PSI - Issue 64

Manish Prasad et al. / Procedia Structural Integrity 64 (2024) 1524–1531 Manish Prasad / Structural Integrity Procedia 00 (2019) 000 – 000

1530

7

SVM

0.885 0.949 0.974 0.191 0.370 0.196

1.012 1.006 1.031 1.696 2.582 2.559

0.147 0.104 0.076 0.552 1.016 0.849

RF

XGBOOST

GEP

fib Bulletin 90

Smith and Teng (2002)

The GEP model theoretically considers the best mathematical functions given by the user to fit the data, while fib Bulletin 90 and Smith and Teng estimate the CCS load as a factor of design shear resistance of the beam. Based on the R-squared and mean score in Table 2, the mathematical expression given by GEP model, fib Bulletin 90, and Smith and Teng do not agree well with the experimental results. On the other hand, as observed in Table 2, ML regression models present better predictions on the CCS capacity than the analytical formulations. All these analytical estimations along with their calibrated factors have been obtained from the experimental observation though some means of fittings or regression and are generally limited by the form of mathematical function and engineering understanding of the problem at hand. ML is not limited by these constraints and can perform thousands of trials to fit the data with an algorithm. Therefore, ML models have shown higher accuracy compared to regression based on mathematical functions. The R-squared scores obtained from ML are the highest, and the mean values of the ratio of experimental failure load and that of calculated by the ML model are the lowest, which means that ML can estimate the value very close to the experimental results. Based on the standard deviation, results from ML methods deviate the least from the experimental results. For the better prediction of new data using ML, R-squared score is chosen to select the best ML model. By comparing the 4 ML methods studied in this work, and based on R-squared score, KNN is the most accurate model to predict CCS failure load (higher 0.97 R squared score, lower 0.078 standard deviation). Moreover, a wide range of variables, which cannot be easily covered by experimental investigations, can be incorporated into ML models. 5. Conclusions A total of 140 four-point load test data for beams strengthened with EB CFRP from past research were collected. The accuracy of some existing analytical models to predict CCS load was checked with the collected data. Due to the limited accuracy of the existing models in predicting the failure load, ML was adopted to address the problem. Four different ML algorithms were used to predict the CCS load. These algorithms were trained on 80 % of the experimental data collected. The trained models were then used to predict the CCS load of the rest of the data (20% of the database). All ML models performed better than the existing analytical models in predicting the CCS load and with an R-squared score higher than 0.88. KNN, with an R-squared score of 0.975, performed the best. Acknowledgements The authors acknowledge the support provided by the Spanish Ministry of Science, Innovation and Universities through the projects PID2020-119015GB-C21 and PID2020-119015GB-C22 funded by MICIU/AEI/10.13039/501100011033; M.P acknowledges the grant PRE2021-100670 funded by MCIN/AEI/10.13039/501100011033 and European Social Fund Plus (ESF+); M.A. acknowledges the grant IFUdG2021/01 funded by the Generalitat de Catalunya through the predoctoral program AGAUR-FI and the FSE+. References Abuodeh, O. R., Abdalla, J. A., & Hawileh, R. A. (2020). Prediction of shear strength and behavior of RC beams strengthened with externally bonded FRP sheets using machine learning techniques. Composite Structures , 234 . Ahmed, O., & Van Gernert, D. (1999). Effect of Longitudinal Carbon Fiber Reinforced Plastic Laminates on Shear Capacity of Reinforced Concrete Beams. ACI Symposium Publication , 1888 , 933 – 944. Al-Ghrery, K., Kalfat, R., Al-Mahaidi, R., Oukaili, N., & Al-Mosawe, A. (2021). Prediction of Concrete Cover Separation in Reinforced Concrete Beams Strengthened with FRP. Journal of Composites for Construction , 25 (4).