In the textiles dyeing industry, the process of color matching is extremely complex due to nonlinearities in the dyeing process. The sources of these nonlinearities are many, including the dye-to-dye interactions and the dye-to-fiber interactions. In this paper, a neural network model is employed to describe the process of color matching. The model is constructed using 31 input neurons, 15 hidden layer neurons, and 3 output neurons which are fully connected. Each of the hidden and output layer neurons is based upon a sigmoid activation function. In order to test the accuracy of the neural network model in comparison to a linear model, the neural network model was trained on 150 experimental samples to 250,000 epochs. The dye-concentration prediction of the neural network model was shown to be twice as accurate in predicting the actual experimental dye-concentrations as the linear model. The neural network algorithm developed in this research effort and described in this paper is an improvement over classical neural network methods in that the approach incorporates adaptive learning rates with an innovative stabilizing algorithm. This results in a reduction in the training time compared to classical neural network approaches. Currently, the neural network can be trained to predict the concentration of three dyes in a color sample; the approach can be extended to predict additional dyes. Further, the neural network algorithm is supported by a graphical-user interface, a non-relational database, an interface to a spectrophotometer, and other software support modules. Results of comparison between the neural network model and an actual color matching system illustrate the viability of the neural network model for color matching.
展开▼
