Optimization of top roller diameter of ring machine to enhance yarn evenness by using artificial intelligence

Transcription

1 Indian Journal of Fibre & Textile Research Vol. 33, December 2008, pp Optimization of top roller diameter of ring machine to enhance yarn evenness by using artificial intelligence M Ghane, D Semnani a, R Saghafi & H Beigzadeh Department of Textile Engineering, Isfahan University of Technology, Isfahan, Iran Received 26 November 2007; revised received and accepted 7 May 2008 Attempts have been made to improve cotton and cotton/polyester (35:75) yarns regularity by optimizing front top rollers diameter of the ring machine using a self-organized Kohonen neural network. The diameters of the top rollers have been reduced in stages and in each stage yarns are produced. The unevenness as well as the imperfections of the produced yarns have been measured. The results show that the unevenness of the yarn decreases with the decrease in top rollers diameter up to optimum diameter beyond which the unevenness of yarns increases rapidly as the top rollers diameter decreases. These optimum values are different in the cases of cotton and cotton/polyester yarns. Kohonen neural network has been applied to observe optimum values of top rollers diameter in case of each yarn type. The optimum diameter of top rollers as estimated by neural network is found to be 27.5 mm for most of the cotton and cotton/polyester (35:75) yarns. Keywords: Artificial intelligence, Cotton, Cotton/polyester yarn, Top roller diameter, Yarn irregularity IPC Code: Int.Cl. 8 G06N 3/02 1 Introduction The studies on the effects of roller drafting parameters on the quality of the yarn have indicated that the drafting quality at the ring frames is governed by the top roller weighting, apron spacing and top roller settings. 1-4 Drafting rollers and aprons, the essential components of the drafting system, are of the great importance in determining process efficiency and yarn quality. The influence of the cots material on the final yarn quality has been established as these components are in a direct contact with the fibres. Many researchers have studied this important subject. Broome 5 investigated the influence of the cots material on the yarn properties and concluded that among the properties surveyed, the cots hardness had the greatest effects on the yarn strength and evenness. Cots usually are made of soft rubber. 5 Tailor and Graham 6 concluded that the front drafting roller load impressed the spinning strength of cotton and the fibre cohesion (resistance to slippage) in the nip of the front drafting rollers is very important. After a period of time the surface of the cots is damaged and cracked in its circumference. This cracks and uneven surface of the cots will cause the slipping of the fibres during the drafting process. The common way to remove the cracks and roughness of a To whom all the correspondence should be addressed. d_semnani@cc.iut.ac.ir/dariush_semnani@hotmail.com the cots surface is grinding the surface of the cots by a grinding machine. In this way, the diameter of the top front roller will be reduced. This, in return, has some influences on yarn quality. The main aim of this work is to study the effect of top roller diameter on the quality of the yarn. The effect of front roller pressure on the yarn properties during the decrement of the top roller diameter has been studied. There are many researches in the field of considering and predicting yarn properties, specially yarn mechanical properties by using simple supervised neural networks. Evaluating and comparing the physical properties of spliced yarns have been presented as an application of neural network techniques in comparison with regression analysis method. 7 Also, the breaking elongation of ring-spun cotton yarns has been considered to predict optimum raw material and spinning characteristics by using various mathematical, statistical, and artificial neural network models, while artificial neural network presented very careful prediction. 8 Prediction of tensile properties of spun yarns and worsted spinning performance is another application of neural network modeling. 9, 10 Optimal neural network architecture and performance of selected network have been 11, 12 considered in prediction of cotton yarn properties. The neural networks presented by previous researchers are supervised networks which are very sensitive in error of training and testing while textile

