A meta-learning approach for vibration signal-based cross-domain wind turbine blade defect detection

Wind turbine blades (WTBs) are critical components for ensuring the safe and reliable operation of wind turbines. However, defect detection across different WTBs remains challenging due to distribution shifts. This paper proposes a time–frequency hard-sample-weighted metric meta-learning network (CHM-ML) for WTB defect detection. The proposed framework integrates three key components: (1) a lightweight time–frequency enhancement network serving as a feature extractor to learn discriminative feature representations; (2) a cross-blade metric module that adopts a weighted prototype strategy to dynamically adjust the contribution of support samples and mitigate negative transfer; and (3) a hard-sample identification module that applies adaptive reweighting to reduce the influence of confusing samples, thereby achieving joint optimization of feature learning and metric learning. Experimental results demonstrate that, under multiple cross-blade settings, the proposed method improves defect detection performance compared with representative baseline methods, indicating its potential for practical deployment.