A Data-Driven Approach for Predicting Industrial Dyeing Recipes of Polyester Fabrics

Polyester is extensively used in the textile industry for fabricating fibers and fabrics. Dyeing for polyester fabrics has a huge demand. Given that dyeing is one of the least environmentally friendly industrial processes, decreasing the attempts for dyeing (a.k.a realizing “one-shot” successful dyeing) is of vital importance to the dyeing manufacturing for polyester fabrics. This can be achieved by accurately predicting the dye concentrations for a dyeing recipe with provided target color information on the polyester fabrics. In this paper, we report a data-driven approach for accurately predicting industrial dyeing recipes of polyester fabrics. We intensively discuss the data preprocessing skills for this purpose. We show that log-transform and using full reflectance spectra for the color as input are two effective preprocessing techniques to improve the model performance. An effective model based on gradient-boosting regression tree (GBRT) has been developed to quantitatively model the relationship between the colorimetric information and the dye concentrations of industrial dyeing data of polyester fabrics. The developed approach can predict dye concentrations for dyeing tasks for polyester fabrics with error at 10–20%.