Textile Research Journal: Problems and Possibilities in Sliver Monitoring
Textile Research Journal: Problems and Possibilities in Sliver Monitoring
Textile Research Journal: Problems and Possibilities in Sliver Monitoring
com/
Published by:
http://www.sagepublications.com
Additional services and information for Textile Research Journal can be found at: Email Alerts: http://trj.sagepub.com/cgi/alerts Subscriptions: http://trj.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav Citations: http://trj.sagepub.com/content/57/3/171.refs.html
171
ponents of Polyester-Cellulosic Fibre Blends, J. Soc. Dyers Colour. 84, 434-435 (1968). 7. Jain, A. K., and Sao, K. P., Analysis of Jute/Rayon
12. Mehta, R. D., A Rapid Method For Determining Cotton/ Polyester Blend Levels, Textile Res. J. 55, 259-261
R. J., and Tripp, V. W., An X-Ray Diffraction Method for the Analysis of Cotton/Polyester Blends, Textile Res. J. 47, 386-388 (1977). 14. Parthasarathy, M. S., Srinathan, B., Sreenivasan, S., Chidambareswaran, P. K., and Sundaram, V., Cotton13.
Blends, Ind Textile J. 93, 81-83 (1982). 8. King, W. D., and Stannonis, D. J., Analysis Methods for Cotton/Polyester Blends, Am. Dyest. Rep. 65, 36-37, 39, 60 (1976). 9. Kumar, Raj, Sharma T. S., and Srivastava, H. C., Quantitative Analysis of Cotton-Polynosic Blends, Textile Res. J. 49, (1979). 673-675 10. Kumar, Raj, and Srivastava, H. C., Analysis of Fiber Blends, Part II: Determination of Blend Composition by Moisture Regain, Textile Res. J. 50, 359-362 (1980). 11. Mayboo, C., Jagasia, S. R., and Radhakrishnan, M., Evaluation of A Micro-Technique for the Quantitative Separation of Components in Synthetic/Natural Fiber Blends with Special Reference to Polyester/Cotton Blends, in "Proc. 15th Technol. Conf. ATIRA, BTRA, SITRA," 1974, pp. 26.1-26.12.
Blending—Changes in Blend Composition During Processing, Textile Res. J. 51, 119-121 (1981). 15. Ruckert, I., Determining Polyester in Blends Using Trichloroacetic Acid, Textile Chem. Color. 153-155 10, (1978). 16. Soignet, D. M., Berni, R. J., and Janssen, H. J., Characterization of Cotton-Polyester Blends by Electron Spectroscopy (ESCA), Textile Res. J. (1976). 46, 763-766
Manuscript received April 7, 1986; aoc~pred Apvi! 11. 1986.
Jute
Monitoring
Sliver monitoring permits the reporting of errors at the time the material goes out of tolerance during processing. This can greatly reduce the waste of material, time, and money, and it can also improve the quality of the product. Adequate monitoring is only possible if the requisite parameters are measured, but most measuring systems react to changes in several parameters and create ambiguity. Currently, none of the parameters such as fiber length, crimp, and silver bulk are measured continuously, even though better system control would be likely if variations in these data could be incorporated into the control devices. This paper discusses these problems in greater detail and suggests a number of possibilities of extending the range of data available for use in monitoring and control systems.
Monitoring equipment draws attention to out-oftolerance material being produced at the time the errors
are
created by the machine in question. It also provides managerial tool to keep track of the performance of machines, materials, and operators [2]. The parameters
to be monitored depend on the interactions between the various machines in the operational flow line and the textile material flowing through. The question of which parameters need to be monitored is discussed in this paper.
172
Sampling methods of quality control, as in common today, involve delays in error detection. This leads to undesirable quantities of substandard material in the production stream before the errors are corrected. These errors can cause problems in later processes and add to the production costs; they also cause a deterioration in quality, which has an adverse effect in the market place. As more highly productive and automated equipment is installed, so the potential for the production of greater quantities of faulty material increases. A quicker response time is needed, and this suggests a need for on-line monitoring. The amount of justifiable investment in a system applied to a single machine is determined by the rate at which textile material passes through that machine. For example, the productivity of a ring spindle is very low, and the justifiable expenditure for monitoring is only of the order of $10 per spindle. On the other hand, a drawframe or card has a high throughput, and expenditures measured in thousands of dollars are justified on purely economic grounds. On this basis it seems to be reasonable to concentrate on monitoring the production of sliver. To carry out its function properly, a monitoring system has to be able to measure appropriate parameters and to identify causes of error as they emerge. In practice, the only on-line measurements that are made today in most staple production systems purport to relate ~ to linear density of the sliver. Sometimes the signals from the transducers are used for monitoring to detect out-of tolerance conditions (i.e., reporting only), and sometimes for autoleveling (i.e., making automatic corrections). In either case, it is important to know to what extent the results truly represent the changing values of linear density. It is also important to know if control of linear density is su~cient to give the desired
use
Unfortunately, inspection of the spectropractice occurs all too infrequently, and often the faults are very strongly developed before action is taken. Weakly developed spikes are easily confused with transient quasi-harmonic disturbances or drafting waves and are often ignored. A useful way of discriminating between random and
machine.
