Hybrid FRP strengthening of reinforced concrete deep beams: Experimental, theoretical and machine learning-based study

Abstract

This paper presents experimental findings from testing seventeen reinforced concrete deep beams, categorized into four groups based on the presence and type of openings. A novel and cost-effective hybrid strengthening scheme is proposed comprising glass chopped mat sheets and eco-friendly basalt FRP sheets (GF-BFRP). Group 1 consisted of solid beams without openings, while Group 2 included beams with circular openings, Group 3 with square openings, and Group 4 with rectangular openings of varying dimensions. Each group comprised beams tested in various strengthening configurations using GF-BFRP layers with and without anchor support. Analysis of failure modes revealed initial flexural cracking in control beams, with beams containing openings exhibiting diagonal cracking and reduced shear capacity. Results revealed that beams with openings experienced a significant reduction in shear capacity. Circular, square, and rectangular openings reduced peak capacity by 26.11 %, 30.67 %, and 31.91 %, respectively, while rectangular openings oriented vertically caused the most substantial reduction at 47.46 %. Strengthening using a single GF-BFRP sheet led to debonding, which was mitigated by anchors, enhancing confinement and reducing diagonal cracking. However, strengthened beams did not recover the original strength of the solid beam, which reached a peak load of 245.51 kN. For instance, the C-W1-A beam achieved a peak load of 173.58 kN, which was 4.31 % lower than its control beam due to the extensive anchor installation. Evaluation of predictive models for shear capacity highlighted discrepancies. None of the existing codes provide expressions that account for the shear contributions of externally bonded FRP systems on beams with opening shape and size implicitly defined. To overcome this issue, machine learning approaches were utilized, employing gradient boosting regression and random forest methods. Data on deep beams, both with and without openings (and without strengthening), was collected from eight studies. The models were trained on this dataset, and predictions were made based on the results of this study. While the gradient boosting regression model tended to overestimate the peak capacity of the deep beams, the random forest model provided predictions that were much closer to the experimental results.