2 366 INDIAN J. FIBRE TEXT. RES., DECEMBER 2008 data are limited in plenty and variety. By those limitations of available data, the use of back propagation neural networks of mentioned modeling methods is not sufficiently acceptable. However, these networks are a little applicable for prediction but are not suitable for optimizing or finding optimum situations of yarn characteristic features for some desired properties. 12 In this work, an unsupervised self-organizing neural network has been designed to find out optimized spinning roller diameter and pressure. 2 Materials and Methods 2.1 Sample Preparation and Testing Cotton and polyester fibres, having the properties : cotton fineness 3.39 µg/inch, mean length mm, bundle strength 22.2 cn/tex; and polyester fineness 1.4 dtex, mean length 38 mm, tenacity 42.1 cn/tex, elongation-at-break 22.3%, were used for the study. The yarns were produced in a carded cotton spinning system. The fibres were blended in blow room stage. INGOLASTD RB10 ring frame with spindle speed of rpm was used to spin cotton and cotton/polyester (35:65) yarns. The linear density and the twist multiplier of the yarns were 20Ne and 3.8 respectively. At the first stage of the experiment the cots diameter was selected to be 29 mm and they were reduced to 28.5, 28, 27, 26.5 and 26 mm in stages by a grinding machine. After reducing the cots diameter the cotton and cotton/polyester yarns were spun in two different ways, namely (i) the pressure of the front top rollers was fixed and no adjustment was made on the ring pendulum arms, and (ii) after each decrement of the top roller diameter the pressure and the pendulum arms of the drafting system were. The other parameters of the ring machine were unchanged during the experiments. The spinning production line for all roving samples was also identical. After the production of yarns, their different properties were measured. Primary consideration of yarn irregularity was given to conventional method of USTER mass irregularity evaluation. The yarns irregularity parameters including U%, neps and thick-thin places per 1000 m of the yarns were determined by means of an USTER Tester 4. As the yarn irregularity is effective on mechanical properties, the yarn elongation and breaking load were also obtained using a Zwick tester. The extension rate and the gauge length of the tensile tester were set to be 5cm/min and 50cm respectively. In each case 30 samples were tested and the mean value was calculated. These experiments are primary evaluation of yarn quality changes based on top roller diameter and pressure. The results of conventional experiments were analyzed to achieve an overall situation of spinning top rollers. 2.2 Algorithm of Kohonen Self-organizing Network To find out optimum situation of top rollers, specially diameter and pressure, artificial intelligence technique was applied on experimental data. Artificial neural networks (ANNs) technique was applied on various data for classification, clustering, prediction and optimization. Many of ANNs are supervised networks which have two sets of input and output data. These kinds of networks are suitable for the prediction of sets of output data based on input ones. Competitive neural nets are kinds of unsupervised self-organizing neural nets which are applicable for classification, clustering and specially optimization of both dependent and independent data. There are many networks, such as Maxnet, Mexiacan Hat, Hamming and Kohonen, in the category of competitive networks. 13 The Kohonen network is a competitive unsupervised network which is very suitable for kinds of data without topology preserving map. This network has self-organizing topology, where the network tries to organize suitable topology and weights of nodes for generation of exemplar vector similar to input vector. The architecture of Kohonen network is shown in Fig. 1. Neighborhood of each node is designated by radii 2, 1 and 0 in a one dimensional topology. The weight vector for a cluster unit serves as an exemplar of the input pattern associated with that cluster. During the selforganizing process, the cluster unit whose weight vector matches the input pattern most closely Fig. 1 Topology of Kohonen ANN used for finding optimum cots diameter

3 GHANE et al.: OPTIMIZATION OF TOP ROLLER DIAMETER OF RING MACHINE 367 (typically the square of the minimum Euclidean distance) is chosen as the winner. The winning node and its neighboring nodes update their weights As the optimum data for cots diameter are desired to adjust top rollers, the results of conventional experiments were introduced to a self-organizing Kohonen network as input vectors. The network was selected in recurrent form (Fig. 1) with four nodes sets of cotton/polyester and 100% cotton yarns in both the situations of and not top roller pressure in seven decreasing cots diameters as 28 input vectors. Each vector included five data of U%, number of thick places, number of thin places, number of neps and breaking load respectively from 1 to 5 elements of each input vector. The training rate of Kohonen network was as 0.5 in first epoch. The training rate was decreased by 0.95 in each epoch to increasing convergence probability of network. The neighborhood radius was selected as 3. In each epoch, distances for all the nodes from exemplar vector "D j " were calculated and the node J with minimum distance of D J min was selected as winner node. All of the weights of J th node were changed by using following equation: W new) = αx + (1 α) W (old) (1) ij ( ij ij where W ij is the weights of node i,j; α, the training rate; and X ij, the input value ij. The vector of competitive indexes C can be calculated using the following equation, considering previous values of each epoch: C = W. (2) i X i where W is the final weight matrix achieved by training of ANN; and X i, the input vector of i th conventional experimental data. The maximum value of final competitive indexes indicates the winner cluster node which is referred to most closely pattern to the optimum situation. 3 Results and Discussion 3.1 Effect of Cots Diameter on U% The unevenness of polyester/cotton yarns and cotton yarns at different top roller diameter and pressure adjustment situations are shown in Tables 1 and 2 respectively. As can be seen in the case of both cotton and cotton/polyester yarns the irregularity (U%) decreases as the cots diameter reduces until a minimum point is reached, and after this minimum point the irregularity of the yarn tends to increase as the diameter of the cots reduces. This minimum point Top roller diameter mm Table 1 Properties of polyester/cotton yarn at different top roller diameter and pressure adjustment situations U% Thick/1000m Thin/1000m Neps/1000m Breaking load, cn Top roller diameter mm Table 2 Properties of 100% cotton yarn at different top roller diameter and pressure adjustment situations U% Thick/1000m Thin/1000m Neps/1000m Breaking load, cn

