Flexural strength prediction of steel fiber reinforced concrete with machine learning

Abstract

Steel fiber-reinforced concrete (SFRC) is a commonly used concrete in structures that need to have better cracking performance and flexural behaviour. The performance of SFRC depends on the concentration of fiber content, geometry and interaction with cementitious matrix. This study examines the mechanical behaviour of SFRC using hooked-end fibers at volume fractions of 0.25%, 0.50%, and 0.75%. Results show that compressive strength was increased at 0.25% fiber volume. The flexural strength improved with low and medium fiber dosage but declined with high contents. These trends demonstrate that the outcomes of experiments are very sensitive to mixture proportions and fiber properties. Due to this complexity, it becomes hard and time-consuming to achieve flexural strength prediction by experiments only. To overcome this shortcoming, the modelling of flexural strength behaviour was done using machine learning. The outcome of the experimental results was combined with published data to compile a dataset of 818 mixes. Three supervised models namely XGBoost, multilayer perceptron, and k-nearest neighbour were trained using k-fold cross-validation. XGBoost performed best and had the highest R2 value of 0.91 compared to the other models tested. The SHAP analysis also established that fiber volume and aspect ratio contributed most to the predicted strength. The combination of these two approaches demonstrates that flexural strength of SFRC may be forecasted using data-driven models with high reliability. It also minimises the reliance on the large-scale laboratory testing which makes the assessment process more effective. Moreover, the developed models was deployed as a web-based application for practical implementation.