Shear Capacity of RC Elements With Transverse Reinforcement Through a Variable-Angle Truss Model With Machine-Learning- Calibrated Coefficients

This chapter addresses the shear capacity prediction of reinforced concrete elements with transverse reinforcement through a hybrid approach in which a mechanical model (i.e., the variable-angle truss model incorporated in the Eurocode 2) is enhanced with two machine-learning-calibrated corrective coefficients aimed at improving the accuracy of the numerical predictions. Genetic programming is adopted to derive closed-form expressions of the corrective coefficients, thus making the proposed formulation suitable for design purposes and usable by practitioners. The model predictive performance and the improvements over alternative code-based formulations is demonstrated through a wide database of experimental results of reinforced concrete beams and columns with plain and hollow sections failing in shear under both monotonic and cyclic loading conditions. The proposed approach leads to numerical-to-experimental shear capacity ratios having mean value close to one and coefficients of variation equal to 32%, 28% and 24% for beams, columns with plain and hollow sections, respectively.