grams in normal
harmonic components is to make arrays of spectrograms and look for channels of identical error wavelength. Figure 1 shows several series of spectrograms taken at various stages in manufacture in a spinning mill. The particular machines were chosen at random, and extensive samples of sliver from each machine were tested on a standard spectrograph. Spectrograms from ten consecutive portions of each sample were recorded, and the results were traced on to another paper to give an array such as shown in Figure 1. Note that very
large spikes
at constant error
wavelengths
are
highly
visible in some of the arrays. When individual spectrograms are used, small spikes cannot be resolved with certainty, and it is common practice to allow the spikes to become large before maintenance is done. Meanwhile many tons of faulty sliver have been produced.
MECHANICAL ERRORS
A mechanical fault in
periodic
and the
source
error that
error wavelength can be used to diagnose the of error. Strongly developed &dquo;spikes&dquo; arising from the mechanical error are usually quite clear, and it is wavelength rather than amplitude that is important. In this case it does not matter much whether or not the transducer truly records linear density. Regular inspection of spectrograms can permit a reduction in the quantity of faulty material produced by the subject
173
Machine faults of importance may be divided into two classes. The first comprises large faults that appear suddenly, and these can be dealt with adequately by conventional systems. The second comprises faults that
emerge
gradually. For example, roll coverings harden irregularly, and mechanical errors in the textile begin to show before the roll becomes obviously damaged. If these events could be detected at an early stage, it
would be much easier to schedule maintenance at a convenient time. To get early warning, it is necessary to separate the weak harmonic patterns from the random signals that are found with textile processing. One possibility is to use arrays of spectrograms as just described. Continuous autocorrelation of the data stream from the transducer is another possibility. Autocorrelation over a sliding interval tends to nullify the random components and leave the harmonics more visible, but autocorrelograms are difficult to interpret.
DRAFTING WAVES
FIGURE 2.
Drafting waves in a properly designed and set drafting device are caused by variations in fiber parameters. The wavelength distributions of these disturbances usually have local maxima, and the local central tendencies are related to the length characteristics of the fiber and the draft involved. Classical theory [ 1, 3, 4, 5] held that the fiber length was one of the most important determinants of the problem, and many writings have assumed that the length distribution of the fiber involved is constant. In fact it is quite variable [8, 9] in many respects. There is a sort of variable blending between fibers of different length. Variations in blend lead to later variations in linear density [8] because of
the shear effects in drafting. Hence variations in fiber length caused by carding or earlier machines introduce latent errors that develop into real errors during later
rayed spectrograms.
Unfortunately, it is not just a question of variations in fiber length, but rather in fiber extent, which is the distance along the line of flow of a convoluted fiber or fiber clump. With well separated fibers that have poor orientation, many are folded and have fiber extents much less than their actual lengths. When there are
fiber clumps, it is possible for the fiber extent to be more than a single fiber length. Sequential measurements of fiber length from breaker-drawn sliver made on a digital fibrograph did not correlate at all well with the corresponding inchto-inch variations in linear density. The results from the digital fibrograph are expressed in terms of span length. If the x% span length is y inches, then this means that x% of the fiber population is over y inches in length. Sliver contains hooked fibers, and the results in the leading and trailing directions always differ because of the way the hooks do or do not comb out in the two directions. The variations in span length were compared to the local linear densities. At the 2.5% span length (the longest fibers), the correlation coefficients was -0.27 in the trailing direction and -0.07 in the leading direction. At 50% span length (the median fiber length), the correlation coefficients were +0.06 in the trailing direction and zero in the leading direction. At best, fiber length did not account for more than 10% of the variations in sliver linear density. In a drawframe, the fibers are not combed out as in a fibrograph test, but are considerably hooked or convoluted. There are considerable variations in fiber extent that do not correlate with the actual fiber lengths. Figure 3 shows a series of determinations of fiber extent in fibers in a card sliver as it was being drafted by two sets of rollers (the way in which the fiber extent was measured will be described later). One can see that the deviations were large and would have caused differ ences in the behavior of the following drafting systems. In other words, the deviations due to fiber folding, clumping, or other deviations from the paravel state were latent errors awaiting development at a later uar.
