Quantitative evaluation method for wrapping effect of vortex yarn based on image analysis

Unlike traditional ring-spun yarn, vortex yarn is wrapped by a layer of wrapping fibers, forming a special wrapping effect that determines the yarn’s properties. Quantitative evaluation of this effect is significant for quality control and diagnosis in manufacturing. However, this currently relies on tracer fibers or manual annotations. In this work, a method is constructed, based on image analysis, to improve the convenience and accuracy of the evaluation. First, a device is built to implement image acquisition and analysis algorithms. Then, the algorithms are constructed using a model based on a convolutional neural network (WRP-UNET++) for wrapping region segmentation associated with a yarn sequential attention mechanism and shape constraints. Finally, a set of evaluation indexes, including wrapping ratio, short wrapping rate (SWR), long exposure rate (LER), and wrapping angle, is proposed to evaluate the wrapping effect quantitatively, considering the area and angle of the wrapping regions. In comparison experiments, the proposed model (WRP-UNET++) outperformed existing image segmentation models, achieving an accuracy of 89.20% and an intersection over union of 83.20% with manual annotations as the benchmark, while recording the lowest mean absolute percentage errors, of 5.31%, 15.74%, 46.82%, and 2.43% for the wrapping ratio, SWR, LER, and wrapping angle, respectively. Additionally, ablation experiments confirmed that the proposed attention mechanism and shape constraints significantly enhance the segmentation results. Regression analysis revealed a significant correlation between the proposed indexes and several yarn properties, including breaking strength, breaking strength coefficient of variation, breaking elongation, abrasion resistance, thin places, hairiness index, and coefficient of variation of diameter, indicating the potential industrial applicability of the proposed method.