Sustainable Machine-Learning Framework for Efficient Color Prediction in Dope-Dyed Recycled PET/PCT

Abstract

Dope dyeing offers an eco-efficient alternative to conventional textile coloration by incorporating pigments directly into the polymer melt, minimizing water and chemical use. This study presents a sustainable data-driven framework that reduces repetitive dyeing trials and resource consumption during color matching of dope-dyed recycled PET/PCT microfiber fabrics. A two-stage hybrid machine learning model-combining k-nearest neighbors (kNN), feature expansion, and residual modeling-was developed to predict subtle color variations within the narrow CIELAB output range inherent to dope-dyed systems. The model achieved R 2 values above 0.83 for L*, a*, and b*, and external validation with untrained dyeing recipes yielded a mean Delta E of 0.65 with visually negligible deviation. By accurately pre-estimating color outcomes, this approach minimizes iterative experiments, energy use, and wastewater generation, contributing to sustainable textile manufacturing. The proposed framework demonstrates that data-driven color prediction can enhance process efficiency and environmental performance in dope-dyed fabric production, supporting circular and low-impact coloration technologies.