Ring Spun Yarn Parameters Impact on Composite Yarn Quality Model

Transcription

1 Rng Spun Yarn Parameters Impact on Composte Yarn Qualty Model Hajer Soud Textle Engneerng Laboratory LGTex of Iset of ksar Hellal, B.P 68, Avenue Hadj Al SOUA, 5070 Ksar Hellal, Tunsa E-mal: Mehd Sahnoun Textle Engneerng Laboratory LGTex of Iset of ksar Hellal, B.P 68, Avenue Hadj Al SOUA, 5070 Ksar Hellal, Tunsa E-mal: Morched Chekhrouhou The Beaux Arts School of ISET of Sfax, 34 Avenue 5 août, Sfax 3069, Tunsa E-mal: morchedchekhrouhou@yahoo.fr Abstract In the present study, we nvestgate the effect of fbers and constructon parameters on the overall rng spun yarn qualty. We develop for the case a global qualty ndex for the rng spun yarn. Ths part of the study was acheved by usng Derrnger and Such desrablty functons. Then, we have appled an artfcal neural networks model to these parameters. We base our research work on a predcton model to represent global rng spun yarn qualty. Ths model s developed by optmzng fbers parameters, yarn count and twst. Fnally, we tred to search the mpact of all these parameters on the neural networks model whle algnng a confdence nterval. Keywords: rng spun yarn; qualty; desrablty functons; neural networks; confdence nterval.. Introducton In spte of the fast-fashon movement, denm clothes are usually found n the major wardrobes. The competton that denm jeans are facng from other product categores s ntensfed by shoppers beng dsapponted about recent changes n the qualty of fabrc and fber content of apparel products. Hence, to satsfy customers, the fabrc should respond to an optmal global qualty. To reach ths goal, dfferent fbers crtera should consequently be satsfed at the same tme. For ths case, we have frstly developed an ndex generalzng rng spun yarn global qualty by studyng nne yarn prncpal parameters. In the second stage, we developed a model for predctng the overall qualty of the present cotton yarn by combnng major fbers parameters. Ths study has been acheved by usng a back propagaton neural networks. The last part of the present work studes the mpact of fbers factors and yarn structural parameter on the neural network model. 2. Materals and methods In order to study rng spun yarn overall qualty, we have used the nternatonal standards usng the Uster Tester 3 and Uster Tensorapd 3 testng systems by consderng the major known yarn characterstcs (Table ). In ths survey, we study the followng yarn aspects (Table ): The statstcal summary of rng spun propertes measurements s gven n table 2 as follow: For fber propertes, we have used a large database wth the followng statstcal summary measurements (Table 3). 26

2 Table : Yarn propertes Yarn propertes Instrument Symbol Tenacty (CN/ Tex) Uster tensorapd 3 RKM Tenacty evenness Uster tensorapd 3 CVRKM Breakng elongaton Uster tensorapd 3 E% Work Uster tensorapd 3 TR Regularty Uster tester 3 U% Number of thck ponts Uster tester 3 THIK Number of thn ponts Uster tester 3 THIN Number of neps Uster tester 3 BOUT Harness Uster tensorapd 3 PILO Table 2: Summary statstcs for rng spun yarn propertes RKM CVRKM E% TR U% THIK THIN BOUT PILO Mnmum value Maxmum value Standard devaton Mean Yarn characterstcs are combned to provde a measure of the composte or overall desrablty D of the mult-response system (equaton ()). The composte desrablty D represents the weghted geometrc mean of the ndvdual desrabltes ( d ) for rng spun yarn responses []; [2]. Table 3: Summary statstcs for fbers propertes Fber property Symbol Mean Standard Mnmum Maxmum value devaton value value Mcronare ndex(µg/nch) Mc Upper Half Mean Length (UHML) (0-3 m) Len Short Fber ndex (%) Sf Strength (CN/tex) Str Elongaton (%) Elg Trash count Tr cnt Greyness (color reflectance) Rd w w w2 D = d d2... w = w ; d wn n () w are the weght of th response. d : Derrnger and Such ndvdual desrablty functon ( d ) by usng the provded goals and boundares for each response [3]. The goal of each response can be one of the three followng choces: 262

