Chapter 7: Machine learning-driven approach to understanding punching shear design in steel fibre-reinforced slabs

Predicting the punching shear strength of steel fibre-reinforced concrete (SFRC) slabs is crucial to the design process and structural safety. Traditional design codes, such as ACI 318-19 (ACI, 2019), Canadian Standards Association (CSA) and Eurocode 2 (BSI, 2004), often provide conservatively high estimates that may not accurately reflect the performance of SFRC slabs in practice.

This chapter aims to evaluate the performance of machine learning (ML) techniques, specifically the gradient boosting regression (GBR), random forest regression (RFR) and k-nearest neighbours (k-NN) models, for the precise prediction of punching shear strength. An ML approach was developed which involved training and testing models using a dataset of SFRC slab parameters and then comparing their performance with conventional empirical and existing analytical methods.

The results indicated that the GBR model produced the most accurate predictions compared to the other ML models and traditional analytical methods, exhibiting a significantly higher coefficient of determination (R²), lower mean absolute error (MAE) and root mean squared error (RMSE). Additionally, the ML models demonstrated effectiveness in providing non-conservative prediction results compared to traditional codes. The study found that ML models could accurately predict punching shear strength in SFRC slabs. Future research should integrate ML with traditional methods to balance prediction accuracy and safety.