Supervised Learning Methods for Modeling Concrete Compressive Strength Prediction at High Temperature.

Area of Science:

• Materials Science and Engineering
• Civil Engineering
• Computational Mechanics

Background:

• Accurate prediction of concrete mechanical properties at high temperatures is crucial for structural safety.
• Supervised learning (SL) offers a promising data-driven approach for modeling complex material behaviors.
• Existing models may not fully capture the intricate relationships influencing high-temperature concrete strength.

Purpose of the Study:

• To develop and evaluate SL models (AdaBoost, RF, DT) for predicting concrete compressive strength at elevated temperatures.
• To compare the performance of these SL models against each other and literature-based models (ANN, ANFIS).
• To identify the most influential concrete mix constituents for high-temperature strength prediction.

Main Methods:

• Utilized experimental data from 207 concrete tests.
• Developed AdaBoost, Random Forest (RF), and Decision Tree (DT) supervised learning models.
• Assessed model performance using R², RSR, MAPE, and RRMSE statistical indices.

Main Results:

• AdaBoost model demonstrated excellent predictive accuracy (R² > 0.9, RSR < 0.5) for concrete compressive strength at high temperatures.
• All developed SL models showed strong correlations between experimental and predicted values.
• Sensitivity analysis identified cement content as the most significant parameter affecting high-temperature strength.

Conclusions:

• Supervised learning methods, particularly AdaBoost, are highly suitable for modeling and predicting the compressive strength of concrete at high temperatures.
• The developed models provide a reliable tool for engineers and researchers in high-temperature applications.
• Understanding the influence of mix design parameters like cement content is vital for optimizing high-temperature concrete performance.