Conventional prediction methods lack temporal resolution and interpretability.
• Fly ash concrete requires accurate shrinkage prediction for sustainable infrastructure.

Purpose of the Study:

• To develop an interpretable machine learning (ML) framework for real-time drying shrinkage prediction.
• To integrate high-frequency experimental data with physically meaningful features.
• To enable continuous, data-driven monitoring of concrete behavior.

Main Methods:

• Utilized a dataset of 79,008 hourly observations from 16 concrete mix designs.
• Trained and evaluated ensemble and nonlinear ML models (Random Forest, Extra Trees, KNN, MLP).
• Employed SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• ML models achieved high predictive accuracy (R² ≈ 0.99) with low error margins.
• SHAP analysis identified multiscale temporal variables (moisture loss, desorption) as key predictors.
• Fly ash incorporation showed a secondary, consistent influence via microstructural refinement.

Conclusions:

• The ML framework enables accurate, real-time drying shrinkage prediction.
• Continuous monitoring can replace periodic testing for quality control.
• This approach supports the development of sustainable concrete infrastructures.