Predicting the Effect of Plasma Treatment on the Fading of Sulphur-Dyed Cotton Fabric Using Bayesian Regulated Neural Network (BRNN) and Gaussian Process Regression (GPR)

Addressing fabric fading in the textile industry, this research investigates plasma treatments as a sustainable solution. The ambiguous relationship between plasma parameters such as air concentration, moisture content, and treatment duration with fading introduces uncertainty in the treatment process, necessitating optimization through precise prediction. Such precision not only curtails costs but also highlights the pivotal role of digital innovations in textile fabrication. Two predictive systems, Bayesian Regulated Neural Network (BRNN) and Gaussian Process Regression (GPR), were developed using tenfold cross-validation to accurately link these parameters to fading results. To handle prediction complexity and small samples, a modular approach was adopted. Four individual BRNN and GPR models predicted CIE L*, a*, b*, and K/S values and were integrated into a single system. The systems’ predictive capabilities were compared, revealing that GPR’s predictive accuracy generally surpassed that of the BRNN model, although its advantage lessened in the testing set, particularly for CIE a* predictions. Nevertheless, it is noteworthy that both GPR and BRNN models demonstrated significant prediction performance on the unseen dataset, with the colour prediction deviating from the actual colour within an acceptable range for industrial applications in most scenarios. Specifically, under a stricter standard of ∆E < 1, the GPR model's predictive accuracy was able to maintain between 79.17 and 87.5%. Thus, the application of the BRNN and GPR models might provide valuable digitalization insights for cotton processing, saving time and cost while efficiently predicting fading effects.