174
This underlines the importance of being able to measure fiber extent continuously during production.
LATENT ERRORS
To show the effects of latent errors, some of the card sliver mentioned earlier was drawn and then converted to roving. Figure 4 shows the short-term root mean square (RMS) error in the roving measured by a capacitive transducer and signal processor. The drift and sinuous variations in the diagram can be related to the wavelengths originally generated in breaker and finhher drawing. The point is, however, that the variations were of such wavelengths that they could not have been generated by faults in the roving frame. The deviations shown were generated in the roving frame as differing populations of fiber passed through.
In our previous work (Lord and Johnson [7]), we have suggested that concentrations of bad fiber break up into subgroups during drafting to create bursts of drafting waves. Logic suggests that latent errors should also affect spinning, but there is an ever-changing fiber population [6] that modifies the behavior and introduces at least one element of confusion. It was necessary to further test the hypotheses. Sliver was marked at the thick spots and was then converted into roving and ring-spun yam. There was remarkably poor correlation between the thick and thin spots in the yam and the linear density of the corresponding portions of roving and drawn sliver. Many of the slub-like thick spots in the yam were associated with particles of trash or clumps of fibers. These inclusions and agglomerations were usually small and were buried in the roving or sliver; they represented only a small percentage of the local mass/unit length. These results reduce the possibility that potential end breaks in spinning can be detected by just measuring the linear density of the sliver. Nevertheless, better control of the variations in linear density of the silver leads to improvements in long-term evenness in yam. To realize the potential oaf the- embedded particles and fiber agglomerations would require the detection and removal of them in a continuous manner during production. Such removal might reduce the end breakage rates in spinning to a large degree, but first there must be a practical means our detecting the unwanted material. If the unwanted material can be detected, it is then worth exploring means of removal from the fiber stream during processing. But detection comes first.
Transducer
Design
This work has already stressed the importance of fiber extent and mentioned the measurement of it. In the conventional measurement of fiber length by fibrograph, a fiber beard is combed out to remove loose fibers and to parallelize the rest. In some systems of measurement, the optical density of the beard at different distances from the clamp is determined, and these measurements are translated into the number of fibers at these various distances. Only long fibers can exist at the distances remote from the clamp; the shorter ones exist nearer the clamp. The number of fibers at each distance from the clamp is used as a measure of the distribution in fiber length. In a drafting system, a fiber beard is held by the back rolls and is continuously depleted by the movement of
175
fibers influenced by the front rolls. The beard is continuously, but slowly, replenished from the input material, and the beard is held in more-or-less equilibrium. The beard is similar to a truncated version of the one in a fibrograph. In these experiments, the fringe was back-lit and viewed by a TV camera, the line-scan of which was in the direction of flow. Each line-scan measured the variation in light intensity in a manner somewhat similar to that in a fibrograph. In this case, however, the fibers were not combed out. Therefore the system was really measuring the fiber extent. The frame-scan measured the variations perpendicular to the flow. The signals from the camera were displayed on picture and wave-form monitors. The wave-form monitor showed the variations in light intensity as amplitude variations, as shown in Figure 3. The picture monitor showed that agglomerations of fiber built up in the draft zone until an avalanche occurred, which then depleted the beard. The process of build-up and depletion continued in a cyclic fashion, although there was some variation in the magnitude and frequency of the avalanches. There were also considerable widthwise variations, and sometimes the flow was divided in unequal parallel &dquo;ribbons.&dquo; It is evident that some of the errors seen in ring spinning must arise because of the failure to divide or remove these fiber agglomerations completely. Furthermore, a surprisingly large number of particles could be seen flowing through the system, and often they were associated with the agglomerations of fiber that accumulated prior to the avalanches. As discussed earlier, the removal of these particles would undoubtedly improve the performance of the ring.frame and possibly also that of the roving and drawframes.
