YANG Shou－ren，S Gordon

（CSIRO Materials Science ＆Engineering，Geelong，Victoria，Australia）

0 Introduction

Choosing the right cotton in preparation of a lay down，which allows an optimum raw material price to be achieved while meeting the needs of a yarn customer，is of great economic importance to spinners.To help spinners achieve this goal，CSIRO has introduced a cotton fibre and yarn quality management software package，Cottonspec，which aims to predict what a good modern mill can expect using a particular quality cotton for a given yarn under the specified spinning conditions［1］.

Cottonspec utilizes a very large database of industrial spinning results and theoretical yarn structure models to achieve high prediction ability.The prediction ability of the Cottonspec software is particularly good for fine count yarns，i.e.Ne 50 and finer.Cottonspec is a user friendly software package that en－ables spinners i to select the most sui cotton that best meets their needs and ii as a quality control tool to benchmark their performance against“best commercial practice”.Cottonspec can also be used as a tool by cotton merchants to demonstrate and promote the value of a particular growth or variety，and by cotton fibre quality researchers to assess the technical merits of new cotton varieties.

Cottonspec is a computer software package that requires input parameters of cotton fibre properties，yarn parameters and limited processing information.Cottonspec predicts yarn evenness，thin and thick places，tenacity and elongation.The working principle of Cottonspec is illustrated schematically in Figure 1.

The Cottonspec software was initially launched at the 2011 Belt wide Cotton Conference［1］.The Cottonspec software has undergone validation trials in collaboration with a number of leading Chinese cotton spinning mills.The trial results have proven that Cottonspec is an extremely useful management tool which is able to give spinners immediate feedback on the yarn quality from the fibre used［2］.

The Cottonspec software has been upgraded for a number of times since its release in 2001.The latest upgrading of the software was in May 2013.Six major industrial spinning datasets，which consisted of 1241 yarn lots from 989 laydowns，were employed for development of the prediction algorithms and testing.The results have demonstrated that the upgraded Cottonspec algorithms have shown greater prediction power than the previous version.

This paper briefly describes some basic concepts of Cottonspec including the best commercial practice，the spinning prediction models and the details of the latest upgrading of the software and the validation trial results.

Figure 1 Schematic diagram of the working principle of Cottonspec

1 Best commercial practice

Yarn evenness is limited primarily by the average number of fibres in the yarn cross－section，because the likelihood of having a given number of fibres present is governed by the statistics of random processes［3］.This is acknowledged by the introduction of the term Index of Irregularity（I），which is the ratio of the measured evenness to the random limit［4］.All spinners run up against the same statistical limit which，for a given yarn count，is determined by fibre fineness（linear density）.How close the yarn evenness value gets to this limit is principally determined by the quality of drafting on the spinning frame.This can depend on the properties of input materials，for example，cotton with high short fibre content（SFC）will have poor Index of Irregularity of both roving and yarn［5－6］.The evenness of yarns approaches but never reaches the random limit，an Index of Irregularity of one.Yarn evenness，in turn，is a major factor in determining yarn strength as it influences the likelihood of thin or weak places［5－6］.Yarn strength（tenacity）also depends on twist and fibre properties，notably strength，length and fineness［5－14］.

There does not seem to be any great differences in the quality of yarn produced on different modern spinning frames.We have noted excellent quality，fine count yarn produced on spinning frames ranging in origin and age.We conclude therefore that the expected yarn quality is predominantly affected by the quality of the cotton used to spin the yarn.Herein is the concept of“best commercial practice”.It assumes the drafting system settings on the spinning frame are optimized and that the machines used from opening through to spinning are well maintained.For a given yarn count and twist，the yarn properties and spinning performance of a good mill are essentially determined by the fibre properties.

2 Industrial spinning trials

Since July 2008，CSIRO has conducted a series of large scale industrial spinning trials in collaboration with a number of leading Chinese cotton spinning mills.Cottonspec algorithms have been developed based on a comprehensive spinning database collected from partner mills over the last few years.The initial version of Cottonspec was developed in 2010 and launched at Beltwide Cotton Conferences［1］.Since then the algorithms have been upgraded for a number of times with the growing spinning database and improved modelling.The latest upgrading of Cottonspec was in May 2013.The upgrading was based on the latest spinning data collected from two partner mills for the period from June 2011 to December 2012.See Table 1 for details of the data provided by partner mills for the upgrading.

