Data-driven modeling for bond strength in reinforced concrete structures

In reinforced concrete (RC) structures bond strength is critical mechanical parameters responsible for the load transfer, cracking behavior, and structural integrity, especially for higher temperature scenarios. Research framed, data driven, data reinforced predictive model is constructed to estimate the normalized bond strength of reinforced concrete under high-temperature scenarios using the techniques of machine learning. A total of 458 data points from various publications were assembled, and for the purpose of training and evaluating the machine learning regression models, the ensemble and the kernel-based and neural network techniques, several machine learning algorithms were utilized. A modeling protocol ensuring robust generalization with standardization, repeated cross validation, and data driven hyper parameter optimization was used for the purpose of comparative modeling. The XGBoost regressor was the most predictive among the 12 regression models created, with a total coefficient of determination of R2 = 0.9417 with a root quadratic mean error (RQME) of 7.9 on the testing data set. Further analysis of important predictive factors, such as degradation of bond strength, and the exposure temperature, provide credible, data driven and rational emphasis for the geometric parameters of the predictive model. The suggested framework illustrates that, for estimating the bond strength of reinforced concrete exposed to high-temperatures, machine learning is not only an effective tool, but also one that provides the assessments in a timely manner. Thus, it can be used in the areas of structural fire assessment, post-fire evaluation, and durability design.