MEASUREMENT OF LINEAR DENSITY USING PNEUMATIC TRUMPETS
All known means of continuously measuring linear density also measure some other parameter or parameters, and there is always a degree of ambiguity in the signals produced by the transducer. For example, the well-known pneumatic trumpet measures the pressure generated when a bulky sliver is compressed as it passes through the throat of the trumpet. A change in bulkiness, or in linear density, causes a change in pressure, but it is not possible with a single point of measurement to distinguish one cause from the other. The variation in pressure generated in such a system was measured on an inch-to-inch basis, and the linear density of that same piece of sliver was weighed on an inch-to-inch
the two sets of data was no better than 64%, which means that the variations in mass explained no more than about 40% of the variations in pressure. Calibration can be used to offset the changes in bulhdness caused by a change of fiber, but changes in bulk caused by previous processing still remain a problem.
basis. In this
case
176
FIGURE 5. Single-element drag40ra type of transducer. Effects of varying linear density and degree of drawing.
MEASUREMENTS WITH MULTI-ELEMENT DRAGFORCE TRANSDUCERS To resolve ambiguities in a practical staple-fiber system, it was necessary to have as many independent sources of information as there were parameters. If bulk and linear density are the two most important parameters, then a tolerable measurement system might be obtained using a transducer with at least two elements. Therefore, the next step was to make a multi-element device, which might give the several signals needed to resolve the ambiguities. Accordingly a three-element system was made as shown in Figure 8a. Again the tests were made on an Instron tensile testing machine. The results are shown in Figures 6 and 9. As mentioned earlier, neither fiber finish type nor fiber fineness had much effect, but the number of times the sliver had been drawn did have a material effect. The total drag force was divided between the three dies according to their diameter and axial position. The appendix shows that the ratio of the forces acting on the entry and exit dies should be a function of original bulk. The ratios were then plotted as shown in Figure 10. While the simple calibration curves did not seem to be much affected by handling the material, the ratios were sensitive in this respect, and it was difficult to achieve good repeatability because of involuntary changes in bulk.
The three-element device was installed in series with Uster evenness tester. Visual correlation between the linear density and force ratio records appeared good, but the ratio was also supposed to be a function of bulk. If the entry force related only to linear density, and the ratio measured bulk independently from linear density, then there should have been no correlation 2 between them. Tests at 25 m/min showed that the r2 value was in fact only 0.02 for a polyester/cotton blend. A two-element die was installed inside a sliver trumpet on a drawframe, and the drag force was measured using strain gauges (see Figure 8b). Two elements were used because of the lack of space in the congested region. Each element was mounted on a thin diaphragm on which the strain gauges were mounted. The devices performed well except for drift. The correlation coefficient between the drag-force on the input die and the linear density of cotton sliver passing through (as measured by weighing) was 75%. In other words, linear density changes explained about 56% of the signals recorded. At the second die, the correlation was poorer. The unexplained variations were thought to be largely due to changes in bulk. The transducers responded to even the shortest error wavelengths. Also, it seemed
177
FIGURE 8.
EIGURE 6. Three-element drag-force type of transducer. Effects of varying fiber finish and fineness.
that the speed at which sliver was pulled through had little effect. A cotton sliver was Suited out in five places at equal intervals and was passed through the dies without being drafted further. The results are shown in Figure 11; the bottom curve is one of the two outputs, and the top curve is a plot of the ratio of forces; the arrows indicate the places where the sliver was euffed out. The ratio did indeed respond to variations in bulk. The variations in mass/unit length after the first yard (only some of which are shown) did not correlate with the variations in the ratio of output signals. A large variation in drag force at 2 yards, which afterwards was found to be due to a local increase in linear density, did not cause a corresponding change in the ratio. Thus the method seems to be capable of discriminating between the effects of changes in linear density and bulk.
MEASUREMENT OF LINEAR DENSITY USING
TEMPERATURE READINGS
We realized that cost of a transducer system is an important factor. We further realized that most sliver manipulating machines work at constant delivery speed. Since work rate is the mathematical product of force X velocity, and since the local temperature is related to the work done in the vicinity, we thought
FIGURE 7.
a
Drag-forces obtained in pulling tow through die to show the importance of bulk.
178
FIGURE 9. Three-element drag-force type of transducer. Distributions of force between the elements as the linear density and degree of drawing are varied.
that measuring the temperature at the throat of a sliver trumpet might provide a means of measuring linear density and bulk. Thermocouples and resistance thermometers are
relatively inexpensive.
three-element transducer.
