Data-driven axial compressive strength investigation of FRP-confined coral aggregate concrete

Abstract

This study utilizes machine learning (ML) method to investigate the axial compressive strength of fiber-reinforced polymer (FRP)-confined coral aggregate concrete (CAC). A dataset comprising 115 samples is created, and eight input features are selected for developing and evaluating ML models. Besides, six empirical formulae are used to compare their performance against the ML models. The SHapley Additive exPlanation (SHAP) algorithm is employed to elucidate the prediction mechanisms of the ML models and to clarify the interactions between the eight input features and the axial compressive strength of FRP-confined CAC. A comparison of evaluation metrics indicates that the empirical model, which is developed for compressive strength of FRP-confined geopolymer-based CAC prediction, outperforms the other five empirical formulas in precision, boasting the highest R² value of 0.84. In comparison, with the exception of the KNN model, the remaining five data-driven ML models exhibit high precision in predicting the axial compressive strength of FRP-confined CAC, with metric R2 values exceeding 0.93 on both the training and testing dataset. Besides, the axial compressive strength of confined CAC is primarily influenced by thickness of FRP layer and unconfined compressive strength of CAC, and the elastic modulus and ultimate strength of FRP are also critical factors. Furthermore, excessive FRP confinement will not further enhance the axial compressive strength of confined CAC, and CAC column with a larger diameter necessitates either a thicker FRP layer or a higher FRP strength to achieve desired compressive strength.