A prediction model for flexural strength of corroded prestressed concrete beam using artificial neural network

The prestressed concrete structures are taking the forefront in recent years due to the innovations in the construction industry. However, corrosion is one of the barriers to the serviceability of the prestressed structures. Therefore, a detailed investigation of the prestressed concrete structure under a corrosive environment is essential. This study uses Resilient Back Propagation with BackTracking Neural Network (RBPBTNN) to estimate the flexural strength of the corroded prestressed concrete beam. Three RBPBTNN-based prediction models are proposed to predict the ultimate load, ultimate moment and deflection. The datasets involving multiple influencing parameters are collected from experimentally verified literature. The best possible RMSE and R² values obtained during the training phase for ultimate load prediction are 3.2834 and 0.9964 and for ultimate moment prediction are 2.6128 and 0.9987 and for deflection prediction are 0.8252 and 0.9992 when K-fold cross-validation is three and training repetition is ten. The final performance measures (MAE, R², RMSE etc) of the prediction results are presented in comparison with other artificial neural network algorithms and it is found that the proposed models are the best fit for the collected datasets.