FREE WOOL-FIBRE SHAPES

Abstract

Keywords:

Introduction and Motivation

Fibre shape is an important factor in many textile processing operations (such as combing yield and throughput), as well as end-user applications. Fibre shape is also an important ingredient in realistic theoretical models of such operations and applications.

A visual estimate of crimp [1] is the most commonly used measure of shape, but it is generally accepted that it is not a complete descriptor of processability or end-product performance. One potential reason for this is that crimp and other automated measurements of fibre curvature [4], such as OFDA [2] and Sirolan-Laserscan [3] measurements, do not capture the three-dimensional nature of fibre shape.

To our knowledge, the explicit mathematical nature of fibre shape is poorly understood. Since fibres are well approximated by 1D space curves, their shape is uniquely parameterised in terms of the curvature (in-plane bending) and torsion (out-of-plane bending) along their lengths, and it is not known what information is not captured by measuring curvature alone. The aim of the experimental study presented in this paper is to understand more fully the nature of fibre shape, in order to suggest practical ways to obtain any ’missing' information, either by alternative measurement techniques or alternative uses of information obtained from current techniques, and in the process build a theory of fibre shape that would be useful for future theoretical modelling.

The experiment involved 18 degreased Perendale wool fibres, nine taken from the fleece of one sheep, and nine from another sheep. The 18 fibres were not chosen at random, since it was felt that experimental errors might swamp the results for very straight fibres, and the image processing was too difficult for highly convoluted fibres. The experiment is only preliminary, due to the small sample size, but interesting results have nevertheless been obtained.

The bending rigidity of a wool fibre is so small that any physical contact, even airflow, can change its shape easily. Photography is therefore an ideal way of determining a fibre's shape.

The fibres were first steamed and slowly dried so that they were as free as possible from internal stresses. Each fibre was then hung from one of its ends so that the only external load was gravity, enclosed in a glass prism to minimise its movement, and a number of photographs were taken to

http://www.autexrj.org/No4-2003/0070.pdf

capture a series of projected images. Enough photographs were taken so that the final 3D space curve was over-determined. Diffuse backlighting and small lens apertures were used to minimise shadows and to maximise the depth of field.

Each print was then scanned to computer using a high resolution CCD scanner. Noise was removed and the fibre images thinned to 1-pixel-wide lines using standard computer packages. Two typical projections are shown in Figure 1. The 3D space curve was then constructed using the set of all projections of a particular fibre. This step was more difficult for convoluted fibres than straight ones.

Figure 1. Two typical projections. The fibre on the left, which has length 71 mm, has a smaller mean curvature than the fibre on the right with length 83 mm

By calculating and measuring the loss of resolution through the entire procedure, and by comparing the final results constructed from disjoint subsets of the set of initial photographs of one fibre, the space curves thus constructed were accurate to within roughly –30 mm (about one fibre diameter).

To determine the curvature and torsion, four adjacent points (a,b,c,d ) are needed. The curvature at

point b (k_b ) and torsion mid-way between b and c (t _bc )are given by

k_b =

where is the vector pointing from a to b , etc. These formulae are the maximally symmetric finite

difference forms which reduce to the Frenet formulae at the infinitesimal limit.

This finite-difference approximation was used to keep close to the current technologies (OFDA and Laserscan), as well as to reduce short-distance experimental noise. Except where noted, all results quoted below are measured at a ‘scale' of 1~mm (which is OFDA's ‘scale'), meaning that the curvature and torsion are calculated at every experimentally-obtained data point – and there are typically around 10,000 of these – by applying Equation (1) with an arc length of 1~mm between the points. The curvature and torsion cannot be defined for points close to the ends of the fibre. A representative example of the curvature and torsion as a function of arc length is shown in Figure 2.

Figure 3 demonstrates that curvature and torsion fluctuate considerably about their mean values for each fibre. This is generic behaviour for wool (since no fibres are substantially helical), which is the constant-k , constant-t shape.

