3-D Electronic Imaging of Fabric Qualities by On-Line Yarn Data

Transcription

1 1 3-D Electronic Imaging of Fabric Qualities by On-Line Yarn Data Code: Investigators: Students: S01-NS2 (formerly I01-S12) Moon W. Suh (NCSU) - Leader Warren Jasper (NCSU) Melih Gunay (NCSU) Project Goals The primary goal of this project is to design and develop a control system for electronically imaging the quality attributes (basis weight, appearance uniformity, physical properties, etc.) of woven and knitted fabrics directly from an on-line yarn mass/diameter measurement system without having to go through the actual fabrication processes. Such a system is to facilitate optimal yarn and fabric development to achieve the desired uniformity or to impart aesthetically pleasing "non-random random effects (chaos)" within the yarn and the resulting fabrics under the data-rich modern manufacturing environment. The secondary goal is to perfect the data reduction and expansion techniques that are amenable to mapping and imaging the micro and macro dimensions of the resulting fabrics suitable for quality prediction, modification and mapping in time domains and two-dimensional spatial domains. The third goal is to perfect a "data fusion" technology based on the optical and capacitor type measurements of yarn diameters and yarn density profiles ideal for imaging the quality attributes of the yarns and fabrics. Abstract This research is aimed at modeling, prediction, imaging and visualization of fabric properties in 2-D and 3-D in accordance with newly defined variance-area curves. The yarn diameter profiles are captured from an on-line measurement system developed within a Lawson-Hemphill CTT. Electronic images are obtained for yarn and fabric quality profiles such as yarn diameter variation, uniformity of fabric surfaces and other physical properties. The variance-area curves have been defined and applied to measure and express the measured fabric qualities in mathematical functions. In addition, we developed a measurement system for obtaining multiple photo images of yarn profiles at different angles along the length of the yarn simultaneously. It was found that, the shape of yarn cross-sections can be approximated by ellipses well with little error. It was also found that the eccentricities or the shapes of ellipse vary along the length of the yarn following a form of statistical distribution. Elliptic approximation of yarn cross-section required the estimation of the major and minor axes and their changes along the yarn length. Furthermore, the angle of the major axes rotated periodically with twists. This discovery allowed us to estimate the number of twists per unit length or the uniformity of twist distribution. The system we developed greatly enhances our capability for imaging and visualization of fabric qualities and provides a means for pinpoint quality mapping for problem isolation required for fabric development.