4 368 INDIAN J. FIBRE TEXT. RES., DECEMBER 2008 is taken as the optimum diameter of the cots for obtaining the minimum irregularity of the yarn. The results also reveal that the optimum diameter of the cots also depends on the pressure of the top rollers. When the pressure of the top rollers is fixed, and no adjustment is made during the experiments, the optimum diameters of cots are found to be about 27.5 mm and 28 mm for cotton/polyester and cotton yarn respectively. However, as the pressure of the top rollers is in each case the optimum diameters of the cots are slightly different; the values are found to be about 28.5 mm for both cotton/polyester and cotton yarns. It may be concluded that when the pressure of the top rollers is in accordance with the top rollers diameter, the optimum diameters are identical for both cotton and polyester/cotton yarns. 3.2 Effect of Cots Diameter on Breaking Load The breaking load of polyester/cotton yarns and cotton yarns at different top roller diameter and pressure adjustment situations are shown in Tables 1 and 2 respectively. As can be seen in the case of both cotton and cotton/polyester yarns the breaking load shows opposite trend in compression to the irregularity. This is expected because the irregularity Table 3 Final weights of ANN nodes and competitive indexes Category Vector name Top roller diameter mm U% Thick places Final weights on nodes Thin places Neps Breaking load Competitive index Class I: polyester/cotton yarn with pressure I I I I I I I Class II: polyester/cotton yarn with non pressure II II II II II II II Class III: cotton yarn with pressure III III III III III III III Class IV: cotton yarn with non pressure IV IV IV IV IV IV IV

