Data and code for "A noise-aware criterion for neural network surrogates of corrosion-induced crack paths in the concrete cover"

Phase-field labels, replicate solutions and code supporting the article "A noise-aware criterion for neural network surrogates of corrosion-induced crack paths in the concrete cover". The dataset contains the crack height labels z(x) at 25 abscissae for 198 cases with two bars at equal height (family A) and 175 cases with staggered bars (family B), solved with a plane-strain AT2 phase-field model on Q4 elements at a mesh size of 0.25 mm in a 100 x 100 mm domain. It also contains 20 replicate pairs per family (original solution and the solution with the bars shifted by half a cell) used to compute the replicate noise floor, the phase-field solver (corr2d.py, fem2d.py) and the scripts for training, the noise-aware criterion, the baselines and the figures (Python 3.12, scikit-learn 1.8.0). See README.md for the file layout.