Machine Learning Models for Predicting Freeze-Thaw Damage of Concrete Under Subzero Temperature Curing Conditions

Area of Science:

• Civil Engineering
• Materials Science
• Computational Science

Background:

• Concrete structures in permafrost regions face significant freeze-thaw challenges.
• Existing methods for assessing frost resistance are insufficient for accurate prediction.
• Machine learning offers a promising approach for evaluating concrete durability under low temperatures.

Purpose of the Study:

• To develop and validate advanced machine learning models for predicting freeze-thaw damage factors in concrete.
• To compare the performance of various machine learning models, including SVM, ELM, LSTM, and RBFNN.
• To introduce an optimized model, SSA-ELM, for enhanced prediction accuracy and interpretability.

Main Methods:

• Utilized Support Vector Machine (SVM), Extreme Learning Machine (ELM), Long Short-Term Memory (LSTM), and Radial Basis Function Neural Network (RBFNN) models.
• Developed an optimized Extreme Learning Machine model using the Sparrow Search Algorithm (SSA-ELM).
• Employed SHapley Additive exPlanations (SHAP) for model interpretability and identified key influencing factors.

Main Results:

• The SSA-ELM model demonstrated superior accuracy in predicting concrete freeze-thaw resistance compared to SVM, ELM, LSTM, and RBFNN.
• SHAP analysis identified the number of freeze-thaw cycles as the most critical factor influencing freeze-thaw damage.
• The SSA-ELM model showed accuracy improvements of 15.46% (SVM), 9.19% (ELM), 21.79% (LSTM), and 11.76% (RBFNN).

Conclusions:

• The SSA-ELM model provides a highly accurate and reliable method for predicting concrete freeze-thaw damage under low-temperature conditions.
• The number of freeze-thaw cycles is the primary driver of concrete damage, while curing humidity has minimal impact.
• This research offers a novel, data-driven approach for assessing the durability of concrete in permafrost environments.