3 Mnmzng a response. Targetng a response. Maxmzng a response. 3. Results and dscusson 3.. Neural network modellng In ths study, we mplemented a back-propagaton neural network [4]. The network structure s composed of one sngle hdden layer connected to an nput layer and an output layer. The nput unts consttute fber propertes summarzed n table3. Snce we have studed nne yarn responses (table ), we should have one neural networks model for Inputs Output each response. In our case, we have reduced these models nto one sngle network model. The output of the model s the geometrc mean of the ndvdual desrabltes of yarn responses (or composte yarn qualty; equaton ). We have tred the back-propagaton neural network wth one hdden layer and dfferent hdden nodes and epochs untl havng a correlaton coeffcent near to one and small and comparable errors test and tran. These 3 condtons determne the good performance of the model [5], [2]. Table 4: Input and output parameters of the neural network model Fber propertes Constructon parameter Rng spun yarn global qualty Mcronare ndex(µg/nch) Upper Half Mean Length (UHML) (0-3 m) Short Fber ndex (%) Strength (CN/tex) Elongaton (%) Trash count Greyness (color reflectance) Yarn count Twst value DG Rng Mc Len Sf Str Elg Tr cnt Rd Nm TORS Fgure : Learnng curve of the artfcal neural network model. As shown n the learnng curve n Fgure, the rght number of epochs reached s 00. In fact, after 00 teratons, the algorthm has converged to a mean square error generated by the tranng data RMSE = 0.03 and a root mean square tran error generated by the test data RMSE test = 0.09 as follow: o RMSE tran : The root mean square error generated by the tranng data (Equaton 2). 263

4 N 2 2 ( D g( x, b) RMSE = tran (2) N N : Number whch represents 80% of database. D : Yarn qualty ndex calculated from the tranng database correspondng to the nput x. g ( x, b) : Yarn qualty ndex calculated by the neural network correspondng to the nput weght value b. y g( x, b) : The ndvdual error o x for a RMSE : The root mean square error test generated by the test data n Equaton 3: P 2 2 ( D g( x, b) RMSE = test (3) P P : Number of test data places = 236 = 20% of nput-output data pars were used as the test set Input varables mpact on neural network model Twst and yarn count effects As shown n fgure 2(a), the mpact of twst (TORS) on the neural network model has shown a notceable varaton of yarn global qualty (DGRng) wth a varaton of 35%. The upper and lower curves show the bounds of confdence nterval. Based on our sample data, the default confdence nterval we used s 95%. In our case, the confdence nterval s respected and s consdered acceptable as these two curves les nsde the nterval 0%. 0,60 0,55 0,50 0,20 0,5 350,00 400,00 450,00 500,00 550,00 600,00 650,00 TORS 0,55 0,50 0,20 0,5 0,0 0,00,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 9,00 20,00 2,00 NM Fgure 2: Twst and yarn count Impact on yarn global qualty model In fgure 2(b), yarns wth hgher counts have conferred a slghter qualty ncrease for the yarn count gamma studed wth a varaton of about 0%. The confdence nterval s also restrant for ths fgure. Nevertheless, we remark for both fgures a slght larger confdence nterval n the extreme borders than nsde the data samples. Ths s reported to an nferor sample sze n the borders. If we want a default confdence nterval of 99%, we should mprove more sample sze. 264

5 Fbers parameters mpact on the neural model Length and mcronare mpact on yarn global qualty model Short fber ndex and trash count mpact on yarn global qualty model 0, ,20 0 0,55 4,00 5,00 6,00 7,00 8,00 9,00 0,00 sf 0,50 0,46 0,44 0,43 0,42 0,4 0,39 0,38 0,37 0,36 0,34 27,00 27,50 28,00 28,50 29,00 29,50 len 4,00 4,0 4,20 4,30 4,40 4,50 4,60 4,70 mc Fgure 3: Length and mcronare mpact on yarn global qualty model The mpact of fber length (fgure 3 a) and mcronare ndex ( fgure 3 b)on the neural network model show a monotone ncrease of the overall yarn qualty (central curve). Yarns wth hgher length and mcronare ndex have generated a better global qualty of rng spun yarn. For Mcronare ndex, whch s a combnason of fber fness and maturty [6];[], the abscssa data vares between 4 and 4.7. Ths varaton covers a large gamma of dfferent orgns of cotton fbers used n the manufacture where the present research was acheved (Indan cotton, Egyptan cotton, Madagascar cotton, Chnese cotton, Bangladesh cotton). In spte of the restrant nterval of varaton of ths parameter, ts mpact on yarn qualty model s relatvely sgnfcant. The confdence nterval s superor of 95%. For both fgures, the confdence nterval s larger n the lmt borders whch are reported to an nferor sample sze. 2,50 5,00 7,50 0,00 2,50 5,00 7,50 20,00 22,50 tr cnt Fgure 4: Short fber ndex and trash count mpact on yarn global qualty model The mpact of short fber ndex sf and trash count trcnt on the neural network model have shown for both curves a slght decrease varaton (about 4%) n the all nterval of varaton of these two fbers parameters (fgure 4). Ths mpact agrees the statstcal results of Ben Ammar S. and Halleb N. [3] and El Moghazy [4]. The confdence nterval s tolerable for both fgures as t s superor of 94% Elongaton and greyness mpact on yarn global qualty model 0, ,00 74,00 75,00 76,00 77,00 78,00 79,00 80,00 rd 265

