Developing a Data-Powered Framework for the Capacity of Concrete Columns Wrapped with Fiber-Reinforced Polymers

Modeling the axial load-bearing capacity of short concrete columns confined with fiber-reinforced polymer (FRP) sheets necessitates consideration of several key factors, such as material properties, geometric dimensions, and the confinement effects provided by the FRP wrapping. These considerations are vital for the design of more durable and sustainable FRP-confined concrete structures. This research presents a comparative evaluation of eight machine learning (ML) classification algorithms and one symbolic regression method aimed at predicting the enhancement in axial compressive strength (Fco/Fcc) of FRP-wrapped short concrete columns with different cross-sectional shapes. The study accounts for variables including size effect (b/bo), aspect ratio (d/b), corner rounding (r/b), wrapping stress (2·t·Ffrp / b·Fco), and wrapping stiffness (2·t·Efrp / ν·b·Ec). A thorough literature review yielded a dataset of 500 experimental results on FRP-confined concrete columns with a variety of concrete strengths, cross-sectional shapes (square and circular), FRP types, and wrap thicknesses. This dataset was divided into a training set of 400 samples (around 80%) and a validation set of 100 samples (approximately 20%). Results showed that the response surface methodology (RSM), gradient boosting (GB), CN2, support vector machine (SVM), k-nearest neighbor (KNN), and Tree models achieved excellent prediction accuracies exceeding 90%, while the RF model delivered very good performance with about 88% accuracy. In contrast, the naive bayes (NB) and stochastic gradient descent (SGD) models underperformed, reaching accuracies below 70%. Analysis using correlation matrices and sensitivity evaluations revealed that confining stress and stiffness were the most significant predictors, followed by corner radius, aspect ratio, and size effect. Notably, the RSM approach was unique in providing a closed-form equation, making it suitable for direct application in design practice.