Figure 2. The curvature (thick line) and torsion, both measured in mm ^-1 , as a function of arc length s ,

-1

measured in mm, for the more convoluted fibre shown in Figure 1. The average curvature is 0.64mm ,

-1 -1 -1

average torsion 0.09mm , mean magnitude of torsion 0.69mm , RMS curvature 0.70mm and RMS

-1

torsion 0.77mm . Roughly 14,000 data points were obtained for this fibre

Relationship between various moments of curvature and torsion

Comparing results from all the fibres measured, it is found that the mean curvature k , and the RMS _k 22

enjoy a linear relationship: k » 1.1k . This is shown in Figure 3, in which it is also clear that the same relation holds for fibres from both sheep. A similar linear relation holds for scales of measurement other than 1 mm.

-1 -1

Figure 3. The RMS curvature (

k ² in mm ) seems to be fairly linear in the mean curvature (k in mm ). The results from the fibres from the two different sheep are distinguished by stars and diamonds, but follow the same linear trend

The mean torsion was always very small (less than 20% of the mean magnitude of torsion) and of

seemingly random sign. The RMS torsion (

) and the mean magnitude of torsion ( t ) could also

be related: t ² » 1.2t

. This is depicted in Figure 4, and a similar result holds for scales of measurement other than 1 mm. However, this relation is much less statistically significant than the

analogous result for curvature, since all quantities related to torsion vary very little over the entire sample.

Figure 4. The RMS torsion (

in mm ) seems to be fairly linear in the mean magnitude of torsion (t in

mm ^-1 ). The results from fibres from the two different sheep are distinguished by stars and diamonds

Fourier analysis

A discrete Fourier transform of the values of curvature at equally spaced points along the fibre k ,

n

n = 0,K, N -1 ( N is of the order of 10,000) and torsion t , is used to analyse the results further:

k _n = ¹ _�k e and t = ¹ _�t e .

A typical power spectrum is shown in Figure 5. Torsion is similar. It is the small frequency behaviour that is of interest here (with wavelengths of more than about 0.5 mm), since most of the power is contained in these modes. The high-frequency modes tend to be overcome by experimental noise, but are damped by the process of measuring curvature and torsion at a non-zero scale.

as a function of j for j > 0 is roughly exponential with a power-law tail. For this particular

~ -1

example, k = 40.7mm , but this point is not plotted

No strong peaks corresponding to characteristic wavelengths were discernible. Instead, the low-frequency behaviour can be adequately modelled by

= a_k exp(-b_k j / L) for 0 < j < N /2 ,

k₀ = Nk and

t ₀ = Nt and t

= a_t exp(-b_t j / L) for 0 < j < N /2 . (2)

Here, L is the length of the fibre (so that j /L is a ’frequency') and the dimensionful constants a_k ,

b_k , a and b are all positive, and may be different for each fibre.

By fitting the power spectra with the above functions for each fibre, it is found that a_k is directly

proportional to the product of the standard deviation s (k ) = with the

a_k » 1.4s (k ) N / L , (3)

as shown in Figure 6 (all dimensionful quantities are measured in mm). This is not completely unexpected, since a_k parameterises the fluctuations of k about its mean. However, it is conceivable

that the dimensionful numerical factor 1.4 mm is an invariant of sheep breed.

The phase of each frequency mode was found to be entirely random for all scales considered. This is important for the generating algorithm below. However, in less than 50% of the fibres was there either a positive correlation (where the phases were similar) or a negative correlation between the phase of

We have been unable to find any strong correlation between the curvature and the torsion of the fibre. Observing such a relation, if it exists, is frustrated by the fact that all quantities related to torsion were

-1

quite uniform over the entire sample of fibres studied. a varied between 8 and 18 mm and b

tt

between 1.1 and 2.4 mm, and there was no obvious correlation between them.

Generating realistic fibre shapes on a computer

The empirical relations found above may be taken as a basis for a model of fibre shape. Therefore, given the unknowns in the power spectra of Equation 2, random phases may be chosen for each

mode k_j and t _j , an inverse FFT may be used to find curvature and torsion, and the space curve generated.

The inputs to the procedure are k , k ² , and t , if required. The latter may be taken to be small

(yielding, say, Nt = t = ae ). The constant is found directly from Equation 3. For large

N , Parseval's theorem yields 2b / L » log(1+ 2·1.4 / L) . Finally, a_t and b may be chosen

arbitrarily within reasonable limits mentioned above.

