Strain variations in edge-restrained walls: Data preprocessing for exploring artefacts and ranges of uncertainty

Early-age cracking in restrained concrete elements remains a major durability challenge in structural engineering. Experimental strain measurements are commonly used to validate numerical models and, potentially, to train machine learning frameworks. However, current preprocessing practices typically rely on simple averaging or smoothing, which assume uniform strain behaviour and suppress meaningful variability linked to structural response. This study demonstrates that such assumptions are not always suitable for edge-restrained walls. Using strain data from four large-scale wall experiments, systematic variability was identified across wall thickness, length, and height, with strain divergence frequently exceeding 100–200 με (microstrains), far greater than the demountable mechanical strain gauge (DEMEC) instrument uncertainty, which is around 10 to 20 με . ,A variability-based preprocessing framework is proposed, using strain range envelopes and coefficient of variation (COV) analysis to preserve meaningful deformation behaviour rather than mask it. The results show that localised restraint, material heterogeneity, and cracking strongly influence strain evolution, invalidating symmetric or uniform deformation assumptions. The study concludes that averaging raw strain data leads to a loss of critical mechanical behaviour information and compromised data quality. The proposed approach enables variability-aware validation of finite element models and provides physically meaningful target bounds for data-driven modelling. This work lays the foundation for reliable hybrid numerical–machine learning strategies for early-age cracking prediction.