EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Machine Learning-Driven Flexural Performance Prediction and Experimental Investigation of Glass Fiber-Reinforced Polymer Bar-Reinforced Concrete Beams

Muhammet Karabulut iD

Polymers · 07.03.2025 · 17 (6) · 713

DOI: 10.3390/polym17060713 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Polymers, online erschienen am 07.03.2025. Verfasst von Muhammet Karabulut. Beteiligte Einrichtung: Department of Structural Mechanics and Analysis, Technical University of Berlin, 10623 Berlin, Germany. Die Arbeit wurde bisher 23-mal zitiert (Crossref) und nennt 45 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study experimentally examines the flexural performance, crack formation patterns, and failure mechanisms of glass fiber-reinforced polymer (GFRP) bar-reinforced concrete beams with varying concrete compressive strengths (low, moderate, and high), addressing a gap in the current literature. Furthermore, it employs an innovative machine learning approach to enhance analysis. Nine RC beams reinforced with GFRP bars, having concrete compressive strengths of low (CC20), moderate (CC30), and high (CC40), each measuring 150 × 200 × 1100 mm, were fabricated and tested under three-point bending conditions. Through the integration of three-point bending tests and machine learning-based prediction models, this study connects experimental findings with advanced analytical approaches. One of the key innovations in this study is the use of eighteen ML regression models implemented with Python’s PyCaret library, achieving an impressive average prediction accuracy of 91.5% for RC beam deflection values. In particular, the Ada Boost Regressor and Gradient Boosting Regressor models performed exceptionally well on GFRP bar-reinforced concrete beams, providing the highest number of consistent and highly accurate predictions, making them very useful tools for GFRP bar-reinforced beam ultimate load-carrying capacity/deflection predictions. The outcomes identified clear failure mechanisms: RC beams with CC20, CC30, and CC40 concrete compressive strengths typically developed a single, large flexural crack at the midpoint. Although the ultimate load-carrying capacity of GFRP bar RC beams improved with higher concrete compressive strength, CC20 and CC30 beams displayed more ductile failure behavior than CC40 beams. The ultimate load-carrying capacity of CC40 RC beams was determined to be approximately 74% higher than that of CC20 RC beams. Regardless of the concrete compressive strength class, the absence of shear cracks and the prevention of sudden failure under bending in GFRP bar-reinforced concrete beams are considered major advantages of using GFRP bar reinforcement.

Schlagwörter

Stichwörter der Autor:innen

ada boost regressorbrittleductileflexural crackglass fiber reinforced polymer (GFRP) bargradient boosting regressormachine learning (ML)reinforced concrete beamshear crackthree-point bending test

Autor:innen und Institutionen

  • Department of Structural Mechanics and Analysis, Technical University of Berlin, 10623 Berlin, Germany.
  • Department of Civil Engineering, Zonguldak Bulent Ecevit University, 67100 Zonguldak, Turkey.

ORCID-Profile: Muhammet Karabulut

Zitieren

APA

Karabulut, M. (2025). Machine Learning-Driven Flexural Performance Prediction and Experimental Investigation of Glass Fiber-Reinforced Polymer Bar-Reinforced Concrete Beams. Polymers, 17(6), 713. https://doi.org/10.3390/polym17060713

Vancouver

Karabulut M. Machine Learning-Driven Flexural Performance Prediction and Experimental Investigation of Glass Fiber-Reinforced Polymer Bar-Reinforced Concrete Beams. Polymers. 2025;17(6):713. doi:10.3390/polym17060713

BibTeX
@article{Karabulut2025,
  author = {Karabulut, Muhammet},
  title = {Machine Learning-Driven Flexural Performance Prediction and Experimental Investigation of Glass Fiber-Reinforced Polymer Bar-Reinforced Concrete Beams},
  journal = {Polymers},
  year = {2025},
  volume = {17},
  number = {6},
  pages = {713},
  doi = {10.3390/polym17060713}
}
Kostenlos starten →