Machine Learning Models for Predicting Flexural Behavior of FRP-Strengthened RC Beams

Abstract

This study's objective is to overcome limitations in current design recommendations by exploring the application of machine learning to predict the flexural behavior of fiber-reinforced polymer (FRP)-strengthened reinforced concrete beams. Although FRP composites have completely changed structural strengthening, it might be challenging to predict bending moments with precision. This work fills the theoretical and experimental findings gaps by utilizing Artificial Neural Network (ANN) models in conjunction with computational techniques and statistical analysis. It includes gathering data, conducting a thorough literature review, and developing three models: Artificial neural network (ANN), Non-linear Regression (NLR), and Linear Regression (LR). Despite other models, the ANN model stands out for its superior performance and accurate predictions. Understanding material characteristics, FRP properties, and beam dimensions is critical in predicting flexural strength. The most significant parameter studied in this research is the overall depth of the beam (h), followed by the variation in bottom flexural reinforcement ( ρ s ). Additionally, the FRP ratio ( ρ f ) and beam width ( b ), which are both regarded as significant attributes, influence the flexural capacity of FRP-strengthened beams. The ultimate moment (M u ) may be predicted by the ANN model with an error range of -20% to +15%, indicating a significant advancement in strengthening approach optimization. This development could reduce the requirement for expensive experimental testing during construction, thereby enhancing the predictive capacity of structural engineering procedures. Furthermore, the design of flexurally strengthened RC beams with FRP may be made possible by depending on this model, specifically the ANN, without the need for experimental effort.