A Novel Concept to Predict Cotton Yarns’ Coefficient of Variation and Hairiness Index by Online Collected Data During Winding Process

Abstract

Coefficient of variation (CV value) and hairiness value are important parameters used to assess cotton yarn quality. In order to accurately monitor yarn quality during production, we use online yarn quality detection as the input data of artificial neural network (ANN) and multiple regression (MLR) model, based on data collected by the electronic yarn clearer in the winding process as a novel concept. A prediction model of offline CV value and hairiness value was established, and the prediction performance of the two models was compared. The results show that the correlation coefficients of the offline CV values predicted by the ANN model reached 0.9832 and 0.9658 (bobbin and cheese yarn), which were much higher than the MLR model’s correlation coefficients of 0.8352 and 0.7877. Furthermore, the importance of online detection data for the prediction of CV value was analyzed. For hairiness value prediction, the correlation coefficients for bobbin and cheese yarn were above 0.923. In fact, the prediction accuracy of the ANN model was substantially higher. Therefore, the ANN model more effectively predicts the quality of offline yarn and achieves real-time monitoring.

Keywords

• artificial neural network
• multiple linear regression
• online detection
• predictive performance
• yarn properties