Data-Driven Models for Predicting the Shear Strength of Rectangular and Circular Reinforced Concrete Columns

Abstract

Predicting the shear strength of structural elements subjected to gravity loads and ground motions is an important component of seismic design. Among all primary structural components, the vital role of columns in load transfer and redistribution, structural stability, and collapse prevention has been well recognized through observations made in the aftermath of past earthquakes. Numerous analytical, numerical, and experimental studies have been conducted to assess the shear strength of RC columns in the past decades. However, there is still a large scatter (i.e., uncertainty) in the predictions of current empirical and numerical models relative to test data. In this paper, novel data-driven models are presented for predicting the maximum shear strength of rectangular and circular RC columns. To this end, two extensive experimental databases for both types of columns were used in the present study. The data were randomly partitioned into calibration and validation sets. The calibration data sets served as the basis for developing linear and nonlinear equations for predicting the ultimate shear capacity of rectangular and circular RC columns through regression analyses. The Monte Carlo method was employed by conducting 106, 107, and 108 realizations to exhaustively examine the optimal parameter space of the postulated equations. The calibrated predictive models were then validated using the validation data sets; and their performances were also compared to existing models. These validation and comparison studies revealed that a new linear model devised and calibrated in the present study achieved very high accuracy, even compared to various nonlinear models considered. It was, moreover, significantly superior to all prior models in predicting the column shear strengths. This linear model, which is based on physical parameters, can therefore be recommended for engineering practice.