2 2 Introduction During the last decade, the textile industry throughout the world has been under extreme pressure to produce high quality products by improving yarn and fabric qualities. This is primarily due to the increased consumer awareness on product qualities and fierce global competition. Of many quality characteristics, perhaps none is more important than the uniformity of yarns and fabric appearance. While the fabric appearance is directly tied to uniformity of the yarn, no mapping was possible until recently due the lack of measurement technology and slow speed of computing. Inherently, all yarns are subject to periodic and random variations. However, their effects on the fabrics are difficult to predict. Currently non-uniformity of fabric appearance is assessed by the standard yarn board that reflects yarn diameter variations. In addition to traditional yarn boards, fabric samples are often produced by actual weaving or knitting although it is expensive and time consuming. Although several authors (Wegener 1986, M.C.Moyer 1992) suggested that quality of fabrics can be predicted by the CV(%) of the yarn used, the CV(%) alone is grossly insufficient to predict the features of irregularities as the single number is not location specific within the fabric. In this research, we showed that aesthetics and appearance qualities of fabrics can be predicted without having to weave or knit from the information obtained on the yarn properties (mass, diameter, etc.) either on-line or off-line through simulation and mapping techniques we developed. Extensive research work was carried out by many researchers in the past to develop a system for characterizing numerous fabric properties, some through introduction of quality indices others by new methods of measurement using time series, Fourier transform and Wavelets. Recently, Snyder (2000) designed a fabric quality rating system called Total Quality Index similar to the KES (Kawabata) System which combined the objective measurements with subjective rating methods. The total quality index by Snyder was composed of three sub-indices; the defect index (N), streakiness index (S), and cloudiness index (C). Noting the difficulty in assessing fabric irregularities by a single value, Townsend and Cox (1951) showed that the relationship between coefficient of variation and the length within which the variance is measured leads to indices characterizing the types of irregularity that has practical significance. Townsend and Cox (1951) and Breny (1953) developed the concept of variance-length curves by computing the CV(%) of yarn mass based on variable unit lengths and then partitioned the total variance into two parts; CV(L) and CB(L). By the shapes of the V(L) and B(L) curves, they tried to characterize the uniformity of the yarn along the yarn axis, and then expanded the traditional CV(%), a point estimate, to a series of numbers expressible in two curves. They showed that an identical CV(%) value can give rise to many different CV(L) and CB(L) curves, providing a much more powerful method for discriminating the qualities of the spun yarns. The variance-area curve concept we apply here is the very same idea. Namely, we want to expand CV(%) of fabric mass into a series of variance components which, in turn, form CV(A) and CB(A) curves. The two curves are hoped to characterize the aesthetic and uniformity characteristics of the fabric with an enhanced discrimination power. In order to generate more realistic yarn signal for fabric simulation, we studied various yarn cross-sections. Considering the variable yarn tension, twists, and diameters, we expected that the yarn cross-sections might not be circular by and large. In reality, it was reported that the yarn

3 3 cross-sections tend to form irregular shapes of close-packed polygons (Hearle, Grosberg and Backer 1969). It was also shown (Tsai and Chu 1996) that the cross-sections of ring-spun and open-end (OE) yarns can be approximated as ellipses, even though the outlines of the actual shapes may be irregular. The estimation of yarn cross-section by conic sections requires application of results from modern computer graphics research. One difficulty in finding the best-fit ellipse is that there are few direct methods for estimating (in a least squares sense) an ellipse from noisy data. Two common techniques involve solving a generalized eigenvalue problem with an associated non-convex constraint or by iteration minimizing the cost function for optimum solution. Techniques also vary in terms of the fit being arithmetic (Gal n.d.) or geometric (Fitzgibbon, Pilu and Fisher 1999). It must be remembered that the eccentricity, which is essentially the ratio of minor axis to major axis, equals 0 if the conic is a circle, 1 if the conic is a parabola, and between 0 and 1 if the conic is an ellipse. Instrumentation Two systems were developed for analyzing yarn profiles. The first system measures yarn crosssections in a constant speed/tension zone. The system is comprised of a Thompson CCD line scan camera, a DIPIX Video Board and a Lawson-Hemphill CTT (Constant Tension Transport) machine. An encoder pulse from the CTT was generated every millimeter to trigger the line scan camera to capture a yarn image. The Thompson CCD line-scan camera can take images at 500 Hz (approx. 24 m/min), with a resolution of 1024 pixels/scan. A software program was developed to interact with the camera and to transfer the captured image to a computer running Linux/RT-Linux for further analysis. To capture two-dimensional continuous images, a second imaging system was developed. This system is comprised of a Pulnix TM progressive scan camera connected to an Engineering Design Group (EDT) frame grabber board. It has a interline CCD with pixels of 1024(H) by 1024(V), of which 1008 x 1018 are active. The camera is capable of taking 15 frames per second, and has a shutter speed of up to 1/16,000 sec. The video output is BW 8-bit RS-422 with a S/N ratio of 50 db. It is important to keep the yarn from distortion for accurate determination of cross-sections. For this, a makeshift jig was built and placed in front of the high-resolution area camera. The yarn was then tightened from both ends as it was rotated along its axes at 6 equal angles (0, 30, 60, 90, 120, and 150 degrees). In every position, the image of the yarn was taken and, by using an image-processing technique, the diameter of the yarn was determined. We have written a user interface shown in Figure 1 to control the system. This user interface is capable of controlling the real-time data capture of the yarn diameters. In addition, the application creates a virtual woven fabric online with the real-time yarn data, providing various types of statistical information for the yarn simultaneously. The front end of the application is written in Java Swing because of its portability and platform independence. The back end which controls and commutates with the line-scan camera, is written in C. The interaction between the Java Virtual Machine and the C runtime environment is achieved through the Java Native Interface.

