ThePrediction of wool Ring Yarn Properties from Fibre Length Blend byUsing Linear Regression Models
<Lecturer’sname and course number>
Inthe paper, some models for foretelling the most crucial ring yarnquality characteristics are made by using AFIS (Advanced FibreInformation System) data. Thread count, roving, and twist propertieshad a selection as predictors because of their significant effect onthe yarn properties. There was a production of 180 ring yarns fromvarying 15 cotton blends on the same ring-spinning machine, on samespinning angles and under the same conditions at Ege UniversityTextile and Apparel Research-Application Centre. Every blend had tospin in four counts (29.53 Tex, 23.63 Tex, 19.69 Tex, and 16.88 Tex)at three independent coefficients of twist (αTt 3639, αTt 4022, andαTt 4404). Linear multiple regression processes followed for theestimates for the properties of the quality yarn. The goodness of fitstatistics showed that our equations had modified R2 values andcolossal R2 (coefficient of multiple determination).
Inaddition to spinning technique, other parameters influence theprocessing behavior. They include processing conditions, the physicalcharacteristics of the fiber, machine parameters, operation stages,final yarn and quality, and production efficiency. The prediction ofthe properties of yarns, with an immense regard for the tensilequalities, forms a fundamental target in a study in the previouscentury. The studies applied two approaches for differentiating theyarn quality. The differentiation is between thread characteristics:a statistical approach, an empirical, and an analytical approach andfiber.
Acommonly used statistical method is the multiple regression method.The method investigates the interdependence of various fiberproperties. Additionally, the method estimates the relativecontribution of a single fiber characteristic to all of the yarncharacteristics. Multiple regression equations have had a discoveryfrom various researchers [1 – 5].
Theapproach, in theory, has foundation on mechanical and physicalprinciples. The principles give us useful information regardingengagements of yarn characteristics and different fiber properties.The models offer huge complications rendering them impossible to usein practicals. The models have a basis on various assumptions, andthe success is hugely dependent on the feasibility of the assumptions[4, 6].
Recently,some researchers [2, 4, 7, and 8] inclined interest in usingartificial neural networks (ANN) to estimate yarn features. Using thelogical structure is beneficial to find out the connection betweenvariables .
Theprimary focus of the investigation was to determine the connectionbetween the fiber measurements obtained by the AFIS instruments andring yarn properties. Additionally, the findings aim to structurebasic models for forecasting yarn properties. Apart from using theAFIS fiber characteristics, roving features such yarn count and yarntwist to model yarn properties came in handy.
Regressionanalysis acts as most basic analytical formula for the forecastingof the connection of a contingent variable and one or more non-variables. The approach is advantageous from its ease in unfoldingthe quantitative affiliation between textile materialcharacteristics, hence the reason for using the approach. At first,the type of connections between preferred characteristics(independent variables) and yarn characteristics (dependentvariables) received individual checks by applying correlationanalysis and curve estimation. Analysis of the statistics indicatedan almost linear relationship between yarn properties and fibertraits. The relationship makes the linear multiple regression modelsselected for this study. The onward stepwise way had a selection forlinear regression study.
Beforethe regression analysis, we tested for colinearity in each variable.The outcomes suggest that there is a strong correlation betweenlength measurements by weight and number based values. All distancemeasurements have excellent correlation coefficients with yarnproperties, but autocorrelation between fiber length measurements canlead to some illogical signs on significant values of the regressionequations. As a result, the regression equations manipulatingexplanatory variables are very uneven. Using the regression equationsgives imprecise results in lone cases. Consequently, we neglectedlength measurements by number (L (n), UQL (n) and SFC (n). We alsoexcluded visible foreign matter (VFM) from the group of variablesbecause of the same reason (it highly correlated with the trashcount).
SPSS11.0.1 software performed the Statistical analyses
4.Results and Discussion
4.1.Predicting Yarn Tenacity
Tensilecharacteristics of a spun yarn are hugely paramount in thedetermination of the yarn quality for they directly have an effect onthe knitting efficiency and the winding, and to the weft breakagesand warp all through weaving. Hence, it is crucial to determine theyarn and fiber parameters sway yarn tensile characteristics and ifprobable, to obtain the active connection linking them. At large,abundant arithmetic and experimental models have had establishmentsto approximate the lone yarn tenacity [3, 10, 11], and CSP (CountStrength Product) [1, 3, 12, 13] using some yarn parameters and fibercharacteristics. Hearle  investigated various arithmetical andempirical studies regarding yarn strength whose publishing wasbetween 1926 and 1965. Hunter  stated that over 200 papers hadpublished about the estimation of yarn quality variables, withinterest on tensile properties, up to the year 2004. Apparently,fiber strength lays a critical factor for yarn tenacity, but AFIScannot measure thread strength. Instead of power, grain diameterbecomes the first property among those of AFIS, besides we discoveredhigh negative correspondence coefficient between yarn tenacity (R =-0.929) and fiber diameter. This adverse correlation implies that thelower the diameter of thread (i.e. greater number of fibers in theyarn cross-section) the greater the yarn tenacity. Our regressionstudy illustrates the connection evidently. Table 3 representsregression coefficients of variables, a significance level of eachvariable and t-values. The array of functions in the table showstheir comparative importance for the representation. Signs (positiveor negative) of regression coefficients of functions demonstrate thetrack of sway. Upper quartile length, neps count, and dust count areadditional crucial fiber measures for yarn tenacity, adding to fiberdiameter. Yarn tenacity lowered from the increase in dust and neps.As expected, toughness rose from the increase in the upper quartilelength. Figure 1 demonstrates the scatter plot of predicted valuesagainst experimental values and regression line of our model.
