关键词: 有色再生纤维, 纤维颜色测量, 纤维配色, 混色预测, 神经网络

Abstract: In 2022, the General Office of the State Council issued “Implementation Opinions on Accelerating the Circular Utilization of Waste Textiles”, which set a target of achieving a recovery and reuse rate of 50% for waste textiles by 2025. With the development of regeneration technology and the enhanced awareness of enterprises towards reuse, the spinnability of recycled fibers has been continuously improved. The production of colored-spun yarn from recycled colored fibers represents a new direction for the development of waste textiles to avoid decolorization treatment.
To realize the color recovery of recycled colored fibers, this paper proposes a digital prediction path for the color mixing process in colored-spun yarn production. First of all, the accurate measurement of fiber color is the basis of color mixing research. Due to the fluffy and deformable nature of fibers, they are typically converted into yarn or woven into small samples for indirect color measurement in actual production. For this reason, this paper adopts 0.1 g fiber carding into the same direction arrangement, tightly opaque state, and with homemade edge length of 2 cm square cardboard for color reading. The RGB values of the colored fibers are collected using digital colorimetry and converted into L*, a*and b* values. Compared with the color measurement results of spectrophotometer and colorimeter, digital colorimetry can directly read the average color of the selected area without the limitation of the test aperture size. The results, converted into color patches, are closer to visual inspection and are suitable for reading the colors of mixed fibers. Secondly, this paper explores the factors affecting the color measurement process of fiber samples. Experiments show that fibers need to be combed into a unidirectionally parallel and opaque state during measurement. To avoid the influence of different pressures on fiber density, the sample mass is fixed at 0.1g, and the same self-made color measurement cardboard is used for measurement. At the same time, in order to ensure the uniformity of the fiber color mixing, it is necessary to use the fiber extensometer to mix the fiber 10 times to ensure the stability of the color measurement results. Finally, this paper employs a neural network for fitting and prediction by comparing the theoretical color mixing results calculated using the color averaging method with the actual L*, a*and b* values of the mixed fiber samples. The input layer consists of the theoretical color mixing L*, a*and b* values calculated after mixing fibers according to a specific ratio, while the output layer represents the actual L*, a*and b* values of the samples. The hidden layer performs nonlinear fitting and prediction on the optimized data. The results show a coefficient of determination reaching 0.9850, with 63% of the samples having a color difference of less than 1. 
This method enables digital control of the color of mixed samples, which not only facilitates rapid prototyping and accelerates new product development but also simplifies digital communication and expression of color information. In the future, we need to increase the number of training samples to improve the accuracy of prediction and, on the basis of ensuring accuracy, expand the data set to include different types of fibers, so as to provide more effective color matching prediction reference for color textile enterprises.

Key words: recycled colored fibers, fiber color measurement, fiber color matching, color mixing prediction, neural networks