4 4 Figure 1. Yarn and fabric profiles created online based on the software developed Technical Approach The process of obtaining accurate measurements of yarn cross-sections presents a difficulty because the process itself may be unstable primarily due to the stability of the yarn at a given angle. While it is impractical to cut the yarn at various locations and examine the cross-sections without distorting the yarn, fast and accurate determination of yarn cross-sections can be achieved indirectly by rotating the yarn segment around its center. Figure 2 shows the projected diameter d when light is shed on a strand of yarn with semi-major and semi-minor axes a and b, respectively. The derivation of the diameter d in terms of a, b and α yield the following equation: d b + a tan ( α) = ab (1) 4 4 b + a tan 2 ( α) Figure 3b shows a typical yarn cross-section that is obtained using a line-scan camera by rotating the yarn along its axis. The dots in Figure 3b mark the measured diameters at various angles. In this case, the yarn s irregular cross-section is shown to be approximated by a best fitting ellipse with an eccentricity measure of 0.46.

5 5 Figure 2. Yarn s elliptic cross-section and it projected diameter a) Mapping a yarn cross-section with hairs b) Fitting an ellipse to a yarn cross-section Figure 3. Yarn cross-sections When several yarn cross-sections were examined as described above, the typical yarn was found to have an elliptical profile with varying eccentricity measures along its length while the major axes were rotating periodically with the twists of the yarn (Jasper, Gunay and Suh 2004). Besides, it was also shown that the twist distribution of the yarn could be estimated indirectly by determining the period of the best fitting model to actual yarn signal as shown in Figure 4. In addition, it was demonstrated that even a regular yarn with elliptic cross sections could have a CV as high as 7.2% due to the rotation of the major axes only. We believe that the elliptic nature of yarn cross-sections and the interaction between the levels of twists and eccentricity play an important role in characterizing the surface properties of the fabrics. Characterization of Textile Structures Characterization of a textile structures could be accomplished in several ways. Although it is a very difficult problem, it is theoretically possible to use a numerical code (analogical to DNA) to represent all properties (aesthetic, strength, handle etc.) of a 3-D structure. One approximation may involve reduction of a known property of the structure to a 2-D matrix array (e.g. mapping the mass distribution of a fabric) and then reduction of this 2-D matrix array into a functional form or a curve (e.g. development of a variance-area or a mean-area curve) and finally creation

6 6 of a quality index from the curve. It must be noted that in every step of the transformations, some information will be lost depending on the technique used. Figure 4. Real vs. simulated yarn signals resulting from rotation of twists In this research, the diameter profiles of yarns are collected and transformed into a twodimensional matrix array (i.e., virtual fabric) by the imaging and mapping system developed which assigns each and every pixel to a specific location x, y within a virtual 1x1 woven fabric This transformation was called basic mapping. While basic mapping was done based on zero warp take-up and filling crimp, it can be modified to simulate woven as well as knitted fabrics with various cover factors. This proto-type system allows us to construct a virtual fabric. As most of the past work on yarn irregularity was centered on mass property of textiles, we attempted to follow the prior path by studying the diameter measurements of the yarn cross sections and linking them to mass irregularities that are known to be highly correlated (Jasper, Suh, Woo, Gunay and Kim 2000). The 2-D images are intended to show a macro view of the fabric mass variation or other quality attributes in a roll of certain size for comparison and rating purposes. First, the virtual fabric, which was constructed using basic mapping, is partitioned into sections, where each section is summed and then normalized by averaging. In some instances, the selection of appropriate partition size and the application of creative transformation were necessary for optimum visualization. The normalization technique used is similar to the one suggested by Martindale (Martindale 1945). In the next step, each section of the fabric was represented with in gray scales based on the normalized value obtained. The difference from the mean value, expressed in either a darker or lighter shade corresponds to a heavier or a lighter area of the fabric, respectively. Figure 5, 6 and 7 illustrate the gray-scale images of various virtual fabrics constructed with real yarn signals by applying the techniques mentioned above. The distribution of the gray pixel indicates the diameter variation and the characteristics of the irregularity. In ideal conditions, we should expect a uniform mass distribution throughout the fabric, but as seen clearly in the following picture, there are regions where the mass densities are quite different from the rest. For the 3-D images, we assumed that the micro images represent the mass density variations for a selected small area.

