Study the effect of ANN splitting ratios and training functions on the prediction of corroded steel-to-concrete bond strength

The structural integrity of the structures depends on the bond between steel and concrete. The strong bond ensures composite action and a unified force-resistant system which results in the efficient distribution of loads and enhances the overall durability and strength of the structure. The present study deals with applying and developing artificial neural network (ANN) models for the prediction of bond strength (BS) between concrete and steel reinforcement. To evaluate the effectiveness of the ANN model with the experimental dataset, the ANN models were initiated through activation functions such as Bayesian regularization, Levenberg–Marquardt, scaled conjugate gradient, and linear regression. The efficiency of the developed machine learning model was determined by comparing it with thirty-one existing analytical models. A violin plot along with a Taylor diagram was also used to graphically present the results. ANN-1 model outperformed other developed ANN models, exhibiting superior precision with a notable coefficient of correlation (R-value) of 0.9859, along with mean absolute error (MAE) of 1.53 MPa and root mean square error (RMSE) of 2.22 MPa. The sensitivity analysis results revealed that the diameter of the bar affects BS predominantly. Also, the ANN model provided a mathematical expression to estimate the BS. The developed ANN model enables the prediction of BS based on considered input parameters and helps engineers in designing and optimization. Additionally, the ANN model contributes to cost-effective and reliable construction practices, all of which can lead to the longevity and safety of the structures. The utilization of ANN based mathematical models offers a user-friendly approach with remarkable precision. Hence, researchers and structural designers can employ this formulation to estimate the BS between steel and concrete.