Machine learning enhanced modeling of steel‐concrete bond strength under elevated temperature exposure

Abstract

This study uses machine learning techniques to investigate the bond strength between steel and concrete under various elevated temperature scenarios. Five distinct machine learning algorithms, including Random Forest (RF), XGBoost, AdaBoost, Decision Tree, Linear Regression, and hyperparameteric optimisations, were used to predict changes in bond strength. The models underwent rigorous optimisation using GridSearchCV to achieve optimal performance. In this study, we evaluated several metrics such as Mean Squared Error, Root Mean Squared Error, Mean Absolute Error, and coefficient of determination (R2) Score to compare and assess the models' prediction capabilities. After optimisation, results indicate that the RF model exhibited exceptional performance in estimating bond strength across different temperature conditions, demonstrating minimal errors and a high R2 Score. Visual comparisons of actual and predicted values further confirmed the efficacy of the RF model in capturing complex fluctuations in bond strength. The findings of this study underscore the potential of machine learning models, particularly the optimized RF method, in accurately predicting bond strength under varying thermal conditions, with promising implications for engineering and construction practices.