Refining crack width predictions in RC beams using FEM and neural network-based surrogate models for crack band size estimation

Concrete cracking is analyzed using nonlinear fracture-mechanics-based constitutive models within a finite element framework. In such simulations, the predicted behavior of reinforced concrete is highly sensitive to the assumed crack spacing or crack band size, particularly when relatively large finite elements are employed. To alleviate this limitation, the present study introduces an approach in which artificial neural network surrogate models are used to estimate the crack spacing in reinforced concrete structures. Model uncertainties in terms of mean and maximum crack width are evaluated against a database of laboratory tests. The influence of reinforcement layout, geometric simplifications and mesh discretization on these uncertainties is examined. Overall, the proposed modelling strategy introduces an advanced tool for assessment of crack widths and mainly crack spacing in reinforced concrete structures at the serviceability limit state.