Failure analysis and on-line damage monitoring based on deep-learning for thermo-oxidative aged 3D angle-interlock woven composites under tension

Abstract

Failure analysis and real-time damage monitoring is of great significance for evaluating the service life of carbon fiber-reinforced epoxy composites structural parts. However, due to the complex microstructure and heterogeneous properties of composite materials, achieving online damage identification in practical applications remains challenging. In this paper, digital image correlation (DIC) technology is used to analyze the failure mode of 3D angle-interlock woven composites including surface strain, damage types. Combining with deep learning network, a comprehensive system deep learning network is developed for identifying, segmenting and analyzing the surface damage of composite materials. In warp tension, transverse cracks initiate in the resin region, with their width confined between two adjacent warp yarns. In weft tension, only transverse cracks originate at the fiber-resin interface within the surface warp yarns and propagate into the surrounding resin. After training, the YOLOv5x model performs well across all categories, with an especially high accuracy of 0.991 in detecting transverse cracks. The trained YOLOv5x deep learning network was used for skeleton extraction, type identification and quantitative statistics for cracks. The statistical analysis shows that the modulus decrease is related to the cracks, and the damage threshold of the composites remains the same across different aging periods.