Artificial neural network-based prediction of steel-concrete bond shear strength under elevated temperature exposure

This study proposes an artificial neural network (ANN)-based predictive model to estimate the percentage normalised steel-concrete bond stress under elevated temperature conditions. The model was developed using an extensive experimental database comprising 393 data points collected from the published literature, incorporating key governing parameters such as material properties, geometric characteristics, and temperature. The proposed ANN formulation demonstrated high predictive capability, achieving a coefficients of determination of 0.97 for the design dataset and 0.89 for the testing dataset. Second-order interaction analysis indicated that fibre type and volume fraction exhibit pronounced interaction effects, while geometric ratios such as length-to-diameter and cover-to-diameter exhibit comparatively weak interaction effects. For practical engineering applications, prediction tuning parameters were also proposed for the ANN-based predictions, using cumulative distribution functions, enabling conservative estimation of percentage normalised bond shear stress with controlled prediction uncertainty. The ANNs based predictions were compared with the existing semi-empirical design models and established ANN-based modelling as a robust and efficient alternative. The proposed approach demonstrates substantial potential for application in modular structural systems, where rapid pre-design evaluation, parameter optimization, and reliable stability assessment are critical.