Axial strength prediction of steel tube confined concrete columns using a hybrid machine learning model. (February 2022)

• Record Type: Journal Article
• Authors: Ngo, Ngoc-Tri; Pham, Thi-Phuong-Trang; Le, Hoang An; Nguyen, Quang-Trung; Nguyen, Thi-Thao-Nguyen
• Is Part Of: Structures. Volume 36(2022)
• Page Start: 765
• Page End: 780
• Publication Date: 2022-02
• Subjects: Machine learning -- Steel tube confined concrete -- Structural design -- Data-driven method -- Concrete structures; Structural engineering -- Periodicals; 624.1
• DOI: 10.1016/j.istruc.2021.12.054
• Languages: English
• ISSNs: 2352-0124

Abstract

Abstract: Estimating the axial strength of steel tube confined concrete (STCC) columns is challenging because it depends nonlinearly on the concrete compressive strength, the yield stress of steel, the column diameter ( D ), the thickness of steel tube ( t ), column length ( L ), D/t, and L/D . This study proposed an optimized hybrid machine learning(ML) model for accurately predicting the axial strength in STCC columns, which integrated support vector regression (SVR) and grey wolf optimization algorithm (GWO). Artificial neural networks (ANNs), SVR, linear regression, random forests (RF), and M5P rule were applied as baseline models. 136 samples of STCC columns infilled with various strength concrete were collected to develop and evaluate the proposed model. The results revealed that the proposed model was the most powerful compared to baseline models. Predicted data produced by the proposed model show the highest agreement with the actual data that confirmed its excellent performance in predicting the axial strength of STCC columns. Particularly, the mean absolute percentage error was 7.00% and the correlation coefficient was 0.992. Similarly, the mean absolute error by the proposed model was 143.47 kN which is the lowest value among 193.25 kN by the RF model, 217.03 kN by the M5P model, 450.00 kN by the SVR model, and 248.88 kN by the ANNs model. The SVR-GWO model improved more than 36 % in root-mean-square error compared to other ML models. This study contributes to (i) the state of the knowledge by examining the generalization and effectiveness of machine learning models in predicting the axial strength in STCC columns; and (ii) the state of practice by proposing an effective hybrid data-driven machine learning model to predict the axial strength in STCC columns which can support the service life of structures.