Measuring of fibre orientation in nonwovens using image processing

ABSTRACT:

The modern textile industry has been facing difficult challenges in creating a high quality manufacturing environment. Such an environment requires improved quality, increased production and flexibility, reduced inventory and minimization of cost. For this purpose the textile manufacturing processes essentially need online, real time, and dynamic controls. Image processing is one such technique that involves the computer processing of pictures or images that have been converted to numerical form. It is a field that is rapidly developing owing to technological advances in the industry today. Using digital image analysis permits a more detailed analysis of basic structural parameters of linear textile products as thickness, hairiness and number of twists, also used to find the fabric defects, fiber orientation in nonwovens, etc. Computer image analysis technique used for measuring the fibre orientation in nonwovens is presented in this paper, together with a brief review of the basic concepts of image processing techniques. Fibre orientation plays an important role in defining appearance, performance, and processability of the nonwovens. It mainly depends on the uniformity of the webs. The nonwovens are expected to have a uniform mass distribution for better performance and aesthetics. Therefore visual assessments of fibre orientation are not reliable and image processing techniques are used to measure them more effectively.

1. INTRODUCTION:

The dynamic development of computer techniques creates broad possibilities for their application, including identifying and measuring the geometrical dimensions of very small objects including textile objects. Using digital image analysis permits a more detailed analysis of such basic structural parameters of linear textile products as thickness, hairiness and number of twists. This technique also enables the estimation of other characteristic features of the external structure of linear textile products, such as twist parameter and linear density coefficient. The process of identifying structural yarn parameters is a significant problem, in the light of both hitherto conducted scientific investigations and current industrial practice. On the basis of the literature considering this problem, we can state that the image processing technique enables images of longitudinal & transversal cross sections of fibres to be obtained, the fibres’ diameters to be further assessed, and images of linear textile products create which allow the observation of possible yarn faults and the determination of their causes. The images obtained can help to create two-dimensional and three-dimensional textile products, including images of spliced yarn-end connections and estimates of their correctness.

2. BASIC CONCEPTS USED FOR DIGITAL IMAGE PROCESSING:

The digital analysis of two-dimensional images is based on processing the image acquirement, with the use of a computer. The image is described by a two-dimensional matrix of real or imaginary numbers presented by a definite number of bytes. The system of digital image processing may be presented schematically as shown in Fig 1.

Digital image processing includes:

• image acquirement and modelling,
• image quality improvement and highlighting its distinguishing features,
• reinstating the desired image features, and
• Compression of image data

2.1. Image modeling:

Image modelling is based on digitising the real image. This process consists of sampling and quantifying the image. The digital image can be described in the form of a two-dimensional matrix, whose elements include quantified values of the intensity function, referred to as grey levels. The digital image is defined by the spatial image resolution and the grey level resolution. The smallest element of the digital image is called the pixel. The number of pixels and the number of brightness levels may be unlimited, although while presenting computer technique data it is customary to use values which are multiplications of the number 2, for example 512 512 pixels and 256 grey levels.

2.2. Image quality improvement:

Image quality improvement and highlighting its distinguished features are the most often used application techniques for image processing. The process of image quality improvement does not increase the essential information represented by the image data, but increases the dynamic range of selected features of the acquired object, which facilitates their detection.

The following operations are carried out during image quality improvement:

•changes of the grey level system and contrast improvement,
•edge exposition,
•pseudo-colorisation,
•improvement of sharpness, decreasing the noise level,
•space filtration,
•interpolation and magnification, and
•Compensation of the influence of interference factors, e.g. possible under-exposure.

2.3. Reinstating desired image features:

Reinstating desired image features is connected with eliminating and minimising any image features which lower its quality. Acquiring images by optical, opto-electronic or electronic methods involves the unavoidable degradation of some image features during the detection process. Aberrations, internal noise in the image sensor, image blurring caused by camera defocusing, as well as turbulence and air pollution in the surrounding atmosphere may cause a worsening of quality .Reinstating the desired image features differs from image improvement, whose procedure is related to highlighting or bringing to light the distinctive features of the existing image. Reinstating the desired image features mainly includes the following corrections:

•reinstating the sharpness lowered as the result of disadvantageous features of the image
•sensor or its surrounding,
•noise filtration,
•distortion correction, and
•correction of sensors’ nonlinearity

2.4. Image data compression:

Image data compression is based on minimising the number of bytes demanded for image representation. The compression effect is achieved by transforming the given digital image to a different number table in such a manner that the preliminary information amount is packed into a smaller number of samples.

3. MEASURING FIBRE ORIENTATION IN NONWOVEN FABRICS USING IMAGE PROCESSING TECHNIQUE:

�Fibre orientation is an important characteristic of nonwoven fabrics, since it directly influences their mechanical properties and performances. Since visual assessments of fibre orientation are not reliable, image processing techniques are used to assess them automatically. The image of a fabric is taken and digitized. The digitized image of the surface of a fabric is progressively simplified by a line operator and an edge-smoothing and thinning operation to produce an image in which fibres are represented by curves of one pixel in thickness. The fibre-orientation distribution of a fabric is then obtained by tracing and measuring the length and orientation of the curves. The image-analysis method for automatically measuring the fibre-orientation distributions is validated by manual measurements made by experts.

