A single-sensor method for structural damage localization in wind turbine blades: Laboratory assessment on a blade segment

An acoustic emission (AE) and graph convolutional network (GCN) based single-sensor method for structural damage localization in wind turbine blades is proposed in this paper. Continuous Wavelet Transform (CWT) is applied to convert AE signals to a sequence of wavelet slices and each of which is treated as a node of graph. Feature extraction for nodes is performed in terms of the wavelet coefficients using one-dimensional convolutional neural network (1D-CNN). Subsequently, Euclidean distance between nodes is calculated to construct the weighted K-Nearest Neighbors Graph (weighted KNN-Graph). The graphs are then utilized as input for the GCN, which extracts deep damage location features. These features are combined with a graph-level regression analysis to determine the coordinates of the damage. Hsu-Nielsen pencil lead break (PLB) tests are conducted on a blade segment and the experimental data are used for 5-fold cross-validation for effective evaluation of the performance of the proposed method. Moreover, the robustness of the proposed method is tested by AE signals with Gaussian white noise. Furthermore, a comparison study on localization accuracy between the proposed method and various methods of convolutional neural network (CNN) is made. The results reveal the high performance of the proposed single-sensor method for structural damage localization.