Prediction of punching shear strength of RC flat plates using a refined machine learning model and a closed-form equation

Abstract

Reinforced concrete (RC) flat plate column-slab connections are highly vulnerable to brittle punching shear failure, which can trigger the progressive collapse of the entire structure. Accurate prediction of the punching shear strength is crucial for ensuring structural safety. However, this prediction remains challenging because this strength exhibits nonlinear dependencies on multiple input variables, along with significant correlations among them. To address these limitations, this study introduces a machine learning (ML) model and a symbolic regression (SR)-based data-driven empirical equation for evaluating the punching shear strength of RC flat plate connections without shear reinforcement. An expanded database comprising 472 experimental specimens was compiled, representing the larger collection of test data than previous studies. The predictive performance of the proposed ML model and SR-based closed-form equation was compared with that of existing design code equations and previously developed ML models using the collected data set in terms of accuracy, dispersion, and bias. Feature selection and Bayesian optimization were implemented to enhance the precision and robustness of the ML model. Validation results from the test dataset demonstrate that the proposed ML and SR models provide more accurate and consistent predictions compared to existing equations and ML models.