Spectrophotometric prediction of pre-colored fiber blends with a hybrid model based on artificial neural network and Stearns–Noechel model

Abstract

The performance of the traditional color prediction model in the color prediction of colored fiber blends is usually not very satisfactory under a variety of conditions, due to the limitation of the trail data and the assumptions used for derivation. In contrast to the traditional model, artificial neural networks (ANN) have an excellent nonlinear mapping ability; however, they also have poor generalization ability if the training data are not sufficient. In this paper a hybrid model, called the Stearns–Noechel (S-N)–ANN model, is proposed, which combines the S-N model with the ANN model. This uses the S-N model first to build the approximate relationship between the recipe and spectrophotometric response of the color blends, followed by optimization with the ANN to achieve higher prediction accuracy and better practicability. Compared with the ANN model, the S-N–ANN model needs less training time with the same training data, yet achieves higher validation and correlation coefficients, indicating that the training of the S-N–ANN model is much easier. The average color difference of the predicted spectrum obtained with the S-N–ANN model was 0.86 CMC(2:1) unit, which was much lower than that obtained with either the ANN model (∼2.21) or the traditional S-N model (∼1.66), indicating that the S-N–ANN model is a more accurate method for the color prediction of colored fiber blends.