Performance-based design of FRP-confined recycled aggregate concrete powered by machine learning techniques

Using fiber reinforced polymer (FRP) to confine recycled aggregate concrete (RAC) has been considered as a value-added and promising solution to improve the inferior mechanical properties of RAC. However, accurate quantification of the axial compressive performance of FRP-confined RAC columns and further inverse design remain a significant challenge, due to the complex failure mechanisms involved and RAC's inherent defects. This study aims to develop machine learning (ML) models in conjunction with multi-objective optimization method for performance prediction and inverse design of FRP-confined RAC columns. A comprehensive database consisting of 213 sets of experimental results on the ultimate conditions of FRP-confined RAC columns (i.e., ultimate compressive strength (f_cc) and strain (ɛ_cu)) was first established. Grey relational analysis (GRA) was then performed to assess the parametric sensitivity of the axial compressive behavior of FRP-confined RAC columns. Subsequently, three different ML models were developed to evaluate the f_cc and ɛ_cu of FRP-confined RAC columns using Bayesian hyperparameter optimization, which showed reasonable accuracy and were superior to existing empirical equations in the literature. The least-squares boosting (LSBoost) models outperformed the other ML techniques with the best prediction accuracy and generalization ability. Lam and Teng's constitutive model was further calibrated with LSBoost, well reproducing the axial stress-strain curves of FRP-confined RAC columns. Finally, combined with the developed LSBoost prediction models, a multi-objective non-dominated sorting genetic algorithm of type II (NSGA-II) was successfully used to obtain the design parameters of FRP-confined RAC columns targeting specific performance requirements. In addition, four data-driven, probabilistic, explicit prediction equations were proposed for practical design of FRP-confined RAC columns using a Bayesian model updating approach. This study offers a ML-powered performance-based design strategy for FRP-confined RAC columns, which will promote efficient development and applications of this innovative composite member.