Estimation of shear capacity for RC short columns using neural networks

Abstract Reinforced concrete (RC) short columns have suffered significant damage during past earthquakes, demonstrating a brittle shear failure. Many design codes lack a specific equation to estimate the shear capacity of RC short columns; consequently, engineers have relied on existing equations for ordinary beams and columns. However, it has been shown that these equations do not accurately estimate the shear capacity of RC short columns. This study aims to develop a feedforward neural network model to predict the shear capacity of RC short columns under uni-directional seismic excitation. A dataset consisting of 57 test results and 258 simulations was prepared. The influential parameters affecting the shear capacity of RC short columns were identified and used as inputs for the neural networks in two scenarios. The first scenario included eight input parameters, while the second scenario contained six input parameters. Results indicated that both scenarios predicted the shear capacity of RC short columns with good accuracy; however, the first scenario yielded smaller prediction errors.