7 7 Figure 5. 2-D macro and 3-D micro images of a woven fabric (Dimension: 1.5 m x 20 m) Figure 6. 2-D macro Images of a woven fabric (Dimension: 0.5 m x 5.0 m) Figure 6 shows a woven fabric of width 0.5 m and length of 5.0 m. The overall mass or diameter variation of the roll can be seen. Figure 7 shows a fabric of size 0.5 m x 1.0 m woven from a NE 17 ring spun yarn. With software we developed, it is very easy to select any region on the fabric and plot a 3-D micro image of the selected region for visualization of local variations. Also, by applying appropriate mathematical transformations to different fabrics, the imaging system can magnify the variations among these fabrics, thus enabling us to view density variations that are normally not easy to see.

8 8 Description of Irregularity in a Spun Yarn as a 2-D Structure The irregularity in a spun yarn has long been investigated and the concept of variance-length curves was studied by several researchers in the past (Cox and Townsend 1951, Olerup 1952, Breny 1953, Suh 1976). Suh expressed the squared CV (L), the ratio between the variance of mass per-unit-length (L) and the square of the corresponding mean, as a function of fiber length Figure 7. 3-D Micro images of a woven fabric (Dimension: 0.5 m x 1.0 m) distribution. While the work was viewed significant, a similar concept has not been widely extended to fabrics for studying the mass variation within certain area of the resulting woven or knitted fabric. The variations in the raw fabric mass are affected by a number of factors mainly the irregularities of warp and weft yarns and the irregularities caused the process itself (Wegener 1986). In parallel to the method used by Suh (1976) for the variance length curve, the variance area curve can be defined as the ratio between the variance of property per-unit-area and the square of the corresponding mean, or square of the more commonly known coefficient of variation. Although, variance area curves were studied before, it has not been applied extensively due to lack of computing power that is available now. Wegener (1986) showed that under ideal conditions the laborious and time time-consuming determination of the fabric irregularity (surface variation) can be avoided by estimating the irregularity of textile surface from the irregularity of the yarn itself. Our initial research suggested that the relationship between yarn irregularity and fabric irregularity is not always proportional and the shape of variance-area curves determines the characteristics of irregularity.

