Bond-Transfer-Controlled Crack-Width Modelling of Reinforced Concrete as a Cement-Based Composite Under Imposed Strains

Reinforced concrete subjected to restrained imposed strains exhibits a cracking response governed by the interaction between concrete tensile behaviour, reinforcement–concrete bond transfer, crack-spacing development, reinforcement stress, and internal force redistribution. This study investigates this bond-controlled response at the material-system level and uses it to assess and extend the EN 1992-1-1:2023 crack-width formulation. A measurement-level database was compiled from published tests, comprising 137 specimens and 1473 filtered observations corresponding to successive imposed-strain or loading stages. The original EN 1992-1-1:2023 model reproduced the general trend but exhibited a systematic, stress-dependent underestimation, especially in the low-to-intermediate reinforcement stress range. To account for this range-dependent behaviour, bond- and stage-sensitive terms were introduced into the crack-spacing and mean strain-difference components. Within the calibration database, the modified model improved the agreement with measured mean crack widths from an R2 = 0.72 to R2 = 0.83 and reduced the underestimation rate relative to the estimated 95th-percentile crack width to 3.4% when measured reinforcement stress was used. Machine-learning analyses were used as diagnostic tools rather than replacements for the mechanics-based model. Gradient Boosting predicted mean crack width directly from imposed strain and material and geometrical parameters with R2 = 0.914, while SHAP interpretation confirmed the dominant role of imposed strain and the secondary influence of reinforcement ratio, bar diameter, and concrete tensile strength. These findings supported a semi-empirical stress-estimation relationship and a complete imposed-strain-based procedure. Within the calibration domain, the final procedure provides a mechanically interpretable, code-compatible estimate of upper-tail crack width that is equal to or greater than the adopted w0.95 reference for most measurement-stage observations.