Abstract: The limit state model of fiber-reinforced polymer (FRP) concrete under axial compressive load can be divided into design and analytical models. The axial compressive stress and strain included in the limit state form the basis of the model parameters. Accurately calculating these parameters can provide a decision-making basis for evaluating the performance of FRP-reinforced concrete structures. Through a comprehensive evaluation of the limit state model performance of FRP-partially confined concrete with a design-oriented approach, it is shown that existing models generally exhibit poor generalizability, low prediction accuracy, and high dispersion. In response to the limitations of existing design models, the compressive strength and ultimate compressive strain of 112 FRP partially confined concrete cylinders were predicted using the XGboost (extreme gradient boosting) machine learning method. The research results indicate that the XGboost model not only overcomes the shortcomings of existing empirical models, such as poor generalizability, low prediction accuracy, and high dispersion but also reflects the importance of various parameters on axial compressive stress and strain. Moreover, compared to existing design models, the computational values based on the machine learning model align better with the experimental values, with significantly reduced deviation and randomness, ensuring the accuracy and stability of the prediction results.

Key words: FRP sheet, compressive strength, ultimate compressive strain, XGboost, confinement concrete