9 Variance Area Curves NTC Project: S01-NS12 (formerly I01-S12) 9 The variance area curves are constructed by plotting the coefficient of variation as a function of the unit area (= A) for predetermined values of A i where i = 1, 2,, n. The size of each unit area is determined by the width (W) and length (L) that can be chosen somewhat arbitrarily. Based on the theories we developed for variance-area curves, CB(A) and CV(A), we compared a dozen spun yarns and the resulting images producible from CYROS system as well as from the mathematically mapped fabrics. A few examples of CV(A) and CB(A) curves are shown below for illustration purposes. Not only do we see the differences in shapes between the two sets of graphs resulting from two yarns, we can also obtain the specific values of the within- and between-area variances at different sizes of A. In addition to the 2-D macro images above, the variance-area curves provide numerical values as additional comparison criteria. Figure 10. Two sets of CV (A) and CB (A) curves for two different yarns Conclusions Several yarn profiles were examined and it was determined that their cross-sections can best be approximated by ellipses, where the semi-major and semi-minor axes of the ellipses rotate with twist. It was also determined that the majority of the optical variation in yarn evenness, for yarn with an eccentricity greater then 0.5, was due to the rotation of the ellipse with the twist. It was also demonstrated that even a regular but elliptic yarn could have a CV(%) as high as 7.2 because of the rotation of the major axes within an ideal yarn. We think that this nature of yarn cross-section and the interaction between the level of twist and eccentricity plays an important role in the properties of textile structures. In order to visualize the uniformity of certain physical properties of a fabric, the yarn has to be property mapped which we call basic mapping. As an example, we can map yarn diameters measured at 1mm intervals onto a fabric woven from the yarn. Using this basic mapping, variance-area curves have been calculated. In addition to the 2- D macro images shown above, this research demonstrated that the variance-area curves have a wide array of applications for characterization and discrimination of certain fabric properties within a reasonable dimension of woven fabrics produced from a spun yarn.

10 Acknowledgements NTC Project: S01-NS12 (formerly I01-S12) 10 We deeply appreciate the technical assistance from Keisokki Company, Japan and Lawson- Hemphill. In addition, we also acknowledge the contributions made by Prof. A. Cherkassky (Shenkar College, Israel) during the early stage of the project development. References Breny, H. The Calculation of the Variance-Length Curves from the Length Distribution of Fibers J. Text. Inst. 44, P1 9, Fitzgibbon, A., Pilu, M. & Fisher, R. Direct Last-squares Fitting of Ellipses PAMI 21(5), , 1999 Gal, O. (n.d.). ellipse.m, Golub & Loan, V. Matrix Computations 3 rd edition, Johns Hopkins University Press, Han, D. Development of Fabric Image Models and Invariance Property of Variance-Area Curve within Fabric M.S. Thesis, College of Textiles, N.C. State University, Raleigh, NC, Hearle, J., Grosberg, P. & Backer, S. Structural Mechanics of Fibers, Yarns, and Fabrics Wiley- Interscience. New York, NY, 1969 Jasper, J., Gunay M. & Suh M. W. Measurement of Eccentricity and Twist in Spun Yarns Textile Research Journal (in review), Jasper, W. J., Suh, M. W., Woo, J. L., Gunay, M. & Kim, H. B. Real Time Yarn Characterization and Data Comparison using Wavelets 2000 Annual Report (I97-S1), National Textile Center. Martindale, J. G. A New Method of Measuring the Irregularity of Yarns with some Observations on the Origin of Irregularities in Worsted Slivers and Yarns J. Text. Inst. 36, T35 47, Moyer, M.C. Using Variance Length Curve Analysis to Determine the Impact of Medium Term Yarn Variations on Fabric Appearance Institute of Textile Technology, Olerup, H. Calculation of Variance-Length Curve for an Ideal Sliver J. Tex. Int. 43, P , Snyder, B. T. Development of a Yarn Quality Rating System for Visual Fabric Qualities Master s Thesis, College of Textiles, North Carolina State University, Raleigh, Suh, M. W. Probabilistic Assessment of Irregularity in Random Fiber Arrays-Effect of Fiber Length Distribution on Variance-Length Curve. Textile Research Journal 46(4), , The Mathworks Inc. Image Proccessing Toolbox. Natick, MA, Tsai, I. & Chu, W. The Measurement of Yarn Diameter and the Effect of Shape Error Factor (SEF) on the measurement of Yarn Evenness J. Text Inst. 87(3), , Townsend, M. W. H. & Cox, D. R. The Analysis of Yarn Irregularity J. Text. Inst. 42, P , Wegener,W. The Irregularity of Woven and Knitted Fabrics J. Text. Inst. 77(2), 69 75, Project Website: