Reliability of artificial neural networks in predicting shear strength of reinforced concrete beams

This paper investigates the reliability of artificial neural networks (ANNs) in predicting the shear strength of shear-critical reinforced concrete (RC) beams using multi-layer back-propagation neural network (MBNN) and radial basis function neural network (RBFNN) paradigms. For this purpose, the ANN models are built, trained, and tested using an extensive database of 181 tested specimens obtained from the technical literature. The data used in the ANN models comprises nine input parameters, namely cross-sectional dimensions, cylinder compressive strength, yield strength of the longitudinal and transverse reinforcing bars, shear-span-to-effective-depth ratio, span-to-effective-depth ratio, and longitudinal and transverse reinforcement ratios. The ACI-318 design equation predicted ultimate shear strengths were compared against the MBNN, and RBFNN predicted results. The comparison results revealed that the RBFNN model could predict the ultimate shear strength of RC beams more accurately compared to the MBNN model and ACI-318 design equation. The reliability of the prediction was independently verified using a shear-critical RC beam fabricated solely for this study and another shear-critical RC beam sourced from the literature outside the collated database.