Ring spun yarn quality prediction using hybrid neural networks

Abstract

Yarn quality is an essential concept that requests the satisfaction of various parameters simultaneously. These parameters are fiber characteristics and yarn properties. Thus, it is necessary to develop model that can encompass all these variables and predict an overall yarn quality index. Since hybrid approaches combining two or more techniques have proved their abilities in processing and predicting accurately various variables over different research areas, this paper reports two hybrid models by combining two different approaches for predicting a new quality index: Back-propagation artificial neural networks (ANN) and fuzzy expert system. The hybrid models are ANN combined with ANN and ANN combined with fuzzy logic. The ANN is used to predict four yarn characteristics namely tenacity, breaking elongation, CVm and hairiness. Then, these four outputs are used to predict a new quality index by means of ANN or fuzzy expert system. Several performance criteria are necessary to evaluate the performance of the established models. They are correlation coefficient (R), root mean square error (RMSE), mean absolute error (MAE) and mean relative percent error (MRPE). The obtained results show that these constructed hybrid models are able to predict yarn quality from the chosen input variables with a reasonable degree of accuracy. Moreover, the hybrid model using two ANN systems performs significantly better than the other one.