Numerical simulation and data-driven analysis on the flexural performance of steel reinforced concrete composite members

• Published in: Engineering structures Vol. 247; p. 113200
• Main Authors: Lai, Bing-Lin, Yang, Lifu, Xiong, Ming-Xiang
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 15.11.2021; Elsevier BV
• Subjects: Algorithms; Artificial Neutral Network (ANN); Composite materials; Ductility; Fiber section analysis; Flexural capacity; Flexural stiffness; Learning algorithms; Limit state design; Machine learning; Machine Learning (ML); Material properties; Mathematical models; Numerical models; Numerical prediction; Performance indices; Reinforced concrete; Reinforcing steels; Robustness (mathematics); Steel; Steel Reinforced Concrete (SRC) members; Stiffness; Fiber section analysis; Steel Reinforced Concrete (SRC) members; Artificial Neutral Network (ANN); Machine Learning (ML); Flexural stiffness; Flexural capacity
• ISSN: 0141-0296; 1873-7323
• DOI: 10.1016/j.engstruct.2021.113200

•A dataset containing 27 SRC beam test is established from literature.•A numerical model based on fiber section analysis is built to predict SRC beam flexural behavior.•Parametric study is conducted to evaluate influences of different design parameters.•Various design codes are evaluated in estimating flexural characteristics.•Artificial neural network method is proposed to predict flexural characteristics. The flexural stiffness and flexural capacity of composite member are decisive indexes for the serviceability limit state and ultimate limit state design. To deepen the understanding of the flexural behavior of Steel Reinforced Concrete (SRC) composite members, a computationally efficient numerical technique based on fiber section analysis is employed with discussion of concrete confinement effect and steel section placement location. A dataset of SRC flexural members from published literature, covering a wide spectrum of geometrical and material properties, was built and compared with numerical predictions. The comparison signifies that the proposed numerical model is able to reproduce the moment-deflection relation of SRC flexural members, giving close estimation of the flexural stiffness, flexural capacity and flexural ductility. Parametric study is then carried out using the validated numerical model by varying concrete class, steel grade, steel section size and other influential parameters. Based on the numerically generated database, Machine Learning (ML) analysis is performed through an Artificial Neutral Network (ANN) algorithm. Compared with current design codes, the ML technique efficiently predicts the flexural capacity and flexural stiffness with high level of accuracy and robustness. As a consequence, it can be employed to intelligently estimate the flexural characteristics of SRC flexural members.