Using Bayesian regulated neural network (BRNN) to predict the effect of plasma treatment on the fading effect of cotton fabric

The utilization of plasma treatment as a technique for fading the colours on textiles is an environmentally friendly approach. However, the challenge lies in the continuous adjustment of its parameters to attain desired fading colour effects. This is due to the unclear mechanism causing changes in fading effect as a result of alterations in plasma treatment parameters. This research endeavours to predict the alteration of fading effect after plasma treatment of cotton fabrics under varying parameters through the development of a prediction system based on a Bayesian Regulated Neural Network (BRNN) with 10-fold cross-validation. Through a modular approach, the inputs for the training of BRNN models include initial values of plasma treatment parameters such as colour depth, air (oxygen) concentration, water content, treatment time, and one of the CIE L*a*b* values and K/S values corresponding to cotton materials. The outputs, trained individually by four independent BRNN models, are the final values of four colour variables: CIE L*, CIE a*, CIE b* and K/S. The models were trained and validated through the Bayesian Regularization Algorithm and 10-fold cross-validation on 192 data sets, yielding fitted coefficients of determination R-2 of 0.9956, 0.9976, 0.9980 and 0.9687, respectively. Approximately 87.5% - 91.67% of predicted colours were within the range of imperceptible or acceptable differences from actual colours. Hence, the developed artificial intelligence system can aid textile finishers in adjusting the plasma colour fading machine's parameter settings and selected recipes, thus enhancing efficiency and reducing cost and trial time.