The main difficulty was that the time constant was rather long. It could not respond to errors produced in the subject machine, but only to changes in the feed material. The use of composite trumpets with nylon bodies and metal liners reduced the time constant to about a minute, but this meant that the device still could not indicate changes within some hundred yards of drawn sliver. Nevertheless, it was still a valuable function. The graph in Figure 12 shows how the temperature at the throat of a trumpet in a short-staple drawframe responded to a card sliver feed made from a deliberately tapered card lap. The variation in linear density is also plotted to facilitate comparison. A thermocouple was embedded in the throat of a sliver trumpet so that the hot junction was in contact with the flowing sliver. The cold junction was embedded at the entrance to the trumpet. Various weights of sliver were drawn and redrawn to create varying bulk (Figure 13). The system responded well to both sorts of variation. A tightly packed throat did not react as strongly to changes in the number of drawings as one that was less tightly packed. Differences in the shapes of the curves imply that it should be possible to separate the variables.
179
drag-force transducer to changes in bulk and linear density during short-staple processiaa.
temperature-type transducer
to
RGURE 12. The response of a temperature-type transducer to long-term changes in linear density.
We tried a two-element thermocouple system on a long-staple pin-drafter. Figure 14 shows the calibration curves. In this case, we found that the pin-drafter bulked the sliver being produced, and the second passage of drawing produced higher temperatures than the first. We also found that running the material through the trumpet again without drawing reduced the temperatures generated. From this it is clear that, while conventional drawing reduces the bulk of the sliver, pin-drafting increases the bulk due to the action of the pins. The bulk so generated is due to the production of fiber crimp and/or the deregistration of crimp. Later compression forces back the crimp into better registration and
so
Signal Processing
FAST FOURIER TRANSFORMS
analyzer equipped with FFT circuits coupled pneumatic transducer forming part of a sliver trumpet used in the laboratory to produce
A Data 6000
was
to a
reduce bulk.
array of periodograms similar to those in Figure I. In this case, however, they related to the sliver as it was discharged from the drawframe (the term &dquo;periodogram&dquo; is used to describe the output from a FFT analyzer and &dquo;spectrogram&dquo; to describe the output of a conventional spectrograph). With good rolls, the array of diagrams was as shown in Figure 15(a). The front top roll was then replaced with one that was slightly
an
180
eccentric, and the array shown in Figure 15b was obtained. The pattern of spikes is quite recognizable, even though they were small in relation to the irregular drafting waves. In all probability, errors of this magnitude would not have been diagnosed with single per-
iodograms or spectrograms. The analyzer was used in conjunction with several types of transducers and produced similar results.
AUTOCORRELATION
FIGURE 14. Calibration curves for a two-element temperature-type transducer used with long-staple 15 dpf carpet fiber.
The Data 6000 analyzer was used to make arrayed autocorrelograms. Many individual correllelograms showed a sinusoidal pattern, even though there were no mechanical errors present. Sequential correllelograms showed sinusoidal patterns with varying wavelengths, and this is thought to indicate a series of quasiharmonic drafting waves. The varying wavelengths
periodograms of drawn sGver with (a) drafting waves, and (b) small mechanical errors as well as drafting waves.
181
thought to be related to the changing fiber population. Over some tens of yards of sliver, the autocorrelations made a complicated pattern such as shown in Figure 16. The amplitude of the patterns was sometimes so large that it was not easy to separate mechanical errors from the drafting waves by this means. It is possible to array the correllelograms to separate the organized mechanical errors from the drafting waves, but then there is no advantage over using the arrayed
were
from the first device to a second one with appropriate biasing and a time constant that was 10 times that of the first one. The signals from the first and second devices are referred to as RMS 1 and RMS2 signals. Inspection of Figure 17 will show how RMS2 varied in a typical record, and the result may be compared to the associated RMS I signal. A strip chart can be used to show the flow of errors as production proceeds.
periodograms.