Table 1 Details of spinning data for the latest upgrading of Cottonspec in May 2013

There are six datasets，which in total consist of 1 241 yarn lots from 989 bale－laydowns.Among the six datasets，dataset 1 and 4 from mill 1 were chosen as the database for the upgrading of the algorithms for compact spun yarn and ring spun yarn，respectively.The remaining datasets were chosen as testing sets for validation purposes.Therefore，emphasis will be put on dataset 1 and 4.

The cottons used in the laydowns were of various origins，including US Pima，Egypt Pima，US SJV，Australian，Xinjing long staple cotton and Indian.For all datasets both cotton and yarn properties were tested at the mills′testing lab under standard conditions，i.e.20℃and humidity of RH 65%.Cotton fibre samples were tested on HVI and some other testing equipment.For this paper only the HVI data will be discussed since currently Cottonspec requires only five HVI data，i.e.Micronaire，Up－Half－Mean－Length（UHML），Short Fibre Contents （SFC），F(xiàn)ibre Tenacity （FT）and Fibre Elongation（FE）.Yarn samples were tested on Uster 3 for yarn evenness，thin and thick places，neps，and Chinese made tensile tester for yarn tenacity and elongation.

2.1 Cotton fibre properties

Details of the statistics of the cotton HVI properties for dataset 1 and 4 are given in Table 2 and 3，respectively.

Table 2 Statistics of HVI cotton fibre properties for 359 compact yarns of dataset 1

Comparison of histogram of Micronaire，UHML and fibre tenacity for datasets 1 and 4 are shown in Figure 2 3 and 4.

Table 3 Statistics of HVI cotton fibre properties for 114 ring spun yarns of dataset 4

Figure 2 Histogram of Mironaire

It is seen from Table 2 and Figure 2 that the Micronaire values for dataset 1 are within the optimum range of between 3.6 and 4.35 with a peak at around 3.8.For dataset 4 the Micronaire values are similar but with two peaks around 3.9 and 4.1.

Figure 3 Histogram of UHML

Figure 3（a）shows that for dataset 1 the mean fibre length has two peaks around 29 mm and 35 mm，which correspond to the upland cotton and Pima（and long staple Xinjiang cotton），respectively.Lots of laydowns were made from blends of Pima and upland cotton and this is clearly seen in the histogram.For dataset 4 the histogram shows two peaks around 30mm and 35mm，reflecting the fact that less blends of upland cotton with Pima（and Xinjing long staple cotton）were used in the laydowns.The histogram also shows that high percentage of long staple cotton was used for the ring spun yarns.

For fibre tenacity the situation is similar to fibre length（see Figure 4）.For dataset 1，there are two distinct peaks，around 31c N/tex and 39c N/tex，which correspond to the two components of upland cotton and long staple cotton，respectively.For dataset 4 there is a peak around 41c N/tex，which corresponds to Pima（and Xinjing long staple cotton）used in majority of the laydowns.

2.2 Spinning details

As shown in Table 1，359 compact spun yarns were from 250 laydowns and 114 ring spun yarns were spun from 114 laydowns.Yarn counts covered a range of from Ne 50 to Ne 100 for the 359 compact yarns and from Ne 44 to Ne 100 for the 114 ring spun yarns.The Histogram of yarn count for dataset 1 and 4 is shown in Figure 5.

The dataset 1 consists of 329 weaving yarns and 30 knitting yarns and the corresponding metric twist factor histogram is shown in Figure 6.The dataset 2 consists of 79 weaving yarns and 35 knitting yarns and the metric twist factor histogram is shown in Figure 7.

Figure 4 Histogram of fibre tenacity

Figure 5 Histogram of yarn count

Figure 6 Histogram of Metric twist factor for dataset 1

3 Results and discussions

3.1 Prediction models

A series of spinning prediction models for staple fibre spun yarns have been developed by Yang and his co－workers for a worsted spinning prediction package Yarnspec［5，15－17］.In principle，these models are applicable to cotton spinning.With a large spinning database available it becomes possible to apply the physical modelling technique to develop robust spinning prediction algorithms for fine count cotton yarns.To illustrate the principle of the work a brief summary of yarn tenacity prediction modelling is described in the following.

