Failure mode classification for hybrid FRP/steel reinforced concrete beams: a soft computing concept based on the numerical model

The utilization of FRP bars in combination with steel bars in concrete structures is gaining increasing attention due to the advantages offered by both materials. Currently, there are no design standards providing guidance for practicing engineers despite the high relevance of this hybrid structural type, particularly in corrosive environments. The goal of this paper is to propose a soft computing approach and provide recommendations for predicting the failure modes of hybrid FRP/steel RC beams. A dataset of 158 experimental observations was prepared for training and testing six machine learning algorithms. The findings indicate that the failure mode prediction by the XGBoost model yields the best performance on the experimental dataset. Moreover, a synthetic dataset of 11,477 samples was generated from the proposed nonlinear model to evaluate the XGBoost model. It was found that even though trained on a small dataset, the classification performance on the synthetic dataset by XGBoost is awe-inspiring, with AUC scores for each class exceeding 0.99. The class label imbalance in both datasets is excellently handled by the proposed XGBoost model. The SHapley Additive exPlanations (SHAP) analysis reveals that material parameters such as the ultimate tensile strength, elastic modulus of FRP bar, compressive strength of concrete, and the amount of reinforcement have the most significant impact on the failure modes. In contrast, parameters related to the cross-sectional dimensions of the beam have less influence. Finally, a quick guide can be drawn from the research results to ensure both ductile failure and load-carrying capacity for design engineers.