4. SPATIAL UNIFORMITY OF FIBROUS STRUCTURES:

Spatial uniformity of fibrous structures can been described statistically using an index of dispersion. The spatial uniformity of web mass is described by dispersion of its surface relief distribution. Surface relief is representative of mass density gradients in local regions. In principle, this concept is similar to topographical surface relief that represents the variations in elevation of the earth’s surface: the higher is the elevation, the greater the mass present on a surface. The surface relief in a local region is representative of the average mass density gradient in that local region. Surface relief of a web can be measured from its gray-scale images taken in transmitted or reflected light or any other radiation. Each pixel of a gray-scale image (with 256 gray-levels or 8-bit) is considered a column of cubes with each cube having area (1 pixel) X (1 pixel) and height equal to one intensity level. Fig 2 shows the translation of square pixels of Fig 3 into columns of cubes. Surface relief of a pixel is the difference in height (or graylevel intensity, G) of two adjacent pixels sharing a side. The surface relief area was calculated as the total number of exposed faces of cube columns in a “quadrat,” which is defined as a rectangular region of interest. Only lateral surfaces were considered and flat tops of columns were ignored in the calculation of relief area.

4.1. Index of dispersion:

The dispersion of the quadrat surface relief area represents the uniformity of mass distribution. Selection of an appropriate measure of dispersion is based on the following criteria:

•it should describe the complete spatial variation in mass(from maximum uniformity to randomness to maximum aggregation)
•it should be relatively independent of total mass and mass density in the region
•it should be standardized

To compare the uniformity two or more spatial distributions (an important aspect from the quality control perspective): a standardized scale should include the confidence limits on the randomness of the spatial distribution. Based on above criteria, the variance-to-mean ratio is selected as the index of dispersion.

Since the surface relief is a measure of the local mass density gradient, a small deviation in the local mass density gradient from the average mass density gradient (small σ2) suggests the presence of a uniform mass distribution (small Id). Consequently, the dispersion index appropriately captures the physical meaning of uniformity of the mass distribution.

For a random pattern, described by Poisson distribution, the variance is equal to the mean, or Id = 1. Therefore, the interpretation of statistics becomes significantly straightforward: Id > 1 means clumped distribution; Id = 1 means random distribution; and Id < 1 means uniform distribution. Spatial randomness (non-uniform, yet not aggregated) was evaluated via fitting the observed distribution to the Poisson distribution. The null hypothesis of randomness is tested by a χ2-statistic obtained from the dispersion metric using the following equation:

Where, n = number of quadrats in the selected web area.

The spatial pattern was considered random if the observed χ2-statistic is between the 95% confidence limits. If the observed value is lower than the lower confidence limit, the spatial pattern is uniform, while a value higher than the upper confidence limit was indicative of an aggregated pattern. The dispersion index was standardized between –1.0 and 1.0.

Quadrat analysis was used to evaluate web uniformity. Based on the desired scale (e.g., length scale from post processing), the image was divided into N x M quadrats. Figure 4 shows an image of adhesive laydown pattern divided into 49 quadrats. Quadrat relief area of each quadrat was calculated using the formula. Both row-wise (representing machine direction, MD) and column-wise (representing cross-direction, CD) mean and variance were evaluated. Mean and variance of MD and CD were pooled from N rows and M columns, respectively.

MD and CD indices of dispersion were calculated as their respective ratios of variance-to-mean. Since most webs have preferred mass orientation in the machine direction, mass distribution is anisotropic. Therefore, anisotropy of mass dispersion (not to be confused with fiber orientation) was calculated as the ratio of CD-to-MD dispersion: a value greater than 1.0 means a greater variation of mass in the cross-direction than in the machine direction and vice versa; a value of 1.0 means isotropic mass distribution. The overall index of dispersion was calculated by pooling MD and CD variance and dividing it by mean quadrat relief of the image as follows: The overall dispersion index calculated from equation (11) was standardized in the range of –1 to 1. In the case of fibrous webs, the statistics were interpreted as the uniform mass distribution more desirable than the random distribution, which in turn was more desirable than the aggregated distribution, unless otherwise needed. Based on above statistics, webs were ranked in the order of their standardized uniformity values.