Figure 7 Histogram of Metric twist factor for dataset 4

Figure 8 Relationship between yarn tenacity and fibre tenacity

Theory and spinning trial results have shown that yarn tenacity is primarily determined by fibre tenacity［5－7］.Figure 8 shows the strong correlations of observed yarn tenacity with cotton fibre tenacity for the two datasets.

To achieve better results of prediction of yarn tenacity，a concept of Normalised Yarn Tenacity（NYT）is introduced：

NYT indicates the percentage of fibre tenacity that has been realised in the yarn tenacity.For staple fibre spun yarn NYT is a function of other fibre properties，e.g.fineness，elongation，length and short fibre contents etc.NYT is also affected by yarn twist level.

To determine the effect of yarn twist on yarn tenacity，a yarn twist curve trial was conducted in October 2010 with collaborations with a partner mill.Three cottons were used in the trial including SJV，US Pima and Brazilian cotton.For each cotton three yarn counts were used：Ne 23，Ne 32 and Ne 40 for Brazilian cotton，Ne 40，Ne 50 and Ne 60 for SJV cotton，Ne 50，Ne 60 and Ne 70 for US Pima cotton.For each yarn count seven twist levels were used ranging from Metric Twist Factor（MTF）95 to 130.In total 63 yarns were spun and there were nine experimental twist curves showing the dependence of yarn tenacity versus twist level.Due to the limited number of twist levels used in the trial not all curves showed tenacity maxima.Strictly speaking，the shape of the twist curve was cotton fibre variety（mainly fibre length）and yarn counts depended.For simplicity，an average of Ne 50 and Ne 60 yarn twist curves was used to develop a theoretical yarn twist model by fitting the experimental data.

The theoretical yarn twist model is shown in Figure 9，which shows that yarn tenacity increases with increasing yarn twist level and reaches a maximum value at approximately 120 Metric Twit level.Note that the vertical axis is the normalized yarn tenacity which is the ratio of yarn tenacity divided by the average yarn tenacity at various twist levels.The normalised yarn tenacity at each twist level is called yarn tenacity twist correction factor.

To exclude the effect of yarn twist on yarn tenacity，a Twist Corrected Normalised Yarn Tenacity（TCNYT）is introduced and defined as：

Theoretically，TCNYT is independent of fibre tenacity and yarn twist.However，experimental results show that fibre tenacity has a“secondary effect”on observed yarn tenacity.In other words，TCNYT is still fibre tenacity dependent.This is because in processing fibre length changes because of fibre breakage.The degree of fibre breakage depends on fibre break to work，which is the product of fibre tenacity and elongation.

Utilising the large size spinning database，a spinning prediction model for TCNYT has been developed，which is a function of predicted yarn evenness，Micronaire，short fibre contents，fibre length，and fibre tenacity.Yarn evenness prediction model has also been developed based on theoretical yarn evenness prediction model and fitted with the data from the large size spinning database.Finally，predicted yarn tenacity can be expressed by the following equation：

Figure 9 Cotton spinning yarn twist model

To facilitate wide applications of Cottonspec by the industry the upgraded Cottonspec prediction algorithms require only five HVI testing data：tenacity，elongation，UHML，SFC% （＜16mm）and Micronaire.However，there is a room for improved prediction accuracy if some other fibre property data becomes available，e.g.fibre fineness and neps count.

3.2 Mill correction factor

The quality of a spun yarn is predominantly affected by the quality of the cotton used to spin the yarn.Having said that，other factors，e.g.the quality of textile machinery and maintenance，machine settings，and operator skills etc do play a role in determining yarn quality，to some extent.

To make Cottonspec a useful quality control tool for a range of spinning mills，it is necessary to introduce Mill Correction Factor（MCF）.For example，predicted yarn tenacity for Ne 60 yarn may be expressed as：

PYT（Ne 60）＝TCNYT（predicted CV%，Mic，SFC，UHML，F(xiàn)T）×FT×TCF×MCF（YT Ne 60）.

For standard Cottonspec the default value of all MCF is one.For a particular mill the MCF value may be adjusted after a certain period of time when enough processing data is accumulated.For a particular yarn property of a given yarn count，the MCF is one minus the average variations between predicted and measured yarn property：

where，n＝the number of yarn lots.

In majority cases MCF is expected to be yarn count dependent.In Cottonspec software there is a panel of MCF with two options：Set default or update MCF.The user is expected to record measured yarn properties against the predicted.When enough processing data is accumulated by choosing the option update MCF the software will automatically calculate MCF and replace the existing MCF with the new one.Alternatively，the user can always choose the set default option and by doing so the software predicts the best commercial practice.

With the prediction models described above prediction algorithms were upgraded using the dataset 1 and dataset 4 for compact yarn and ring spun yarn，respectively.Applying the prediction algorithms the calculated yarn evenness and tenacity vs.the measured are shown in Figure 10 for the 359 compact spun yarns and Figure 11 for the 114 ring spun yarns，respectively.It is seen that the calculated yarn evenness and tenacity is highly correlated to the measured with the square of correlation coefficient being 0.83 for yarn evenness and 0.92 for yarn tenacity for the 359 compact spun yarns，and 0.91 for yarn evenness and 0.86 for yarn tenacity for the 114 ring spun yarns.Standard errors and relative standard error for yarn evenness and tenacity prediction are given in Table 4.

Table 4 Square of correlation coefficients and prediction errors with and without MCFs

Figure 10 Calculated vs.measured for 359 compact spun yarns

Figure 11 Calculated vs.measured for 114 ring spun yarns

4 Validation of Cottonspec algorithms

4.1 Mill 1

Dataset 2 （360 compact yarns from 336 laydowns）and dataset 5 （87 ring spun yarns from 87 laydowns）of mill 1 are used as testing sets to Validate Cottonspec prediction algorithms.

Applying standard Cottonspec（MCF＝1）the predicted yarn evenness and tenacity against the measured are shown in Figure 12 for dataset 2，and Figure 13 for dataset 5.

Figure 12 Predicted vs.measured for 360 compact yarns of dataset 2

Figure 13 Predicted vs.measured for 87 ring spun yarns of dataset 5

As may be seen from Figure 12 that，for dataset 2 the predicted yarn evenness and tenacity is highly correlated to the measured with the square of correlation coefficient being 0.79 for both yarn evenness yarn tenacity.The prediction errors are also relatively small，being 0.39%for yarn evenness and 1.03 c N/tex for yarn tenacity.The relative prediction errors are 3.05%for yarn evenness and 4.47%for yarn tenacity，which are reasonably good given the fact that these two datasets are independent on the database used for upgrading of the algorithms and the data are industrial spinning trial results rather than laboratory experimental data.

For dataset 5，F(xiàn)igure 13 confirms that the prediction algorithms work well for the 87 ring spun yarns with the square of correlation coefficient being 0.84 for yarn evenness and 0.72 for yarn tenacity.The prediction errors are quite small being 0.13%for yarn evenness and 0.58c N/tex for yarn tenacity，which are even better than for dataset 2.The relative prediction error is excellent：1%for yarn evenness and 3%for yarn tenacity.

Note that the two datasets used here are independent on the database used for the upgrading of the algorithms.Furthermore，the data used are industrial spinning trial data rather than laboratory experimental type data.These results demonstrate that the prediction algorithms work well for a good modern mill，like mill 1.

Although the prediction accuracy is quite good it can still be improved further by using MCF.MCF is automatically worked out by choosing the update MCF option.With MCF the predicted yarn evenness and tenacity vs.the measured are plotted in Figure 14 for dataset 3，and Figure 15 for dataset 5.The correlation coefficient between the measured and the predicted yarn evenness and tenacity as well as the prediction error are summarised in Table 4.

Figure 14 With MCF predicted vs.measured for 360 compact yarns

Figure 15 With MCF predicted vs.measured for 87 ring spun yarns

Figure 14 and 15 show that the prediction accuracy has been greatly improved by using MCF：for dataset 2 the square of correlation coefficient is improved from 0.79 to 0.83 for yarn evenness and from 0.79 to 0.88 for yarn tenacity，while for dataset 5 the square of correlation coefficient is improved from 0.84 to 0.85 for yarn evenness and from 0.72 to 0.86 for yarn tenacity.As may be seen that the standard error is also greatly reduced：for dataset 2 the standard error is reduced from 0.39%to 0.35%for yarn evenness and from 1.03 c N/tex to 0.79 c N/tex for yarn tenacity，while for dataset 5 the standard error is reduced from 0.13%to 0.08%for yarn evenness and from 0.58 cN/tex to 0.49 c N/tex for yarn tenacity.In terms of relative standard error，the improvement for dataset 2 is from 3.05%to 2.77%for yarn evenness and from 4.47%to 3.44%for yarn tenacity，while for dataset 5 the improvement is from 0.99%to 0.65%for yarn evenness and from 3.08%to 2.56%for yarn tenacity.These results clearly demonstrate that MCF is a powerful function that can significantly improve the prediction accuracy.

4.2 Mill 2

Applying standard Cottonspec（MCF＝1）to dataset 3（184 compact yarns from 120 laydowns）and dataset 6（137 ring spun yarns from 96 laydowns）the predicted yarn evenness and tenacity against the measured are shown in Figure 16 and 17，for dataset 3 and dataset 6 respectively.The correlation coefficient between the measured and the predicted yarn evenness and tenacity as well as the prediction errors are summarized in Table 4.

Figure 16 and 17 clearly show that yarn quality for mill 2 is significantly poorer than for mill 1：

Figure 16 Predicted vs.measured for 184 compact yarns of dataset 3

Figure 17 Predicted vs.measured for 137 ring spun yarns of dataset 6

measured yarn evenness is greater than that of predicted（on average about 3%for compact yarn and 5%for ring spun yarn），while measured yarn tenacity is smaller than that of predicted（on average about 7%for compact yarn and 9%for ring spun yarn）.

The large prediction errors and the low correlation coefficient between the predicted and the measured yarn quality（Table 4）are largely due to the poor status of quality control in the mill.Actually，the processing system at mill 2 is similar to that of mill 1 but the quality control status is much lag behind.Cottonspec can be used as a quality control tool to benchmark a mill′s performance against“best commercial practice”.The poor spinning performance of mill 2 is clearly revealed as shown in Figure 16 and 17.With the quality gap revealed the most important task for mill 2 is to improve its yarn quality by optimising spinning processing.

Quality improvement takes time for a spinning mill.During the course of quality improvement the prediction accuracy for mill 2 can be improved by choosing the update MCF option.With MCF the predicted yarn evenness and tenacity vs.the measured are plotted in Figure 18 and 19.The correlation coefficient between the measured and the predicted yarn evenness and tenacity as well as the prediction errors for dataset 3 and 6 are summarized in Table 4.

Figure 18 With MCF predicted vs.measured for 184 compact yarns of dataset 3

Figure 19 With MCF predicted vs.measured for 137 ring spun yarns of dataset 6

Figures 18 and 19 show that the prediction accuracy has been significantly improved by using MCF：for dataset 3 the square of correlation coefficient is improved from 0.60 to 0.69 for yarn evenness and from 0.51 to 0.69 for yarn tenacity，while for dataset 6 the square of correlation coefficient is improved from 0.71 to 0.73 for yarn evenness and from 0.60 to 0.61 for yarn tenacity.As may be seen from Table 3 the standard error is also greatly reduced：for dataset 3 it is reduced from 1.18%to 0.62%for yarn evenness，and from 1.03 c N/tex to 0.64 c N/tex for yarn tenacity，while for dataset 6 the standard error is reduced from 1.26%to 0.36%for yarn evenness and from 1.03 c N/tex to 0.62 c N/tex for yarn tenacity.In terms of relative standard error，the improvement for dataset 3 is from 8.71%to 4.60%for yarn evenness and from 5.32%to 3.30%for yarn tenacity，while for dataset 6 the improvement is from 9.20%to 2.60%for yarn evenness and from 5.92%to 3.56%for yarn tenacity.Again，the results clearly demonstrate that MCF is a powerful function that can significantly improve the prediction accuracy.

