Prediction of Punching Shear Capacity for Fiber-Reinforced Polymer Concrete Slabs Using Machine Learning

The punching shear capacity of slab structures is a critical design parameter. The existing specifications in the majority of international reinforced concrete design standards for slab punching shear capacity are based on experiments of steel reinforced slabs. For fiber-reinforced polymer concrete slabs in particular, these conventional design methods may be insufficient to effectively estimate their punching shear capacity and the interaction of many influencing variables impacting punching shear capacity. In this study, several linear regression models and machine learning algorithms, namely support vector machine (SVM) models, were applied to predict FRP slabs' punching shear capacity accurately. A dataset of 103 points was gathered from experimental studies in literature and was used to train the models. The performance of each utilized model was assessed based on coefficient of determination (R
²
), root mean square error (RMSE), and mean absolute error (MAE) of the predicted punching shear capacity. The results demonstrated that the cubic SVM model outperformed all other models used in this study, with an R
²
value of 0.95, RMSE of 74.61 kN, and MAE of 49.17 kN. Then, a sensitivity study was conducted, providing valuable insights on the influence of selected variables, namely slab effective depth (d), loading area (A), slab length (L), and concrete compressive strength (fc') on the punching shear capacity of FRP slabs. The results suggested that increasing the concrete compressive strength in FRP concrete slabs is advised at high d values (d > 145 mm), as increasing the fc' is more pronounced at higher slab effective depths. Moreover, a more pronounced effect at d > 145 mm was observed. Increasing the loading area reduces the punching shear capacity to a loading area/effective depth ratio of around 400. Above the ratio 400, increasing loading area increases the punching shear capacity.