An Extreme Gradient Boosting approach to estimate the shear strength of FRP reinforced concrete beams. (November 2022)

• Record Type: Journal Article
• Authors: Le, Hoang-Anh; Le, Duc-Anh; Le, Thanh-Tung; Le, Hoai-Phuong; Le, Thanh-Hai; Hoang, Huong-Giang Thi; Nguyen, Thuy-Anh
• Is Part Of: Structures. Volume 45(2022)
• Page Start: 1307
• Page End: 1321
• Publication Date: 2022-11
• Subjects: AFRP Aramid Fiber Reinforced Polymer -- AI Artificial Intelligence -- BFRP Basalt Fiber Reinforced Polymer -- CFRP Carbon Fiber Reinforced Polymer -- CV Cross-Validation -- η Learning rate -- FRP Fiber Reinforced Polymer -- FRPRC Fiber Reinforced Polymer Reinforced Concrete -- GFRP Glass Fiber Reinforced Polymer -- MAE Mean Absolute Error -- MAPE Mean Absolute Percentage Error -- mcw min child weight -- ML Machine learning -- R2 Coefficient of determination -- RC Reinforced Concrete -- RMSE Root mean square error -- SHAP SHapley Additive exPlanations -- SS Shear Strength -- VFRP Vinyl Fiber Reinforced Polymer -- Vu Shears Strength -- XGB Extreme Gradient Boosting; Machine learning -- Extreme Gradient Boosting (XGB) -- FRP reinforced concrete beams -- Shear strength; Structural engineering -- Periodicals; 624.1
• DOI: 10.1016/j.istruc.2022.09.112
• Languages: English
• ISSNs: 2352-0124

Abstract

Abstract: Although fiber-reinforced polymer (FRP) bars have gained much attention recently, their material characteristics, particularly the elastic modulus, are markedly different from conventional reinforcement. This explains why classical calculation models cannot correctly capture the behaviors of FRP-reinforced concrete (RC) beams in structural engineering. As a novel approach, this study aims to construct a machine learning (ML) model for predicting the shear strength (SS) of FRP-RC beams with and without stirrups. An Extreme Gradient Boosting (XGB) model is selected thanks to its effectiveness and robustness in tackling many complicated problems. An experimental database encompassing 453 examples is utilized, with input variables covering the FRP-RC beams' geometry, the mechanical properties of concrete, and two reinforcement components. The selection of hyperparameters of XGB model is first conducted, followed by a prediction performance assessment using common statistical measures. Additionally, this study conducts feature importance analysis and employs SHAP values technique to discover the inputs versus output relationship, as well as to identify the most significant features affecting the SS of FRP-RC beams. The outcome of this research, which lies in developing a reliable and accurate ML model that can handle both FRP-RC beams with and without stirrups, may favor the application of ML models to various real-world engineering problems.