ROOT MEAN
Conclusions
We have shown that variations in fiber extent are large, especially in breaker drawing. These variations create not only direct errors, but also latent ones that will erupt in later processing. Thus the control of the early machine-fiber interactions could have a very significant effect on the product as it passes through each of the subsequent machines. Added to this is the observation that sliver contains embedded material and agglomerations of improperly opened 6ber, which can produce faults in later processing. We have demonstrated that undesirable inclusions can be detected, lf they could then be removed from the product stream at the point of detection, considerable advances in quality control might result. We have further shown that the outputs of at least two types of transducers are ambiguous, and at least two major factors influence such output. These are bulk and linear density. Where linear density is the norm by which the quality is judged, then variations in bulk cause the output of the transducer to be in ernor. In many new spinning systems, researchers are recognizing that fiber orientation is important to the operation
SQUARE VALUES
Product variance introduced by a given machine is of the performance of that machine. Variance in the feed material confuses the issue. A bandpass filter was used to pass only the signals relevant to the particular machine. Once filtered, the RMS values of the error (i.e., the square root of the variance) provided a useful measure of the performance of the machine in terms of error production. The output from a capacitive type of transducer was filtered by appropriate resistance capacity elements in the electrical circuit and was applied to a RMS semiconductor device. The RMS error values of the signals produced by the transducer varied during production as the fiber population changed, as shown in Figure 4. Frequently there were quite severe random peaks as shown in Figure 17, and these often correlated with the local thick spots. The variation in the RMS signal over a longer interval is a sensitive indicator of problems. Accordingly, the root mean square value of the RMS signal was also recorded by applying the output
a measure
182
FIGURE 17. Typical records of variation in RMS error levels in slivec RMSI root mean square of the running error over a sliding interval of 2 yards, RMS2 root mean square value of RMS l over a sliding interval of 20 yards, x response time for signal processing.
= = =
.
of the machines at high speeds. Some machine makers even recommend standards of bulkiness of the sliver,
but until there has been no way of scientifically measuring this bulk on a continuous basis. Multi-element transducers offer a way of resolving some of the difficulties. Long-term variations can be detected by using temperature measurement in sliver trumpets. With more work, these developments could lead to the development of more sophisticated and effective control systems.
now
Let
Ap
As
cross-sectional
area
cross-sectional
state
area
A1 = A2 F1 =
=
F2
throughway area of die number 1 throughway area of die number 2 drag-force in pulling sliver through die 1 drag-force in pulling sliver through die 2
(F + y) to A during compression a constant relating (F + y) to A after relaxation polynomial index during compression polynomial index after relaxation
assume
ACKNOWLEDGMENTS
Grateful acknowledgements
given
to Dr. T. K.
Ghosh for his work in adapting the Data 6000 analyzer for the specific uses, and also to G. Grover, A. Marathe, E. McLean, and R. Gaynor for their help in making measurements and operating the various pieces of
K2
x
z
equipment.
For simplicity,
exist between (F +
Appendix
The lateral compression/relaxation forces acting on
sliver
or
tow
probably produce
hysteresis
curve.
Starting from a given state, the compression curve has one characteristic, but the relaxation curve has another.
and
183
where f means &dquo;a function of.&dquo; Equations 1 and 2,
Substituting for y in
die, and there was always a degree of relaxation. Conwe concluded that any of the force ratios should respond to bulk. Also the forces acting on dies 2 and 3 should respond more to variations in bulk than those acting on die 1.
sequently
Equations 4 and 5.
Bulking affects As/Ap and K 1 /K2, but since As/Ap is quite large and is raised to the power x or z (both of which are negative), the term containing As/Ap has very little effect on the outcome. Thus a reasonable approximation is to write
Literature Cited
A. R., The Irregularity of Materials Drafted on Cotton Spinning Machinery and its Dependence on Draft, Doubling and Roller Setting, J. Textile Inst. 46, T529 (1955). 2. Douglas, K., Improving Quality and Productivity in
1.
41
(May 1984).
3. Foster, G. A. R., The Causes of the Irregularity of Cotton Yams, J. Textile Inst. 41, 357 (1950). 4. Goto, H., Ichino, S., and Kurozaki, S., Fiber Motion in Roller Drafting, J. Textile Inst. T389 (1957). 48, 5. Grishin, P. F., A Theory of Drafting and its Practical Applications, J. Textile Inst. 45, T167 (1954). 6. Lord, P. R., Short Fibres and Quality Control, J. Textile
Inst. 76, 389 (1985). 7. Lord, P. R., and Johnson, R., Short Fibres and
Quality
Control, J. Textile Inst. 76, (1985). 145 8. Lord, P. R., Stuckey, W. C., Xiuye, Yu, and Grover, G., Deblending in Roller Drafting. J. Textile Inst. 76, 339 (1985). 9. Yamashita, S., Kawabata, S., and Kawai, H., Study on Blend Irregularity Caused by Drafting, J. Textile Mach. Soc. Jpn. Eng. Ed. 14, 1, 57 (1968).
Mamucript rtaeived April 19. 1986; awepted Seprembn 9. I9~6.