An exponential smoothing multi-head graph attention network (ESMGAT) method for damage zone localization on wind turbine blades

An exponential smoothing multi-head graph attention network (ESMGAT) method is proposed for structural damage zone localization on wind turbine blades. This marks the first time that multi-head graph attention networks are introduced into the AE source zone localization task. This method introduces two key innovations: (1) Damage zone localization for wind turbine blades using only a single sensor. (2) Exceptional localization performance in the presence of noisy AE signals. First, the original AE signals are processed using exponential smoothing to effectively smooth them out and eliminate noise. Next, the smoothed AE signals undergo decomposition through continuous wavelet transform (CWT), and the resulting wavelet coefficients are utilized as node features. Euclidean distances between node features are calculated to assess the connectivity within graphs. Additionally, a new aggregation method is introduced for multi-head graph attention networks to enhance the robustness of the proposed method under noisy conditions. Finally, the effectiveness of the ESMGAT method is validated using the dataset from pencil lead break (PLB) tests conducted on a segment of a wind turbine blade.