Predicting Bond Strength of FRP Bars in Concrete Using Soft Computing Techniques

Abstract

Fiber-reinforced plastic (FRP) rebars can be the futuristic potential reinforcing material in place of mild steel (MS) rebars which are highly prone to corrosion. However, the bond properties of the FRP rebars are not consistent with those of mild steel rebars. Therefore, determination of bond strength properties of FRP rebars becomes essential. In this study, an investigation was conducted on 222 samples for bond strength data set for FRP rebars using various soft computing techniques such as multilinear regression, random forests, random tree, M5P, bagged-M5P tree, stochastic-M5P, and Gaussian process. Outcomes of accuracy assessment parameters, i.e., CC, MAE, and RMSE, suggest that bagged-M5P tree-based model is outperforming than other developed models CC, MAE, and RMSE whose values are 0.9530, 0.8970, and 1.2531, respectively, for testing stages. On assessing the data and the results, it was found that GP_PUK model is more appropriate than GP_RBF-based model for predicting the bond strength of FRP (MPa). On comparison of the RF and RT models, it was concluded that RF-based model performs better than RT models with CC, MAE, and RMSE values of 0.9427, 0.8674, and 1.3424, respectively, for testing stages. The results of the study also suggest that bagged-M5P model attains higher correlation with lesser RMSE values. Taylor diagram also verifies that bagged-M5P model performs better than other developed models. Sensitivity analysis suggests that bar embedment length to bar diameter (l/d) is the most influencing parameter for the prediction of bond strength of FRP.