PREDICTION OF THE INCREASE IN YARN UNEVENNESS AFTER WINDING PROCESS USING STATISTICAL AND ARTIFICIAL NEURAL NETWORK MODELS
Technical University of Liberec
This paper investigated the prediction of the increase in unevenness of two types of yarn: Ne 30/1 CVCM (combed yarn Ne 30/1, 60% Cotton 40% Polyester) and Ne 30/1 COCM (combed yarn Ne 30/1 100% Cotton) after winding by artificial neural network (ANN) and by statistical models. Four technological winding parameters: the winding speed (Z1), the load on the friction discs of the yarn tensioner (Z2), the distance between the bobbin and the yarn guide (Z3) and the pressure of the package on the grooved drum (Z4) were used as the input parameters to investigate yarn unevenness after winding. The research results showed that by using statistical models, within the selected research range, four investigated technological parameters influenced the increase in unevenness of the two mentioned yarns. The regression coefficients represented the influence of each technological parameter on the increase in yarn unevenness: the winding speed parameter has the most influence on the increase in yarn unevenness with the biggest value coefficients b1 which was 1.2339 for the Ne 30/1 CVCM yarn and this value was 0.6996 for the Ne 30/1 COCM yarn. Moreover, the increase in yarn unevenness predicted by ANNs obtained a higher coefficient of determination (R2), while the mean square error (MSE) and the mean absolute error (MAE) were lower than the ones predicted by statistical models.
Unevenness, Artificial neural network, Regression function, Predicting yarn unevenness, Winding