Haghighat, E.; Johari, M. S.; Etrati, S. M.; Tehran, M. A. Study of the Hairiness of Polyester-Viscose Blended Yarns. Part III - Predicting Yarn Hairiness Using an Artifcial Neural Network. FIBRES & TEXTILES in Eastern Europe, 2012, 20, 1(90) 33-38. 33 Study of the Hairiness of Polyester-Viscose Blended Yarns. Part III - Predicting Yarn Hairiness Using an Artifcial Neural Network Ezzatollah Haghighat, M. Safar Johari, *Seyed Mohammad Etrati, M. Amani Tehran Department of Textile Engineering, Amirkabir University of Technology, Tehran, 15875-4413, Iran *Corresponding author: S. M. Etrati E-mail: elham@aut.ac.ir Abstract The hairiness of blended yarns is infuenced by several parameters at the ring frame. For this reason, it is necessary to develop a model based on experimental evidence that includes all known processing factors. The generalised from of this model is a candidate for predict- ing yarn hairiness. In this paper, an artifcial neural network and multiple linear regression were used for modelling and predicting the hairiness of polyester-viscose blended yarns based on various process parameters. The models developed were assessed by applying PF/3, the Mean Square Error (MSE), and the Correlation Coeffcient (R-value) between the actual and predicted yarn hairiness. The results indicated that the artifcial neural network has better performance (R = 0.967) in comparison with multiple linear regression (R = 0.878). Key words: polyester-viscose blended yarn, hairiness, artifcial neural network, ring frame parameters. n Introduction Several authors have studied yarn hairi- ness and the effect of fbres and process parameters on it. Barella [1] stated that yarn hairiness is defned as the fbre ends and loops protruding from the main yarn body. Hairiness is one of the most impor- tant yarn characteristics, which affects weaving, knitting, dyeing and fnishing processes in textiles [2]. The importance of yarn hairiness as a factor infuencing the handle, appearance, thermal insula- tion, and pilling propensity of fabrics is well known. Hairiness is generally con- sidered as a negative attribute of spun yarns. However, some hairiness is also required for specifc yarns to produce good handle and comfort properties [3 - 5]. Beltran et al [6] studied the in- fuence of the hairiness of worsted wool yarns on the pilling propensity of knit- ted wool fabrics. The results suggested that a relatively large reduction in yarn hairiness was needed to achieve a mod- erate improvement in fabric pilling, and that the nature of yarn hairiness was also a key factor in infuencing fabric- pilling propensity. Canoglu and Tanir [7] studied the hairiness of polyester/ cotton blended yarns with different blend ratios. They found that among the yarns produced, the best result was obtained from the blend yarn with a polyester/ cotton ratio of 33/67. Altas and Kadoglu [8] investigated the effect of cotton f- bre properties and linear density on yarn hairiness. They found that yarn hairiness increases with an increase in yarn linear density. According to Pillay [9], torsional rigidity, fexural rigidity and fbre length are the major cotton fbre properties for determining the level of yarn hairiness. the basis of spinning parameters. There- fore, this paper presents a feed forward backpropagation model of an ANN and another MLR model for predicting the hairiness of polyester-viscose blended yarn at the ring frame based on signif- cantly effective parameters of the spin- ning system. n Evaluation of the models Artifcial Neural Networks The development and use of neural net- works is part of an area multidisciplinary study that is commonly called neural computing but is also known as connec- tionism, parallel distributed processing and computational neuroscience. The ANN is a powerful data-modeling tool that is able to capture and represent each kind of input-output relationship [17]. In the feld of textiles, artifcial neural net- works have been extensively studied dur- ing the last two decades. In the feld of spinning, previous researches have con- centrated on predicting yarn properties and spinning process performance using fbre properties or a combination of fbre properties and machine settings as the in- put of neural networks [18]. Multi-layer perceptron neural networks are responsible for approximately 80% of all practical application. A typical feed forward network with a single hidden layer is shown in Figure 1 (see page 34). In the MPL, the units are arranged in distinct layers, with each unit receiving weighted input from each unit in the pre- vious layer. A neural network is usually trained so that a particular input leads to a specifc output. The process of training Viswanathan et al. [10] demonstrated the relationship between fbre quality param- eters and yarn hairiness. The effect of processing factors such as the drafting system, winding section factors, and yarn parameters have already been reported [11-12]. Some researches have attended to the pre- diction of yarn hairiness using an Artif- cial Neural Network (ANN) and multiple linear regression (MLR) models based on fbre characteristics, yarn properties and processing factors. Khan et al [13] evaluated the performance of multilayer perceptron (MLP) and MLR models for predicting the hairiness of worsted-spun wool yarns from various tops, yarns, and process parameters. Their results indi- cated that the MLP model predicted yarn hairiness more accurately than the MLR model. Jackowska-Strumillo et al [14] stated that the hairiness of cotton – poly- ester blended yarns could be predicted using an MLP artifcial neural network based on the characteristics of feeding streams. Beltran et al. [15, 16] examined pattern recognition algorithms as a prac- tical alternative to existing experimental techniques for the prediction of spinning performance, in which they successfully predicted worsted spinning performance with an ANN model and showed that the MLP approach was slightly better than the other approaches. Their results also indicated that the ANN method provided a more precise mill specifc spinning per- formance than the traditional experimen- tal technique. Few researchers have studied the predict- ing of the hairiness of blended yarns on