Predicting the properties of polyester/viscose blended open-end rotor spun yarns by establishing artificial neural networks and statistical models

The aim of this study is to develop Artificial Neural Networks (ANN) andStatistical models in order to predict the polyester/viscose blended open-end rotoryarn properties before the yarn production.For this purpose, seven different blend ratios of polyester/viscose slivers wereproduced and these slivers were spun with four different rotor speed and fourdifferent yarn counts in rotor spinning machine. Total number of bobbin producedwas 224, however there were 112 different types of bobbins because of thereplication. The physical and mechanical properties (breaking force, tenacity,breaking force, elongation, unevenness, thin place, thick places, neps and hairiness)of these yarns were measured with related test equipments.NeuroSolutions software was used for constructing ANN and backpropagation feed-forward Multi-Layer Perceptron (MLP) network having sigmoidtransfer function and momentum learning rule was used as ANN model. The ANNmodels, having the least prediction errors were selected as the best model for eachyarn property. Design Expert software was used for statistical analyses and simplexlattice design was carried in which mixture*process crossed models were developed.With the regression equations obtained from these models, yarn properties could bepredicted. In conclusion, both ANN statistical models can be used for the predictionof yarn properties, however, the predictions of ANN gave more reliable results thanstatistical models.