Automated identification of defect morphology and spatial distribution in woven composites

The performance of heterogeneous materials, for example, woven composites, does not always reach the predicted theoretical potential. This is caused by defects, such as residual voids introduced during the manufacturing process. A machine learning-based methodology is proposed to determine the morphology and spatial distribution of defects in composites based on X-ray microtomographic scans of the microstructure. A concept of defect "genome" is introduced as an indicator of the overall state of defects in the material, enabling a quick comparison of specimens manufactured under different conditions. The approach is illustrated for thermoplastic composites with unidirectional banana fiber reinforcement.