4.1.1. Implementation:

The standardized index of dispersion was tested on real as well as simulated patterns by dividing the patterns into quadrats of different areas ranging from 2 mm2 to 100 mm2. For a given quadrat size and number, the same image was sampled multiple times by shifting the position of horizontal grid lines by a known amount (10% of height) in the downward direction (CD) while keeping the column positions constant. This was done to minimize the dependence of dispersion index on quadrat positioning, which could potentially render the same pattern clumped or uniform. The variance and mean of all samplings were pooled to evaluate the standardized dispersion index and dispersion anisotropy. Statistical stationarity of images was assumed during multiple samplings of the same image. The dispersion evaluation algorithm was implemented using FORTRAN 90, and critical values of χ2 distribution were evaluated using IMSL routines. Images of webs were taken using digital cameras with field of view at least 5 cm x 5 cm. Nonwoven webs were imaged under reflected white light with a resolution of 84 pixels/cm. Colored images were converted to gray-scale using digital color separation that kept the channels having maximum amount of intensity information (evaluated using information entropy).Histogram equalization (flattening) was done to remove any illumination intensity variation within and between images. Removal of illumination variation (lighting instability) via histogram flattening improves the repeatability of data. In all images, lighter regions represented denser regions of webs. Dispersion evaluation was then performed on the preprocessed images.

Uniformity evaluation was performed on four simulated patterns and four nonwoven webs. Figure 5 shows a simulated 30 gsm, 4.1 denier, continuous fiber web, while Figure 6 shows a simulated 20 gsm, 10.5 denier, staple fiber web. Both simulated images had a resolution of 500 pixels/cm. Of the four nonwoven webs tested, there were two spunbond (15 gsm and 60 gsm), one carded-thermobond (31 gsm), and one spunlace (58 gsm) nonwoven.

4.1.2. Results:

Results show that patterns with different mass densities can be compared using the same standardized index. Furthermore, with an increase in quadrat area, patterns tend to become more uniform. This result is expected since larger size quadrats will average out the variation of relief in local neighborhoods and the pattern will appear uniform.

Figure 7 shows uniformity grading of nonwovens as a function of quadrat area. Five uniformity grades are defined in Figure 7 with the grade 1 assigned to the most uniform distribution and the grade 5 assigned to the least uniform distribution. All four real nonwovens had standardized index of dispersion between -1.0 and -0.5 (between grades 1 and 5), making them uniform patterns. However, simulated nonwovens were far more uniform than the real webs for all quadrat areas. All real nonwovens were least uniform (greater than grade 2 uniformity) for 2 mm2 size quadrats and most uniform (grade 2 uniformity) for 100 mm2size quadrats.

Figures 9 depict the anisotropy in mass uniformity as function of quadrat size. Mass anisotropy increases with mm2. Interestingly, for 100 mm2 quadrat size, simulated clumped pattern had maximum uniformity since the aggregates are uniformly spaced for 10 mm x 10 mm quadrat size. Mass anisotropy increases with increasing quadrat size for nonwovens .Figure 9 also shows spunlace web to be more dispersed in MD than in CD. Again, such anisotropic behavior is not captured by the standardized index of dispersion in Figure 7.

The standardized index of dispersion is reasonably independent of basis weight. Therefore, the standardized index can be used to compare nonwovens of different basisweights (from less than 10 gsm to more than 100 gsm). The standardized scale can be correlated to customer specifications or pre/post-processing variables. Furthermore, the standardized scale has been combined with the five qualitative grades to categorize and differentiate uniform patterns. Therefore, it can also be used for quality control purposes. Anisotropy of mass uniformity can be measured for different webs. When combined with standardized index of mass uniformity, the anisotropy ratio can be a very useful tool to differentiate between different webs with a similar standardized index of dispersion. A robust test methodology can be developed by combining the standardized index of dispersion with the mass anisotropy ratio for quality control as well as characterization of mass distribution in nonwovens, adhesive laydown patterns, or any other substrate. In principle, the methodology can be applied to any two-dimensional pattern or data set to quantify spatial uniformity.

5. CONCLUSION:

The development of computer techniques offers many opportunities for its applying in textile science and practice. Use of computer image analysis, among other techniques, has enabled the identification of geometrical dimensions of very small textile objects. Using image correction techniques allowed the elimination of structural faults in the fabric which earlier would have been ignored. Applying image correction techniques enables a detailed identification of the structure and geometry of linear textile fabrics. Elaborating the digitisation algorithm, combined with numerical methods, allows the numerical characteristics of a textile product’s structure to be obtained. Qualities of the textile materials have gained importance in the present quota free trade which is highly competitive. So use of these computer techniques has become mandatory which will help in improving the quality.

6. REFERENCES:

1)AUTEX Research Journal, Vol. 6, No 1, March 2006
2)Textile Research Journal, Sep 2002
3)Aggarwal, R. K., Kennon, W. R., and Porat, I., “A Scanned-laser Technique for Monitoring Fibrous Webs and Nonwoven Fabrics,” Textile institute.
4)Ericson, C. W. and Baxter, J. F. “Spunbonded Nonwoven Fabric Studies”
5)B.K.Behera, “Image Processing in Textiles” Textile institute.
6)Russ, J. C., “The Image Processing Handbook,”
7)www.inda.org
8)www.textiles.edu
9)www.corel.com
10)www.cyto.purdue.edu