If this procedure yields points of negative curvature, another random data set may be generated, or it may be noted that a space curve of negative curvature -k

t is physically indistinguishable from a curve with positive curvature + k and torsion -t . The effect

of this sign change on the power spectra is very small and certainly within the accuracy of the whole procedure.

An example of the curvature and torsion of a computer-generated fibre is shown in Figure . Here,

-1 2 -1

k= 0.64mm and k= 0.70mm were chosen with N = 14000 and the fibre length of

L = 83mm . The above formulae give k = 41mm , = 5.0mm and b = 1.8mm . The torsional

parameters t = 0 , = 11mm and b = 2.3mm were chosen. Therefore, this fibre should be

similar to that depicted in Figure 2.

Figure 7. k (bold) and t (both measured in mm ) as a function of arc length s (in mm) generated by the procedure described in the text

Summary and Discussion

This experimental work, which measured the space curve of 18 free fibres, has shown that the power spectra of curvature and torsion can be well approximated by exponentials. Moreover, curvature (including all its statistical fluctuations along the fibre's length) seems to be completely parameterised in terms of its mean and RMS values. The torsional properties of the fibres were found to be comparatively constant. The phases of the spectra were random and largely uncorrelated between the same modes of curvature and torsion.

These results are important, for they suggest that a parameterisation of fibre shape in terms of only a handful of parameters probably exists and, because the low frequency modes completely dominate, only comparatively low-resolution images are needed to capture this information. Measuring the small number of important parameters for a batch of fibres should help infer the processability and end-product performance for that batch.

Unfortunately, no strong conclusions could be made about torsion because the torsional parameters tended to be constant over the sample of 18 fibres. This uniformity may be an intrinsic feature of wool fibres, or it may have been caused by the non-random sampling (which did not appear to affect the curvature in the same way).

The featureless exponential power spectrum is surprising because a ’wavy’ periodic pattern is often evident in the staple, even for Perendale fibres. However, the paths of individual fibres are fairly irregular when tracked through the staple (as may be inferred from Figure 1), which suggests that the expected peak in the power spectra will be rather weak. A superposition of the power spectra of the 18 fibres showed no obvious peak at non-zero frequency, which suggests that the position of this peak probably depends on the staple.

Torsion sometimes appeared slightly step-wise along the fibre's length – remaining roughly constant in magnitude but changing sign occasionally – especially at large scales. This is not captured by the

exponential power spectrum (or t � t ) but again, because of lack of data, no other tell-tale peaks

were observed. This could be the cause of the wavy pattern in the staple.

Finally, it would be an interesting exercise to explore the relationship between the bulk of a single fibre and parameters such as k , s (k ) , a_k , etc. Bulk could be defined as the ratio of the volume of the smallest parallelepiped that will enclose the fibre, to its length. There is a relation between staple bulk

(and fibre diameter) and curvature [5]: it would be interesting if something like Equation 3 emerged as a condition of maximum fibre bulk.

Acknowledgements

We thank Peter Durrant of WRONZ for helping with the photographic procedures and for providing the necessary equipment. CM was supported by the New Zealand Wool Board, project 95WR02. BT was funded by a New Zealand Wool Industry Charitable Trust PhD scholarship. AW was funded by FRST postdoctoral fellowship WROX0010.

1. Alexander, E., Menachem, L., Muhsam, H. V. and Shiloh, M. Definition and Measurement of Crimp of Textile Fibres, Textile Res. J. 26, 606-617 (1956).
2. Brims, M. A., New OFDA Developments and the Use of OFDA as a Projection Microscope, IWTO Technical Committee Meeting, Istanbul, May 1993, Report 22.
3. Dabbs, T. P., van Schie, H. and Glass, M. The Effect of Fibre Curvature on Laserscan Diameter Measurement, IWTO Technical Committee Meeting, Nice, December 1994, Report 2.
4. Edmonds, A. R., Measurement of Fibre Curvature: A Review of Work to Date, Wool Tech. Sheep Breed. 45, 227—234 (1997).
5. See references in [4]. Also, Sumner, R. M. W., and Upsdell, M. P., Factors Associated with the Prediction of Core Bulk from Fibre Diameter and Fibre Curvature of Individual Fleeces, Wool Tech. Sheep Breed. 49, 29-41 (2001).