Damage detection and localisation in wind turbine blades under temperature variations

This study presents a data‑driven, semi‑supervised framework for damage detection in wind turbine blades (WTBs). Autoregressive (AR) and multivariate AR models are employed to extract features from acceleration and strain time‑series signals. The approach is evaluated on an existing laboratory dataset from a composite WTB subjected to progressive cracking at multiple locations under varying temperature conditions. Results show that acceleration measurements are highly effective for detecting the onset of damage, while strain gauges provide superior accuracy for localising damage. The framework demonstrates high sensitivity to subtle structural changes, enabling early and reliable detection. Its scalable and automated design highlights strong potential for deployment in offshore wind applications and broader structural health monitoring contexts.