Hybrid regression and machine learning model for predicting ultimate condition of FRP-confined concrete

Behrooz Keshtegar, Aliakbar Gholampour, Duc Kien Thai, Osman Taylan, Nguyen Thoi Trung

Research output: Contribution to journalArticlepeer-review

Abstract

The accurate design-oriented model for concrete confined with fiber-reinforced polymer (FRP) is important to provide safe design of this composite system. In this paper, the response surface model (RSM) is coupled with support vector regression (SVR) for developing a novel hybrid model, namely RSM-SVR, with the aim of predicting the ultimate condition of FRP-confined concrete. Predictions obtained by the proposed model were compared with those by six empirical models and two data-driven models of RSM and SVR for database containing 780-test column results with circular cross section. Statistical analysis reveals that the proposed RSM-SVR model predicts the compressive strength and corresponding axial strain of the concrete confined with FRPs more accurately in comparison with the existing models. The results also show that RSM-SVR and SVR models provide stable predictions of strength and strain enhancement ratios for lateral confining ratio of >1 while the other models exhibit chaotic model error. The high accuracy and stable predictions by the proposed model are achieved based on its high flexibility and robustness in capturing the effect of lateral confining pressure as the interaction between the concrete core and FRP jacket in comparison with the existing models.

Original languageEnglish
Article number113644
Number of pages12
JournalCOMPOSITE STRUCTURES
Volume262
DOIs
Publication statusPublished - 15 Apr 2021
Externally publishedYes

Keywords

  • Fiber-reinforced polymer (FRP)
  • FRP-confined concrete
  • Hybrid model
  • Response surface method (RSM)
  • Support vector regression (SVR)

Fingerprint Dive into the research topics of 'Hybrid regression and machine learning model for predicting ultimate condition of FRP-confined concrete'. Together they form a unique fingerprint.

Cite this