Machine learning-based models for predicting the shear strength of synthetic fiber reinforced concrete beams without stirrups

Abstract

Stirrups can be added to reinforced concrete (RC) beams made of plain concrete to increase the shear strength providing better performance against embrittlement. If the need for shear stress is less than the strength of RC beams, reinforced with the minimum number of stirrups, then this choice might not be the most cost-effective. Synthetic Fiber Reinforced Concrete (SyFRC) in Reinforced Concrete (RC) structures is limited due to the lack of predicting models for this emerging material in current literature. However, recent advances in concrete technology have revealed more significant benefits of SyFRC, including high deformation capacity and less corrosive concrete. This study predicts shear strength of SyFRC beams without stirrups using the ACI 318-19 equation in addition to Machine Learning (ML) methods such as LightGBM, XGBoost, and Gene Expression (GEP). A database of 102 tested SyFRC specimens were compiled, processed, and evaluated. With R2 values of 98.91% and 97.22%, respectively, the LightGBM and XGBoost outperformed the other examined algorithms by having the least predictions discrepancy and highest accuracy values. As the ACI 318-19 equation does not account for the fibers volume ratio and shear span-to-depth ratio effects on the shear strength contribution, it projected shear strength with the lowest degree of accuracy, with R2 = 75.5%. The feature importance analysis revealed that these two factors, in addition to the effective beam depth, beam width, longitudinal steel reinforcement ratio, and concrete compressive strength should not be negligible in shear strength prediction. For forecasting the shear strength, a closed-form GEP-based model was suggested. The proposed GEP model has a little lower prediction accuracy with R2 = 88.4%. The performance of the four examined models was evaluated from various perspectives. The analysis shows that, apart from the ACI equation, all considered models effectively predict the effects of shear span-to-depth ratio. This is critical for investigating deep and slender beams, and size effect, which are critical for beams with high effective depths. The current study's findings should give practitioners a solid platform for making precise and straightforward assessments of shear strength.