Shear capacity of FRP-RC beams: XML modeling and reliability evaluation

The accurate prediction of the shear capacity of concrete beams reinforced with FRP bars remains a significant challenge due to the distinct mechanical characteristics of FRP, including its linear-elastic behavior and relatively low stiffness compared to steel. To address this, the present study develops two machine learning (ML)-based solutions—a practical multilayer perceptron (MLP) model with a minimalist structure and an enhanced MLP model—for predicting the shear capacity of FRP-RC beams reinforced with both longitudinal FRP bars and FRP stirrups. A curated dataset of 144 experimental tests conducted between 1993 and 2021 was used for model development and validation. Both models achieved superior predictive accuracy compared to existing empirical equations, with goodness-of-fit values of 0.93 and 0.97 and root mean square errors of 27.5 kN and 14.5 kN, respectively, while avoiding systematic over- or underestimation. Feature ranking and sensitivity analysis showed that beam dimensions (effective depth and width) have the strongest positive correlations with shear capacity (0.90 and 0.65). The FRP stirrup ratio and bend radius-to-diameter ratio also influence shear capacity (around 0.45), while the span-to-depth ratio has a moderate negative correlation (–0.51), consistent with the mechanics of FRP-RC beams. SHapley Additive exPlanations (SHAP) analysis revealed that beam section dimensions, span-to-depth ratio, and FRP stirrup properties are the most influential input variables. The enhanced MLP model captured more complex feature interactions, while the practical model preserved interpretability and is better suited for design applications. Reliability analysis performed using Monte Carlo simulation showed that the practical model has a lower probability of failure ($P_f$ = 0.0035) compared to the enhanced model ($P_f$= 0.045), indicating its conservative and robust nature. Sensitivity reliability analysis highlighted variations in beam dimensions, FRP stirrup properties, and concrete compressive strength significantly affect structural reliability, providing useful insights for design optimization. Overall, the proposed solutions offer accurate, interpretable, and robust tools for predicting the shear capacity of FRP-RC beams, with promising applications in structural design and safety assessment.