4.2.Predicting Yarn Elongation
Fewforecast models have associated with elongation of the strand(cotton) yarns. Aggarwal has proposed mathematical models [15, 16],Frydrych , and Żurek et al. . Hunter developed ANN modelsand statistical models  formed by Majumdar . In linearregression analysis, there should be a direct connection of thedependent variable and every lone variable. Our curvature estimationanalysis revealed the roving count (Tex) value would relate in adirect manner to yarn elongation following the quadratic form below.
Wedisregarded roving count from the group of variables and added QRvvalue. Table 4 illustrates our multiple linear regression analysisfindings. Yarn count mostly influences the breaking elongation. FiberDiameter and Upper Quartile Length are imperative fibercharacteristics for the yarn flouting elongation. Extra vitalmeasures are roving unevenness, yarn twist, and roving count. Rovingirregularity and fiber diameter offer negative effects.
Someresearchers [3 – 5] finalized that yarn elongation has principalinfluence from fiber strength and fiber elongation. It is impossibleto measure fiber potency and fiber elongation on an AFIS device.Therefore, the projecting power of the representation and the R2value are comparatively low down. Figure 2 shows the rife of figuresabout the regression line.
4.3.Predicting Yarn Unevenness
Cross-sectionalthread disparity acts as the fundamental explanation fordisproportion of the fiber. The yarn unevenness has a decisiveinfluence on the thread parameters, and to the Spinning method, themachine parameters, and from the yarn count. Hunter  and Ethridgeet. al.  came up with some representations to resolve yarnunevenness by use of constant fiber variables. In Table 5, linearregression analysis results are presented. Our studies show nepscount and dust affect yarn unevenness. The direction of impact ispositive for fiber diameter, neps, dust, roving unevenness, theroving number, and short fiber content. That means enlargedparameters improved the yarn consistency. Fibre diameter shows thenumber of fiber in yarn cross section. Reduced fiber diameteraugmented the regularity since there was an increase in thecross-sectional length of the fiber. Hunter  showed that increasesin short fiber content increase yarn`s thin places. The draft ratioon ring rotation depends on the roving count. We see that reducedroving count lowered yarn unevenness. However, we must note that inthe research, drafting ratio was between 18.80 and 54.20. As shown inFigure 3, our model shows very high prediction ability.
4.4.Predicting Yarn Hairiness
Hairinessis an undesirable property of the measure of yarn characteristics.There are acceptable measuring devices that determine the hairiness.The devices are comparably new and include the Zweigle HairinessTester and the Uster Tester. Owing to their newness, there has beenthe lesser publication of research articles to estimate the hairinessby manipulating fiber parameters. Table 6 gives the Regressioncoefficients of the variables and the significance level.
Weobserve that fiber diameter is the most imperative factor influencingthe fiber hairiness. Finer kinds of cotton produce less yarnhairiness. Thread count (Tex), roving unevenness, roving count (Tex),and increased nep content enlarged yarn hairiness but yarn twist andUpper quartile length lowered yarn hairiness. Longer fibers givelittle opportunity to project from the yarn stiff to form a hair.Figure 4 shows the disperse plot of predicted values versusinvestigational values and regression line of our model.
Analysesof variance (ANOVA) took place to control the fitness of theregression equations. The ANOVA test results for all models are inTable 8. This chart sums regression and residual sums of squares, Fvalues and a significance level of regression (p-value), and meansquares. The goodness of fit statistics and ANOVA tables prove thatthe predictive powers of our models are extremely high andsignificant at α = 0.01 significance level. The nice results oflinear regression models in elucidation of yarn properties show thatthe relationships between yarn characteristics and our variables(fiber properties, yarn count, roving properties, twist, and yarncount) are nearly linear. Our models pointed out that rovingcharacteristic has a significant effect on all yarn properties.Almost all key factors for yarn properties are Yarn twist and count.In the fiber properties, fiber diameter is the most importantparameter as it maintains significance in every regression equation.Upper quartile length, short fiber content, dust, and neps counts areother important fiber parameters for yarn properties. Our studiesshow that the successful prediction of the yarn properties can befrom the usage of AFIS fiber properties.
Page 2 of 2