Machine learning models to predict the punching shear strength of reinforced concrete flat slabs

Modelos de aprendizado de máquina para previsão da resistência à punção de lajes lisas de concreto armado

abstract:

Punching shear failure is caused by shear stress concentration in the slab-column connection of reinforced concrete flat slabs. As it is a brittle failure, it is crucial to understand how this mechanism works and to correctly predict the resistance of slabs subjected to it. In this paper, machine learning-based models were developed and compared to predict the punching shear resistance of reinforced concrete interior slabs without shear reinforcement. The models were based on 373 experimental data of interior slabs. Artificial neural network, decision tree, random forest and extreme gradient boosting algorithms were employed. The input variables considered herein were the effective depth of the slabs, flexural reinforcement ratio, effective width of the columns, concrete compressive strength and steel yield strength, and the target variable was the punching shear strength. The results for the punching shear resistances obtained by the developed models, as well as those obtained by employing models presented in five reinforced concrete design codes, were compared to the experimental data. All machine learning models showed coefficient of determination above 0.95 for test data. As for the design code models, large discrepancies were observed between them, with the Brazilian code showing more accuracy than the others in predicting the failure load of the slabs.

Keywords:
predictive models; structural safety; design provisions; computational modeling; XGBoost