A hybrid artificial intelligence model to predict the color coordinates of polyester fabric dyed with madder natural dye

Abstract

Color matching is an important issue in textile dyeing especially when facing with natural dyes. Generally, the expert engineers evaluate the fabric’s color visually and try different dyeing parameters to achieve that based on the trial and error method and their own experiences. In this study a database was created by evaluating the color of polyester fabrics dyed with madder root at various dyeing conditions using reflectance spectrophotometer. Then, the relations between dyeing parameters and color coordinate values known as l*, a* and b* were analyzed statistically. After that, each color coordinate was modeled separately based on dyeing parameters using linear regression as the old-typical method and artificial neural network (ANN). To increase the ANN model accuracy, the genetic algorithm was implemented to optimize ANN’s parameters. The results indicated that ANNs with two, one and two hidden layers are the accurate predictor tools for l*, a* and b* values, with mean absolute percentage error of 0.67, 1.29 and 1.27, respectively. In the next step, some color coordinates were selected randomly and it was tried to find the corresponding dyeing parameters using multi objective optimization based on the obtained ANN models. A comparison between the actual and obtained dyeing parameters showed the high efficiency of proposed method to determine the dyeing parameters in order to achieve the desired color.