Assessing distortion in carbon fiber woven fabrics based on machine vision

Abstract

Distortion in carbon fiber woven fabrics significantly impacts composite mechanical performance through defective fiber tow distribution. This work proposes a machine vision method to locate defective areas, identify defects, and describe fiber tow distribution patterns. A back-lighting imaging system was designed to minimize surface reflection interference, enabling high-quality fabric image acquisition. We developed the Light Transmission Algorithm (LT) analyzing voids at fiber tow intersections to calculate void-to-fabric ratios, providing qualitative and quantitative distribution indicators. A defect recognition method combining isometric and random feature sampling enables segmentation of abnormal fiber distribution regions through standard sample comparisons. The system achieves 95%–100% identification accuracy. The proposed methods demonstrate strong interpretability and robustness in assessing the quality of carbon fiber woven fabrics, addressing critical challenges in local defect detection while enabling comprehensive distribution analysis.

Keywords

• Fiber tow distribution
• carbon fiber fabric
• machine vision
• defect detection
• defect segmentation