Adequate bonding between rebar and concrete is the key to ensuring the reliable performance of RC structures. This rebar-concrete bond behavior directly influences the structural load carrying capacity and structure failure mode. It is found empirically that bond behavior is affected by many factors, including concrete cover, transverse reinforcement, bar spacing, bar size, bar geometry, concrete properties, steel stress and yield strength, bar surface condition, and etc. While many past studies have focused on the prediction of bond strength, how those factors influence the bond failure mode (i.e., pullout failure or splitting failure) is not well investigated, particularly when the concrete is not well confined and/or corrosion is present.
The goal of this research is to propose a probabilistic model to predict bond failure mode considering corrosion. The model development is based on a group of bond testing result of beam end specimens with various rebar size, corrosion levels, covers, and stirrup confinement. This study adopts logistic regression and lasso logistic regression, where the failure mode is the categorical dependent variable and the aforementioned factors that could influence the bond behavior are the independent variables. In lasso logistic regression, a penalty factor is used for remove insignificant variables; in logistic regression, a model selection is applied to select variables that are needed for an accurate and practical model. The developed model can be used for bond failure mode prediction, and the selected variables can help developing the optimal design and/or retrofitting strategies. Lastly, the proposed bond model is employed in the nonlinear finite element models of intact and corroded RC beams to investigate the importance of bond failure mode prediction model on evaluating flexural behavior of the beams.