A novel automatic crack classification algorithm of 3-D woven composites based on deep-learning U-Net model

The crack classification and statistics of composites under different loading conditions are significant to damage mechanism analysis. For computed tomography (CT) images with complex backgrounds and low resolution of composite materials, it is time-consuming to obtain different types of cracks by manual extraction. In this paper, a novel crack classification algorithm combining the U-Net model and morphological operation, was proposed to automatically extract cracks in warp, cracks in weft, cracks in resin and interfacial cracks. The efficiency of the algorithm is thousands of times higher than that of manual extraction, and the algorithm accuracy is more than 88%. Based on the proposed method, the crack distribution in composites can be quantitatively analyzed. The statistical results show that the proportion of interfacial cracks is up to 8%, extremely higher than that of other cracks. It is expacted that the proposed method can be used to quantitatively analyze the internal damage state of composites under different loading conditions.