The Application of Machine Learning Algorithms to Bond Strength between Steel Rebars and Concrete Using Bayesian...

Area of Science:

• Civil Engineering
• Materials Science
• Computational Science

Background:

• Bond strength between steel reinforcement and concrete is critical for structural integrity.
• Existing empirical models may not fully capture the complex factors influencing bond behavior.
• Accurate prediction of bond strength is essential for safe and efficient structural design.

Purpose of the Study:

• To estimate the bond strength between steel rebars and concrete using machine learning (ML) algorithms.
• To compare the performance of different ML algorithms, including random forest (RF), support vector regression (SVR), and extreme gradient boosting (XGBoost), enhanced with Bayesian optimization (BO).
• To develop a simplified ML model for practical application in structural engineering.

Main Methods:

• Utilized a dataset of 401 beam tests incorporating six impact factors.
• Implemented and compared RF, SVR, and XGBoost algorithms integrated with Bayesian optimization (BO).
• Employed Shapley additive explanation (SHAP) to interpret the ML model's predictions.

Main Results:

• The Bayesian optimization-XGBoost (BO-XGBoost) model demonstrated superior accuracy compared to empirical models.
• BO-XGBoost achieved R², MAE, and RMSE values of 0.87, 0.897 MPa, and 1.516 MPa on the test set, respectively.
• A simplified model with three input variables (rebar diameter, yield strength, concrete compressive strength) was developed for ease of use.

Conclusions:

• Machine learning, particularly BO-XGBoost, provides a highly accurate method for predicting steel-concrete bond strength.
• The developed simplified model enhances the practical applicability of ML in structural analysis.
• ML algorithms can effectively predict complex interfacial mechanical behaviors, leading to improved model accuracy and reliability.