Deep learning-based analysis of damage mechanisms in 3D angle-interlock woven composites under variable impact conditions

Abstract

This study presents an innovative method to improve deep learning segmentation of warp and weft yarns in composites, overcoming the shortcomings of existing deep learning techniques in accurately defining yarns. The method entails threshold screening of yarn area and aspect ratio, combined with morphological opening operations and an improved watershed algorithm to enhance the segmentation map’s accuracy. The findings indicate significant improvements in both continuity and accuracy. An examination of failure modes across various impact energy levels indicates that weft yarns mainly absorb energy and support loads; however, weak interfacial adhesion between yarns and resin leads to debonding, which is the main failure mode. At increased impact energies, cracks develop within the composite components rather than at interfaces. This implies that improving the interfacial bond between yarns and resin could strengthen impact resistance. Based on these observations, the study suggests utilizing resin with superior bonding characteristics to enhance the material’s impact resistance and longevity.