A deep learning model for the punching shear strength of prestressed concrete slabs

Abstract

Abstract Failure of concrete slabs in shear is a devastating incident which could result in catastrophic disasters suddenly without warning. Though, the use of prestressing in concrete slabs reduces the warning even further, thus leading to an increased probability of causalities. For several decades, researchers have investigated this phenomenon. However, so far, no robust physical model capable of simulating such complicated behavior exist. Thus, with the increase of experimental testing for concrete slabs under punching shear, the machine learning method is a viable option for almost accurate prediction of the concrete slab capacity. In this study, Decision Tree Regression (DTR) and Deep Neural Network (DNN) as machine learning models were developed and evaluated to explore the applicability of predicting the capacity of concrete slabs subjected to punching shear. The findings reveal that the DNN model achieves a high prediction accuracy of 95.3% for the Coefficient of Determination $$r^2$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msup> <mml:mi>r</mml:mi> <mml:mn>2</mml:mn> </mml:msup> </mml:math> and demonstrates a lower mean absolute error (MAE) value. Based on the analysis results, the prestressing vertical component and the prestressing stress were found to be the most influential on the punching shear capacity after the depth. Also, the proposed DNN model best emphasises the representation of the experimentally observed patterns.