Abstract: Glass fiber bundle is a whole composed of hundreds of small glass fibers. Because of this structure, it is difficult to identify yarn-breaking defects in the production process of filament winding products. In order to solve this problem, a method based on machine vision is proposed to detect the defects of glass fiber bundles and the location of defects. The real-time image of the glass fiber bundle on the yarn road is captured by the industrial camera, and the image is transmitted to the computer. The image of each frame of the glass fiber bundle is processed by the OpenCV library, and the outline and defect characteristics of each glass fiber bundle are obtained. According to the defect characteristics, whether the glass fiber bundle is completely or partially broken is judged by the defect detection algorithm, and the location of the defect is determined by using the KNN algorithm. The movement rate of glass fiber bundle is 1 m/s, and 600 images are collected at a frame rate of 30 fps for experimental verification. The detection data show that the comprehensive accuracy is up to 96.6%, which meets the requirements of glass fiber bundle defect detection.

Key words: machine vision, glass fiber bundle, image processing, defect detection, KNN classification algorithm, composites