LAUSR

The ultimate bond strength of corroded steel reinforcement and surrounding concrete critically affects the load carrying capacity and eventually serviceability of the reinforced concrete structures. This study constructs and verifies a data-driven method for estimating ultimate bond strength. The proposed method is a hybridization of least squares support vector regression (LSSVR) and differential flower pollination (DFP) computational intelligence approaches. Since the problem of ultimate bond strength prediction involves nonlinear and multivariate data modeling, the LSSVR is employed to infer the mapping function between ultimate bond strength and its influencing factors of concrete compressive strength, concrete cover, steel type, diameter of steel bar, bond length, and corrosion level. Moreover, in order to overcome the very challenging task of fine-tuning the LSSVR model training, the DFP algorithm, as a population-based metaheuristic, is utilized to optimize the performance of the LSSVR prediction model. A dataset including 218 experimental tests has been collected from the literature to construct and verify the proposed hybrid method. Experimental results supported by the Wilcoxon signed-rank test point out that the hybridization of LSSVR and DFP can deliver predictive results (root-mean-square error = 2.39, mean absolute percentage error = 33.82%, and coefficient of determination = 0.84) superior to those of benchmark models including the artificial neural network, the multivariate adaptive regression splines, and the regression tree. Additionally, a software program based on the LSSVR model and the DFP optimization result has also been developed and compiled in Visual C#.Net to ease the model implementation. Hence, the hybrid model of DFP and LSSVR can be a promising alternative to assist engineers in the task of evaluating the health of reinforced concrete structures.