5 Summary

Cottonspec，a commercial cotton fibre and yarn quality management software package，has been developed.The package utilizes very large industrial spinning datasets and comprehensive theoretical modelling.Cottonspec is underpinned by a basic concept of best commercial practice.Validation trial results have demonstrated that Cottonspec works well for a good modern spinning mill with predicted yarn evenness and tenacity highly correlated to the measured yarn quality.Prediction accuracy is further improved by using MCF.

Acknowledgements：The authors wish to acknowledge Chinese partner mills in conducting the spinning trials and for providing a large amount of spinning data.The work was supported by the Australian Government，through the Department of Agriculture，F(xiàn)isheries and Forestry（DAFF），Cotton Research and Development Corporation（CRDC），Cotton Catchment Communities Cotton Research Council（CCC CRC）and CSIRO.

［1］ YANG S，GORDON S，WU L.Cottonspec——A cotton fibre and yarn quality management tool［C］//Proc Beltwide Cotton Conferences，National Cotton Council，Atlanta GA，USA，2011：1486－1494.

［2］ YANG S，GORDON S，WU L.Preliminary results of cottonspec validation trials［C］//Proc 2012 Beltwide Cotton Conferences，National Cotton Council，Orlando，F(xiàn)lorida，USA，2012：1428－1435.

［3］ MARTINDALE J G.New method of measuring the irregularity of yarns with some observations on the origin of irregularities in worsted slivers and yarns［J］.J Text Inst，1945，36：35－47.

［4］ SIMPSON J.Discussion of irregularity index as a measure of cotton yarn evenness［J］.Textile Research J，1965，35：854－855.

［5］ YANG S，LAMB P R.The art of spinning prediction－modelling yarn performance in worsted spinning［C］//Proc 2nd China International Wool Conference Xi′an 1998 348－366.

［6］ SPENCER－SMITH J L.The estimation of fibre quality［J］.J Text Inst，1947，38：257－272.

［7］ HEARLE J W S.Theory of mechanics of staple fibre yarns，structural mechanics of fibres：Yarns and Fabrics［M］.London：John Wiley ＆ Sons，1969.

［8］ FRYDRYCH I.A new approach for predicting strength properties of yarn［J］.Textile Research J，1992，62：340－348.

［9］ BOGDAN J F.The predictions of cotton yarn strength［J］.Textile Research J，1967，37：536－537.

［10］ FIORI L A，SANDS J E，LITTLE H W，et al.Effect of cotton fibre bundle break elongation and other fibre properties on the properties of a coarse and a medium singles yarn［J］.Textile Research J，1956，26：553－564.

［11］ PATEL G S，PATIL N B.Studies on some physical parameters of cotton fibres and their influence on breaking strength［J］.Textile Research J，1975，45：168－172.

［12］ VIRGIN W P，WAKEHAM H.Cotton quality fibre properties，PartⅣ：The relation between single fibre properties and the behaviour of bundles，slivers，and yarns［J］.Textile Research J，1956，26：177－191.

［13］ ZUREK W，F(xiàn)RYDRYCH I，ZAKRZEWSKI S.A method of predicting the strength and breaking strain of cotton yarn［J］.Textile Research J，1987，57：439－444.

［14］ FRYDRYCH I.Strength properties of cotton blend yarns［J］.Textile Research J，1994，64：198－208.

［15］ YANG S，HUMPHRIES H.Application of Sirolan yarnspec in the Chinese wool industry［C］//70th IWTO Shanghai Meeting，Tech Ctee Report CTF No.3，Shanghai，2001.

［16］ YANG S，HUMPHRIES H.Mill－specific yarnspec and its applications［C］//Proc 3rd China International Wool Conference，Xi′an，2002.

［17］ YANG S，WANG K，HUMPHRIES W，et al.An innovative approach to worsted spinning prediction－modelling yarn performance using artificial neural networks with built－in know－h(huán)ow［C］//Proc 11th International Wool Research Conference，Leeds，UK，2005.