Achieving robust damage mode identification of adhesive composite joints for wind turbine blade using acoustic emission and machine learning

Abstract Interest in damage mode classification of composite structures by Acoustic Emission (AE) inspection technique and clustering analysis by machine learning has been increasingly growing. Furthermore, hyperparameters in clustering analysis promote the need for more robust clustering algorithms. This paper presents a clustering method by fast search and find of density peaks (CFSFDP), where robust identification for different damage modes can be achieved by means of similarities of AE signals. Based on the clustering analysis, matrix cracking and shear failure of the adhesive layer are demonstrated to be fundamental and characteristic damage modes, respectively. Meanwhile, similarities of AE signals for various damage modes in the subspace of AE features are explored in detail. The interface debonding (fiber/matrix interface debonding and adhesive failure) and the cracking of polymer (matrix cracking and cohesive failure) behave similarly in the subspace of selected AE features. As a different damage mode, fiber breakage is shown to be more similar to delamination, in contrast with other damage modes in the subspace. Moreover, effects of the selection of cluster number, the metric of spatial similarity and the value of cutoff index on the clustering results are shown to be negligible.