Machine learning approaches for estimating concrete shear strength in FRPreinforced members without shear reinforcement

Machine-learning techniques have significantly advanced in structural design, offering efficient, precise, and dominance over conventional methods. FRP bars, with favorable physical attributes, are extensively used as alternative reinforcement in various structural members. Shear modeling in these members gains importance due to the brittle nature of shear failure, leading to conservative shear strength estimates in current codes. Numerous design parameters, such as crosssection dimensions, shear span to effective depth ratio, concrete compressive strength, and axial stiffness of FRP bars, influence shear strength. Consequently, efficiently estimating the shear capacity of these members using traditional mathematical approaches is exceptionally challenging. This study aims to develop and assess the effectiveness of Artificial Neural Networks (ANNs) - Multilayer Perceptron (MPNN) and General Regression (GRNN) - and Support Vector Machine (SVM) with Radial Bias Function (RBF) techniques in predicting concrete shear capacity of FRP-reinforced members without stirrups. Models' findings, along with various code provisions, compared with shear testing outcomes of 555 specimens, revealed GRNN, SVM, and MPNN consecutively outperformed existing code formulas in performance, efficiency, and precision. The parametric study showed that GRNN accurately delineates the interaction of design variables on shearstrength, with a greater potential to forecast variable behavior despite its complexity and sensitivity.