Ultimate Conditions Prediction and Stress–Strain Model for FRP-Confined Concrete Using Machine Learning

Fiber-reinforced polymers (FRP) as an exterior reinforcement material are generally utilized to enhance the effectiveness of the current and new structures. FRP-confined concrete is characterized by its better mechanical properties. Accurately predicting the ultimate conditions and stress–strain responses of FRP-confined concrete make sense to achieve superior reliability and optimized functionality of structures. In this study, four prediction models based on machine learning, containing support vector regression (SVR), back-propagation neural network (BPNN), generalized regression neural network (GRNN) and extreme learning machine (ELM), were established, and their prediction performance were compared to achieve accurate prediction of the ultimate conditions of FRP-confined concrete. Moreover, a BPNN-based model to predict the stress–strain responses of FRP-confined concrete was proposed. A carefully evaluated and scrutinized database containing 384 FRP-confined concrete specimens under compressive load from various open-access sources was used to train these models. The results showed that these prediction models were more accurate in their predictions than the design-oriented models. Moreover, GRNN and SVR had superior prediction accuracy, followed by BPNN. The machine learning-based predictive models proposed in this study served as a valuable reference for the rapid prediction on the mechanical properties of FRP-confined concrete.