Machine-learning-enhanced variable-angle truss model for shear capacity assessment of reinforced concrete elements

Shear failure of reinforced concrete (RC) elements are typically more brittle and sudden than flexural failure. Consequently, it is of utmost importance to estimate properly the shear capacity of RC members when assessing and retrofitting existing structures as well as in the design of new constructions under later loads, for instance due to earthquakes. Historically, the development of shear capacity equations for RC beams and columns has originated from the conceptualization of a resisting mechanism. Recently, the use of data-driven approaches based on standard regression techniques has evolved by exploiting machine learning techniques. In contrast, this research ad-vances a hybrid approach to formulate a shear capacity equation for RC beams and columns with rectangular/square cross-sections. This approach enhances a mechanics-based, code-conforming formulation through the integration of a machine-learning-aided approach. Specifically, the Ge-netic Programming technique is employed to enrich the shear capacity equation based on a varia-ble-angle truss resisting mechanism. This integration results in the formulation of novel expres-sions for the two key coefficients governing the concrete contribution. The performance of the newly derived equation is assessed for beams and columns with both solid and hollow cross-sections under uniaxial shear. Some future research lines are drawn concerning ongoing experi-mental and theoretical efforts to extend the proposed model to biaxial shear. Hence, this research establishes a unified shear capacity equation for RC beams and columns. Additionally, it demon-strates the advantages of merging mechanics-based and data-driven methods. This integration proves beneficial in developing capacity equations, as it preserves the physical meaning as well as the general validity of the resisting mechanism while enhancing the accuracy through a machine learning technique.