Data-driven structural performance assessment of FRP-confined concrete under environmental exposure

Environmental exposure plays a critical role in the structural performance of fiber-reinforced polymer (FRP)-confined concrete cylinders and recent advances in machine learning (ML) have opened new avenues for improving predictive accuracy in civil engineering. FRP confinement has the dominant influence on compressive strength (CS) and data-driven models have proven effective in simulating intricate and multivariable responses of the complex structural behaviour. However, existing stress-strain analytical models do not account for environmental exposure conditions, which limits their potential for application in real-world conditions. They also require strong assumptions and extensive experimental validation, creating barriers for widespread use in engineering practice. Therefore, this study developed and validated a collection of ML models such as linear regression, artificial neural networks, decision trees, random forest, gradient boosting regression trees, eXtreme Gradient Boosting (XGB), and Optuna-XGB, to predict FRP-confined CS under different exposure conditions. The models were trained on a comprehensive dataset of experimental results, with Optuna-XGB outperforming both ML peers and five conventional analytical models in predictive accuracy. SHAP feature analysis and partial dependence plots suggested CS was significantly impacted by FRP layers, concrete grade, FRP type, cylinder diameter, and exposure type. These results highlight the ability of ML to go beyond the confines of traditional models and offer a more precise and flexible way of predicting the performance of FRP-reinforced concrete in environmental conditions. • Extensive database on FRP-strengthened concrete cylinders under exposure conditions was created. • Machine learning identified influential parameters on strength of FRP-strengthened cylinders. • Data-driven approach offers more accurate alternative to traditional models. • Coupled mechanics and environmental exposure effects captured using machine learning approach.