Predictions of mechanical properties of Fiber Reinforced Concrete using ensemble learning models

Fiber Reinforced Concrete (FRC) significantly improves the tensile, crack resistance, and durability of concrete by adding fiber materials, making it highly promising for applications in structural reinforcement, repair and new construction projects. The strength of FRC is a crucial indicator of its mechanical properties. This study utilizes ensemble learning techniques to predict and analyze the compressive strength, splitting tensile strength, and flexural strength of FRC. A total of 1589 sets of compressive strength data, 1137 sets of splitting tensile strength data, and 1061 sets of flexural strength data were collected from existing literature. Four ensemble models (GBDT, Xgboost, LightGBM, RF) were used to establish prediction models and analyze influencing factors for these three FRC mechanical properties, and the results were compared with traditional models. The results show that the ensemble models can effectively solve the FRC strength prediction problems compared to traditional models. The GBDT regression model performed best in predicting compressive and flexural strength, with R2 values of 0.939 and 0.867, respectively. The Xgboost regression model performed best in predicting splitting tensile strength, with an R2 value of 0.944. Further analysis using SHAP revealed that cement and water significantly impact FRC strength. The inclusion of fibers significantly affects the splitting tensile strength and flexural strength of FRC but has a minimal impact on compressive strength. The study not only demonstrates the feasibility of using machine learning algorithms for FRC strength prediction but also provides a fast and reliable means for FRC strength prediction, offering more reliable data support for the design and application of FRC materials, aiding practical engineering applications. •A large-scale strength database for FRC has been established.•Models have been developed for the predictions of the strengths of FRC.•Compared to traditional models, the ensemble models exhibit higher accuracy.•SHAP analysis reveals the internal principle of the models.