5 GHANE et al.: OPTIMIZATION OF TOP ROLLER DIAMETER OF RING MACHINE 369 and breaking load are inversely related to each other. The breaking load of the yarn increases as the cots diameter reduces until a maximum point is reached, and after this maximum point the breaking load of the yarn tends to decrease as the diameter of the cots reduces. This maximum point is taken as the optimum diameter of the cots for obtaining the maximum breaking load of the yarn. The results also reveal that the optimum diameter of the cots also depends on the pressure of the top rollers. When the pressure of the top rollers is fixed, the optimum diameters of cots for obtaining the maximum breaking load are found to be about 27.5 mm and 28 mm for cotton/polyester and cotton yarns respectively. However, as the pressure of the top rollers is in each case the optimum diameters of the cots are slightly different; the values are found to be about 28.5 mm for both cotton/polyester and cotton yarns. Although the optimum cots diameter is found based on the results of graphs for different situations of pressure and not, the real optimum value is not accessible which can be presented as suitable cots diameter for most of the situations and yarn types. Thus, the graphical analysis cannot present general investigation of different properties of U%, thick places, thin places, neps and strength of yarn as breaking load. Specially, all of these properties are dependent together. Therefore, it is needed to find out optimum diameter with neural networks of self-organizing Kohonen to ensure the best diameter as real optimum value. For this purpose, the results of experiments are arranged in four classes of polyester/cotton and 100% cotton yarns in both situations of and not top rollers pressure. After that, the arranged data are introduced to ANN as input vectors. After training of neural network, optimum vector is selected based on the final minimum distance, where the network reaches to convergence situation while weights change is less than threshold of The training rate is done for epochs in stop error of weight changes equal to The training rate is found to be 0.5 in first epoch and in the last epoch it reaches to The competitive index moves towards the maximum value during training process for winner cluster node, while competitive indexes of other nodes are arranged lower than winner index. After training process the maximum value which is most closely to winner index is the optimum choice. This means that the competitive index is maximum when U%, thick places, thin places and neps values are minimum and strength value is maximum. The results of final calculated weights and competitive index (which is offered to yarn quality) are presented in Table 3. The maximum value of competitive index is for cots diameter of 27.5 mm where polyester/cotton yarn is produced with pressure of top rollers. Also, the pressure situation in cots diameter of 28 mm for 100% cotton yarn presents second maximum value of competitive index as This may be due to the difference in frictional and mechanical properties of cotton and polyester fibers. However, the optimum cots diameter is in neighborhood of 28 mm based on competitive indexes. 4 Conclusions The results of conventional experiments show that the unevenness of the yarn U % decreases with the decrease in top rollers diameter up to an optimum diameter and after that the (U %) of yarns increases rapidly as the top rollers diameter decreases. These optimum points are different in the cases of cotton and cotton/polyester yarns. Kohonen neural network has been applied to forecast optimum value of top rollers diameter in case of each yarn type. The result of neural network suggests that the optimum cots diameter is 27.5 mm for both types of yarns. Also, the cots diameter can be to 28 mm in case of cotton/polyester yarns. Industrial Importance : In the spinning ring frames, the surface of the top roller cots is damaged after a period of the operation time. The common way to remove the cracks and roughness of the cots surface is grinding. In this way, the diameter of the top roller will be reduced. An artificial intelligence technique has been introduced to find the optimum diameter of cots after grinding for both the cotton and polyester/cotton yarns. The optimum cots diameter is very important in the period of cots grinding operation. References 1 Balasubramanian N, A study of the irregularities added in apron drafting, Text Res J, 39 (1969) Balasubramanian N, Optimize apron spacing to get better yarn quality and ring frame performance, BTRA Scan, 8 (1972) 3. 3 Newton F E, Burley S T & Laferney P E, A new approach to evaluation of cotton fiber quality, Text Res J, 43 (1973) 1. 4 Sripathy R, Top roller pressure, its effect on end breaks and yarn quality under varying cot hardness, Proceedings, 13th Joint Technological Conference of ATIRA, BTRA, SITRA and NITRA (ATIRA, Ahmedabad), 1972, 75.

6 370 INDIAN J. FIBRE TEXT. RES., DECEMBER Broome R W, An investigation of the effect of cot material, cot hardness, and roll pressure on yarn properties and drafting efficiency in Saco Lowell Duo-Roth spinning system, M Sc thesis, Institute of The Textile Technology, Charlottesville, Virginia, Taylor R A & Graham J S, The influence of front-roll coverings on the strength of cotton, Text Res J, 49 (1979) Cheng K P S & Lam H L I, Evaluating and comparing the physical properties of spliced yarns by regression and neural network nechniques, Text Res J, 73 (2003) Majumdar P K & Majumdar A, Predicting the breaking elongation of ring spun cotton yarns using mathematical, statistical and artificial neural network models, Text Res J, 74 (2004) Zeng Y C, Wang K F & Yu C W, Predicting the tensile properties of air-jet spun yarns, Text Res J, 74 (2004) Beltran R, Wang L & Wang X, Predicting worsted spinning performance with an artificial neural network model, Text Res J, 74 (2004) Babay A, Cheikhrouhou M, Vermeulen B, Rabenasolo B & Castelain J M, Selecting the optimal neural network architecture for predicting cotton yarn hairiness, J Text Inst, 96 (2005) Chattopadhyay R & Guha A J, Performance of neural networks for predicting yarn properties using principal component analysis, J Appl Polym Sci, 91 (2004), Fausett L, Fundamentals of Neural Networks (Prentice Hall International, New York), Bertsekas D P, Network optimization (Athena Scientific Publishing, Athens), Kohonen T, The self organizing map, Proceedings of IEEE, 78 (1990) 1464.