Prediction and analysis of punching shear capacity in steel fiber reinforced concrete slab using machine learning

The study of punching shear capacity in steel fiber reinforced concrete (SFRC) slabs is significant for enhancing the safety of slab-type structures. It helps engineers optimize structural designs and ensures the stability of buildings under various loading conditions. Although numerous models have been proposed to predict punching shear capacity, their limited consideration of influencing factors often results in low prediction accuracy, making them insufficient for practical engineering applications. This study proposes a Backpropagation Neural Network (BPNN) model optimized by the Grey Wolf Optimizer (GWO) to improve prediction performance. By optimizing the initial weights and biases of the network, the model achieves faster convergence and higher predictive accuracy, effectively capturing the complex nonlinear relationships among input variables. The input features include slab thickness, effective depth, loading pad length, and concrete compressive strength, among others, and the output variable is the punching shear capacity of the slab. A dataset of 144 experimental samples was used for training and validation. To enhance the interpretability of the model, the SHapley Additive Explanations (SHAP) method was introduced to quantify the contribution of each input variable to the model's output. The results indicate that when the slab thickness exceeds 120 mm, the compressive strength is above 50 MPa, and the steel fiber volume fraction is greater than 1.0 %, the punching shear performance of SFRC slabs is significantly enhanced. In contrast, when the slab thickness is less than 80 mm, and the compressive strength is below 30 MPa, the strengthening effect of steel fibers is minimal. The proposed BPNN-GWO model achieved high prediction accuracy on the validation set, with a determination coefficient (R²) of 0.987 and a root mean square error (RMSE) of 12.64, both of which are significantly better than those of traditional empirical models, demonstrating the applicability and superiority of the proposed method in predicting the punching shear capacity of SFRC slabs.