Design of concrete-filled steel tubular columns using data-driven methods

By leveraging the merits of structural steel and concrete materials, concrete-filled steel tubular (CFST) structures have been increasingly used in the composite construction of bridges and high-rise buildings. However, their design equations are more complicated than those of steel and reinforced c...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Journal of constructional steel research Ročník 200; s. 107653
Hlavní autoři: Degtyarev, Vitaliy V., Thai, Huu-Tai
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier Ltd 01.01.2023
Témata:
Boosting algorithms
CFST columns
Machine learning
Reliability analysis
Resistance reduction factor
Structural design
Structural design
Reliability analysis
Resistance reduction factor
CFST columns
Machine learning
Boosting algorithms
ISSN:0143-974X, 1873-5983
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:By leveraging the merits of structural steel and concrete materials, concrete-filled steel tubular (CFST) structures have been increasingly used in the composite construction of bridges and high-rise buildings. However, their design equations are more complicated than those of steel and reinforced concrete (RC) structures, especially for circular columns under eccentric loading. Therefore, the use of emerging data-driven approaches will help structural engineers ease the design process. This paper explores the use of data-driven design methods as alternatives to conventional mechanics-based design models. Five boosting algorithms, including adaptive boosting (AdaBoost), gradient boosting machine (GBR), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical gradient boosting (CatBoost), are employed to develop predictive models for four different types of CFST columns (i.e., circular columns, rectangular columns, circular beam–columns, and rectangular beam–columns). These predictive models are trained using the most up-to-date and comprehensive database collected from over 3,200 test specimens. Reliability analysis is conducted to calibrate the resistance reduction factors for three different design frameworks (i.e., the US, Eurocode, and Australian frameworks) to ensure that the newly developed predictive models meet the target reliability indices required by different design frameworks. A web-based design tool is also developed to promote the practical use of data-driven methods for the design of CFST columns. •ML models for predicting the resistance of CFST columns and beam–columns were created.•The models were based on five boosting algorithms and a 3,208-test database.•The models outperformed the design provisions of AISC 360, EC 4, and AS/NZS 2327.•Resistance factors were determined from reliability analyses for three design frameworks.•A web application based on the ML models was created and deployed to the cloud.
ISSN:0143-974X
1873-5983
DOI:10.1016/j.jcsr.2022.107653