Improved BPNN models based on different algorithms to predict the flexural capacity of corroded RC beams

Abstract

Reinforced concrete (RC) beams used in marine and saline-alkaline environments experience a decline in their load-bearing capacity over time because of environmental corrosive agents. This is a complex and time-consuming process, and the corrosion of the steel reinforcements is uneven. Hence, accurately determining the actual flexural capacity of corroded RC beams is challenging. This study predicts the flexural capacity of corroded RC beams using an improved back propagation neural network (BPNN) model with various algorithms, including Cuckoo Search (CS), Seagull optimization algorithm (SOA), whale optimization algorithm (WOA), and particle swarm optimization (PSO). The study aims to develop a robust predictive model for the flexural capacity of corroded RC beams. The dataset comprised 228 samples; ten parameters, including the geometric and mechanical properties of the RC beams, reinforcement parameters, and corrosion rate, were selected to predict the flexural capacity of the corroded RC beams. For the evaluation of the performance of the improved models, following indicators such as mean absolute error (MAE), root mean square error (RMSE), over-fitting analysis (OFA) and R� were used. In addition, k-fold cross-validation is used to further verify the model�s accuracy. Comparing the performance metrics of each model showed that the PSO-BP model has the highest R� and the lowest RMSE and MAE values, which are 0.98, 2.979 kN�m, and 1.913 kN�m, respectively, indicating a strong correlation between the predicted results and experimental values. The accuracy and reliability of the PSO-BP model were further validated through visual graphical results such as Taylor diagrams, violin plots, and multiple histograms. Sensitivity analysis indicated that the area of the tensile reinforcement, effective depth of the beam cross-section, and corrosion rate of the longitudinal reinforcement significantly affected the flexural strength of the corroded RC beams. When determining a reinforcement scheme for corroded RC structures, using machine learning to predict the flexural capacity of RC beams can significantly reduce time, labor, and resources.