PSI - Issue 70

N. Vignesh Kumar et al. / Procedia Structural Integrity 70 (2025) 453–460

455

(Hassanein et al., 2023). A comprehensive review examined 88 models for predicting the behavior of FRP-confined concrete, evaluating their performance against a large experimental database. The analysis unveiled significant variations in the accuracy of prediction across various models, emphasizing the challenges in developing an analytical formulation to be acknowledged by all (Cui et al., 2024; Li et al., 2023). In recent years, ML techniques have gained popularity as a reliable alternative for predicting the FRP confined concrete strength. The use of Artificial Neural Networks has shown promising results in predicting the axial stress-strain behavior of FRP-confined concrete elements, outperforming traditional modeling approaches (Zarringol et al., 2023). Studies have leveraged Support Vector Regression techniques to predict the compressive strength of FRP-confined concrete, demonstrating favorable agreement with experimental observations (Amin Mozumder et al., 2016). Comparative studies of various ML approaches for concrete strength prediction have been conducted in recent decades. Studies have compared various regression analysis, artificial neural networks, support vector machines, and ensemble models for predicting concrete strength, concluding that tree-based models generally outperform individual models (Yue et al., 2024). However, similar comparisons for FRP confined concrete strength prediction are limited in the review. Developments in ML for FRP-confined concrete have expanded the available techniques. Ensemble learning with uncertainty quantification provides more predictions and reliability metrics for design (Shang et al., 2025). Interpretable machine learning models offer balanced accuracy and physical consistency, addressing concerns about black box models (Atf et al., 2025). Transfer learning allows models to train on one FRP type to adapt to other FRP type. Researchers have used comprehensive database of FRP-confined concrete properties and test results to develop and validate predictive models. This dataset is leverages in the present study, using a pre-processed subset to ensure high-quality data for ML model development (Realfonzo & Napoli, 2011). The present study banks upon existing research by providing a comprehensive comparison of five different ML approaches, including the relative undermined Gaussian Process Regression approach, for FRP confined concrete strength prediction. In addition to that, sensitivity analysis is conducted in this study that offers invaluable insights into the influential parameters which can aid both model development and practical design considerations.

Table 1. Statistical characteristics of the dataset parameters for CFRP datapoints. Parameter Minimum Maximum

Mean

Standard Deviation

Diameter (mm)

51

406 4.01

144.74

30.86

Height to Diameter ratio

1.97

2.07

0.35

Unconfined Concrete Strength (MPa)

15.20

169.70

44.61

21.26

FRP thickness (mm)

0.09

3.00

0.62

0.65

FRP tensile strength (MPa)

108.50

1392.77 303.60

860.06

355.81

Confined Concrete Strength (MPa)

32.90

83.43

41.29

3 Methodology 3.1 Database and Preprocessing

This study uses the experimental database compiled by Realfonzo and Napoli (2011), aggregating the results of 450 FRP-confined concrete cylinders from over 30 research works. This literature compiled database, not generated by authors, provide diverse experimental scenarios. Post preprocessing to remove inconsistent or incomplete data and retaining only FRP rupture failures, the final database comprised 277 datapoints (183 CFRP, 94 GFRP). Key parameter statistics are provided in Tables 1 & 2. Input variables included cylinder diameter (D), height-to-diameter ratio (H/D), unconfined concrete strength (fco), FRP thickness (tfrp), and FRP tensile strength (ffrp), calculated using strain efficiency factor ke = 0.31. The target variable was confined concrete strength (fcc). Post to the preprocessing, the final databased comprised of 277 datapoints, including 183 wrapped with CFRP and 94 wrapped with GFRP. Table 1 and 2 pertains to the statistical properties of the key inputs in the final database of the CFRP and GFRP datapoints. The input variables selected for the prediction model has cylinder diameter, height to diameter ratio, Unconfined concrete strength of concrete, FRP thickness, FRP tensile strength which is calculated