A Hybrid Data-Driven and Physics-Based Approach for Shear Capacity Estimation of Steel Fiber Reinforced Concrete Beams

Abstract

Estimation of shear strength of steel fiber reinforced concrete (SFRC) beams is difficult because of the complex nonlinear interactions between fiber, geometry, and reinforcements. In this study, the hybrid machine learning (ML) and finite element analysis (FEA) method are utilized for estimating the load-carrying capacity (Pmax) of SFRC beams. A database of 257 SFRC beams was collected from literature and was pre-processed. Seven ML methods, including Decision Tree, AdaBoost, Linear Regression, k-Nearest Neighbors, Random Forest, Stochastic Gradient Descent, and Gradient Boosting, were applied on the database through the k-fold cross-validation technique. The best model was selected as Gradient Boosting due to R² > 0.93. By conducting SHAP analysis, longitudinal reinforcement ratio, height of the beam, and shear span-to-depth ratio were found to be the most important parameters. Using nonlinear FEA in Abaqus, the results of ML models were validated by comparing them with experimental and simulation results (∆ ≈ 4.79% and ∆ ≈ 0.14%, respectively). Through parametric studies, the positive influence of the volume fraction of fibers and reinforcement ratio and negative influence of shear span-to-depth ratio were determined. The suggested ML-FEA approach can be considered accurate and efficient for practical SFRC shear design.