Prediction of Color Coordinates of Cotton Fabric Dyed with Binary Mixtures of Madder and Weld Natural Dyes Using Artificial Intelligence

In this study, cotton fabrics were dyed with different combinations of aluminum potassium sulfate (eco-friendly mordant), besides weld and madder as natural dyes. Then, the L*, a* and b* color coordinates were measured. The statistical analysis indicated that all three mentioned materials have significant effect on the color coordinates of the dyed fabric samples. In the next step, it was tried to model the relation between the concentration of mentioned materials and each color coordinate separately using regression method, artificial neural network (ANN), fuzzy logic and support vector machine (SVM). Moreover, in order to increase the models' accuracy, the genetic algorithm, particle swarm optimization (PSO) and gray wolf optimization (GWO) were applied to optimize the parameters of ANN, fuzzy logic and SVM. Mean absolute percentage error (MAPE) was calculated to assess the model's accuracy. It was revealed that the regression method indicates an acceptable accuracy only for L*, but the other models can predict all color coordinates with high accuracy. Finally, it was found that in prediction of L* and b*, ANN optimized with GWO presents the most accurate model with MAPE of 1.29% and 2.51%, respectively. While in the case of a*, ANN optimized with PSO possess the least MAPE (4.65%).

Automated summary

In this study, cotton fabrics were dyed using aluminum potassium sulfate, weld, and madder as natural dyes. The color coordinates (L*, a*, b*) were measured and it was found that all three materials had a significant effect on the color of the fabric samples. Regression, artificial neural network (ANN), fuzzy logic, and support vector machine (SVM) models were used to predict the color coordinates based on the concentration of the dyeing materials. Optimization techniques such as genetic algorithm, particle swarm optimization (PSO), and gray wolf optimization (GWO) were applied to improve the accuracy of the models. The results showed that the ANN optimized with GWO had the highest accuracy for predicting L* and b*, while the ANN optimized with PSO had the least mean absolute percentage error for predicting a*.