Data-Driven Spectral Prediction of Black Dyeing in Recycled Polymer Microfibers via Multi-Output Regression

Area of Science:

• Textile Chemistry
• Color Science
• Data-Driven Modeling

Background:

• Accurate color prediction is crucial for textile dyeing processes, especially for black shades in recycled microfiber fabrics.
• Traditional color prediction relies on scalar values, limiting spectral accuracy and physical interpretation.
• Developing data-driven models for spectral prediction is essential for improving color reproducibility.

Purpose of the Study:

• To present a spectrum-based data-driven modeling framework for predicting wavelength-dependent K/S spectra.
• To enable direct interpretation of color reproducibility by predicting the full K/S spectrum (400-700 nm).
• To provide practical support for data-driven dye recipe design and quality validation in microfiber dyeing.

Main Methods:

• Formulated the problem as a multi-output regression task, predicting K/S values at each wavelength.
• Systematically compared eight regression models for black-shade data with high inter-wavelength correlation and limited spectral variation.
• Selected partial least-squares regression (PLSR) as the final model due to its performance.

Main Results:

• The PLSR model demonstrated high agreement between predicted and measured K/S spectra.
• Conversion to CIELAB coordinates resulted in a mean color difference ($\Delta E_{ab}^{*}$) of 0.79, indicating excellent color reproducibility.
• The framework effectively modeled black-shade dyeing data with a limited sample size.

Conclusions:

• The proposed spectrum-based data-driven framework accurately predicts K/S spectra for black dope-dyed recycled microfiber fabrics.
• The model maintains both spectral agreement and perceptual accuracy, crucial for color reproducibility.
• This approach offers practical benefits for color control, process efficiency, and quality validation in microfiber dyeing.