ANÁLISE ESTATÍSTICA DO ERRO DE MODELO DE LAJES DE CONCRETO ARMADO SUBMETIDAS À PUNÇÃO

Abstract

This paper statistically analysis the model error associated with the theoretical models presented in the standards ABNT NBR 6118 (2023), ACI318 (2022), EUROCODE2 (2023), and FIB MC2020 (2024). Model error is defined as the ratio between experimental results and predictions by the normative criteria. The analysis considers the ultimate limit state for punching of reinforced concrete flat slabs without shear reinforcement, based on 218 experimental results from the literature. Partial safety coefficients were disregarded, allowing the estimation of mean ultimate punching resistance. The normative model results were compared to experimental data, and statistical error metrics including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Coefficient of Determination (R²), and Normalized Root Mean Square Error (NRMSE) were calculated to assess accuracy and variability. Among the standards, FIB MC2020 showed the closest agreement with experimental results, exhibiting the lowest MAE, RMSE, and NRMSE, and a high R², with a mean model error of 0.997. In contrast, ACI318, NBR 6118, and EUROCODE2 underestimated punching loads on average by 30.31%, 15.54%, and 7.32%, respectively, with higher MAE, RMSE, and NRMSE values. These findings indicate that FIB MC2020 provides the most accurate estimation of punching resistance, while ACI318 tends to significantly underestimate slab strength. The study highlights the importance of selecting normative criteria that reflect experimental behaviour for safe and reliable structural design.