Flexural behavior of reinforced concrete beams and prediction of failure stages

Abstract

Reinforced concrete (RC) beams are fundamental flexural members in engineering structures. The mechanisms of flexural failure and shear failure differ significantly and directly affect structural safety design. This study systematically investigates the mechanical responses of RC beams with two distinct reinforcement ratios under flexural failure and shear failure. A BP neural network model based on the Bayesian regularization algorithm is developed to predict the failure mode and load state of RC beams. Experimental results of flexural failure indicate that in RC beams, the tensile steel yields first (at a load of 60 kN, the steel strain reaches 1500 µε), followed by concrete crushing and the formation of a plastic hinge. The load‑deflection curve exhibits a pronounced yield flow stage (deflection jumps from 10.10 mm to 15.15 mm), demonstrating ductile failure with early warning characteristics. In contrast, shear failure tests show that under a short shear span ratio, the beam achieves a higher load‑carrying capacity (up to 140 kN, approximately 2.2 times that of the flexural beam) through an arch action. However, the failure is sudden and brittle: as the load increases from 120 kN to 140 kN, the deflection jumps by 4.75 mm, accompanied by a sharp increase in steel strain. Furthermore, the developed BP neural network model takes concrete strain as input and load as output. The regression values for training and testing are all close to 1 (0.994 and 0.996 for the flexural model; 0.999 and 0.998 for the shear model), with small mean square errors. The results demonstrate that the neural network model can predict the load and failure stage of RC beams with high accuracy, providing a reliable basis for engineering design against flexure and shear.