Developing machine learning model to estimate the shear capacity for RC beams with stirrups using standard building codes

Abstract

Shear failure in reinforced concrete (RC) beams with a brittle nature is a serious safety concern. Due to the inadequate description of the phenomenology of shear resistance (the shear behavior of RC beams), several of the existing shear design equations for RC beams with stirrups have high uncertainty. Therefore, the predicted models with higher accuracy and lower variability are critical for the shear design of RC beams with stirrups. To predict the ultimate shear strength of RC beams with stirrups, machine learning (ML)-based models are proposed in the present research. The models were created using a database of 201 experimental RC beams with stirrups gathered from earlier investigations for training and testing of the ML method, with 70% of the data being used for model training and the rest for testing. The performance of suggested models was evaluated using statistical comparisons between experimental results and state-of-the-art current shear design models (ACI 318–08, Canadian code, GB 510010–2010, NZS 3101, BNBC 2015). The suggested machine learning-based models are consistent with experimentally observed shear strength and current predictive models, but they are more accurate and impartial. To understand the model very well, sensitivity analysis is determining as input values for a specific variable affect the outcomes of a mathematical model. To compare the results with different machine learning models in training and testing R2 , RMSE and MSE are also established. Finally, proposed ML models such as gradient boost regressor and random forest give higher accuracy to evaluate the shear strength of the reinforcement concrete beam using stirrups.