6 Fgure 5: Elongaton and greyness mpact on yarn global qualty model The neural network model shows a slght ncrease effect of color reflectance on yarn global qualty (fgure 5) (0,38 to ). Generally, cotton color reflectance defnes fber sellng to get a better yarn or fabrc whteness degree or to facltate dyeng. Through ths result, cotton reflectance has also an nfluence on composte yarn qualty. Ths result agrees El Moghazy statstcal study of rng and open end yarns [4]. For fber elongaton mpact (fgure 5 b), the global yarn desrablty ndex reveals a smlar varaton as color reflectance of about 5%. For both fgures, the confdence nterval s respected and s superor n the central of the curves Fber Strength mpact on yarn global qualty model The neural network model developed has conferred to a yarn global qualty varaton of 8% (fgure 6). Ths result concdes wth the real database. Hence, yarns wth hgher fber strength present a better yarn strength so a better yarn global qualty. The confdence nterval s acceptable n the mean of the nterval of fber strength ([25; 35]). For hgher fber strength, the confdence nterval s about 90%. To get a better confdence nterval, data samples needs a lttle more mprovement , ,00 5,50 6,00 6,50 7,00 7,50 8,00 8,50 elg 4. Concluson To study the qualty of rng spun yarns, whle consderng fber and constructon parameters, we have used a combnaton of two approaches: the desrablty functons and neural networks. The frst approach has contrbuted to one global yarn qualty ndex. The second approach has allowed to predct ths ndex. In a thrd stage of ths study, we tred to show how the neural network model works aganst these parameters. In fact, we have not studed the effect of a number of fber and constructon parameters on global rng spun yarn qualty, but rather we tred to vew the mpacts of these factors on the neural network model. In addton to confdence nterval, ths has showed, somehow, a manner of the model effcency. 5. References [] Soud H., Babay A., Sahnoun M., Chekrouhou M., Rng spun and slub yarns qualty optmsaton by usng the desrablty functon, Autex Research Journal, Vol. 8, No3, 72-76; September 2008 [2] Soud H., Sahnoun M., Babay A., Chekrouhou M., Slub Yarn Qualty Optmzaton by Usng Desrablty functon and Neural Networks, Journal Of Appled Scences, ISSN / DOI: /jas. 20 Asan Network for Scentfc Informaton. 20 [3]Derrnger GC, Such R., Smultaneous optmzaton of several response varables, Journal of Qualty Technology, 2(4):p ;980 [4] Dreyfus G., Martnez M., Samueldes M., M.B. Gordan, F. Badran, S. Thra, L. Hérault. Réseaux de Neurones: Méthodologe et Applcaton, Edtons Eyrolles, Pars, vol.. (2002) [5] Msahl S., Hadj Taeb A. et Sakl F. "A new approach for predctng the knt global qualty by usng the desrablty functon and neural networks" Journal of Textle Insttute; vol. 97; N ; pp: 7-23; 2006 [6] Harrngton EC., The desrablty functon, Industral Qualty Control, pp: ; (965) [7] El Mogahzy Y., Broughton R., Lynch JR. and W.K.. A statstcal approach for determnng the technologcal value of cotton usng HVI fber propertes, Textle Research Journal, vol 60 n 9; pp ; September 990. [8] Derrnger GC, Such R., Smultaneous optmzaton of several response varables, Journal of Qualty Technology; vol 2; n 4; pp 24 29; ,00 26,00 27,00 28,00 29,00 30,00 3,00 32,00 33,00 34,00 35,00 str Fgure 6: Fber strength mpact on yarn global qualty model 266

7 [9] Abd Jell R., Zeng X., Koel L. and Erwuelz A., Predcton of plasma surface modfcaton of woven fabrcs usng neural networks, IJARTEX [0] Chattopadhyay R., Applcaton of neural network n manufacture, Indan journal of fbre & textle research, vol 3 pp60-69, march 2006 [] El Mogahzy Y., Roy M. Broughton Jr. and W. S. Perkns,Clemson: M.S. Eson, C. D. Rogers, H. Behery, S.R. Matc-Legh,NCSU: Moon W. Suh, Wam Oxenhaum, Jon P. Cotton Fber Qualty: Characterzaton, Selecton, and Optmzaton Fle: A92Cl PI(s): Auburn: RustAnnual Report Endng, 993 [2] Ramesh Mc, Rajamanachan R. et Jayaraman; The predcton of yarn tensle propertes by usng artfcal neural networks, Textle Research Journal; vol 86; n 3; pp ; 995. [3] Ben Ammar S., Halleb. N., Predcton of the mechancal behavour of open end and rng spun yarns, Journal of Appled Scences; vol 9; n 8; pp ; [4] El Mogahzy Y., Broughton R., Lynch JR. and W.K.. A statstcal approach for determnng the technologcal value of cotton usng HVI fber propertes, Textle Research Journal, vol 60 n 9; pp ; September 990. [5]P.J Morrs, J. H. Merkn and R.W. Rennell, Modellng of yarn propertes from fbre propertes, Journal of Textle Insttute 90 part No3; :