The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Quality of ladies’ Relationship between the knitted fabrics warm/cool feeling of fabric and the subjective evaluation of the quality 7 of ladies’ knitted fabrics Takako Inoue and Akira Nakayama School of Life Studies, Sugiyama Jogakuen University, Nagoya, Japan, and
Received 14 November 2008 Accepted 12 May 2009
Masako Niwa Nara Women’s University, Nara, Japan Abstract Purpose – The purpose of this paper is to analyze the relationship between the warm/cool feeling of the heat properties of fabrics and the subjective evaluation of the quality of ladies’ garment fabrics. Design/methodology/approach – Regression analysis is conducted using stepwise block regression applied to the expert judges’ judgment value total hand value, using six blocks of the mechanical properties and one block of the initial maximum values qmax of the heat flux of the heat properties of spring and summer tailored-type jacket fabrics, as the seven blocks of fabric properties, including the secondary term of each property. Findings – The results of the regression analysis show that the qmax values do not affect the subjective evaluation of the quality of spring and summer tailored-type jacket fabrics. The results of the regression analysis of ladies’ knitted fabric properties applied to the subjective evaluation value have confirmed that the qmax values affect the subjective evaluation of the quality of ladies’ knitted fabrics. Originality/value – This paper usefully describes the relationship between the warm/cool feeling of fabric and the subjective evaluation of the quality of ladies’ knitted fabrics. Keywords Mechanical properties of materials, Clothing, Fabric testing, Heat measurement Paper type Research paper
Introduction For the objective evaluation of fabrics that are used in diverse ways in ladies’ garments, we have been conducting analyses using basic mechanical properties, and heat properties such as heat transport properties and air permeability. The subjective evaluation of the quality of ladies’ garment fabrics involves not only basic mechanical properties, but also heat properties. In this study, we analyze the relationship between the warm/cool feeling of the heat properties of fabric and the subjective evaluation of the quality of ladies’ garment fabrics. We also clarify the effects of the warm/cool touch feeling on the subjective evaluation of the quality of ladies’ woven fabrics and knitted fabrics. Experimental Test sample collection The test samples were 134 types of woven fabrics for spring and summer tailored-type jackets which, we have used in previous research (Inoue and Niwa, 2009), and ladies’
International Journal of Clothing Science and Technology Vol. 22 No. 1, 2010 pp. 7-15 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011008767
IJCST 22,1
8
knitted fabrics newly collected from around the world (70 types for spring and summer and 65 types for autumn and winter). The samples of ladies’ knitted fabrics were collected broadly in diverse ways covering various uses. The mechanical properties of woven fabrics for spring and summer tailored-type jackets were measured in a previous research (Inoue and Niwa, 2009) (see the Appendix). Sensory tests of fabric warm/cool touch and subjective hand-evaluation of the quality of fabric Sensory tests of the fabric warm/cool touch were conducted to investigate the relationship between the measured qmax value and the fabric warm/cool touch. The tests were conducted once in the summer and once in the winter (for a total of two times). The room temperature at the time of the tests was 258C in the summer, and 208C in the autumn. Sensory tests of the subjective hand-evaluation of the quality of the fabrics were also conducted two times. The subjects representative of the consuming public were 20 females, and five males were included as reference subjects. Sensory tests of the fabric warm/cool touch were made using ranking numbers 5 (feels cool), 4 (feels slightly cool), 3 (no distinction between warm/cool touch), 2 (feels slightly warm), and 1 (feels warm). The total hand value (THV) evaluation was standardized following the standardization of the total hand, using ranking numbers 5 (excellent), 4 (good), 3 (average), 2 (below average), and 1 (poor). The initial maximum values qmax of heat flux and air resistance The initial maximum values qmax of the heat flux of the fabric heat properties were measured using a KES-F7 thermo labo, and these values were used as the values related to fabric warm/cool touch (Imai et al., 1987). The air resistance AR was measured using a KES-F8 air permeability tester. Results and discussion The test samples are 134 types of woven fabrics for spring and summer tailored-type jackets which have been used in the previous research (Inoue and Niwa, 2009) of the objective evaluation of the quality of ladies’ garment fabrics to compare ladies’ knitted fabrics. For the spring and summer ladies’ knitted fabrics, cotton and blended fibers each make up more than 30 percent of the group, while other fibers used in this group include linen, silk, and polyester. The blended fibers consist of cotton or rayon blended with polyurethane or nylon. More than 50 percent of the autumn and winter ladies’ knitted fabrics are blended fibers; other fibers used in this group include cotton and wool. Many of the blended fibers consist of wool or acrylic blended with polyurethane or nylon. Weft-knit fabrics comprised 94 percent of the spring and summer fabrics and 96 percent of the autumn and winter fabrics. Figure 1 is a distribution map of the qmax value and fabric weight. The weight per unit area of autumn and winter ladies’ knitted fabrics is higher than that of spring and summer ladies’ knitted fabrics. The distribution is wide, from heavy to light. The qmax values of spring and summer ladies’ knitted fabrics are higher than those for autumn and winter ladies’ knitted fabrics, and the distribution is slightly wider than that of autumn and winter ladies’ knitted fabrics.
Quality of ladies’ knitted fabrics
qmax (kW/m2)
2
1
9 6 5 4 3 9 10
2
3
Weight (mg/cm2) Notes: : Spring and summer knitted fabrics (n = 70); : autumn and winter knitted fabrics (n = 65)
The qmax values when a knitted fabric is arranged in two layers were measured using the same procedure. A difference in the qmax values between two- and one-layers knitted fabrics was not recognized for spring and summer ladies’ knitted fabrics. However, the distribution was narrow when the knitted fabric was in two layers. For autumn and winter ladies’ knitted fabrics, the qmax values when the knitted fabric was in two layers show a tendency to be lower than the qmax values of one-sheet fabric. From the results of the correlation of the sensory test of the warm/cool feeling in summer and autumn, the reproducibility of the warm/cool feeling of ladies’ knitted fabric is recognized collectively. The relationship between the mean value of the sensory tests with the 25 “consuming public” judges and qmax values is shown in Figure 2. A correlation with the qmax values was recognized collectively in the summer test and the autumn test. However, the correlation coefficient in the summer test is slightly lower, and the same results are seen in research by Imai et al. (1987). The correlation coefficient between the mean value of the female sensory test and the qmax values is higher than those between the mean value of the male sensory test and the qmax values, and for both the summer test and the autumn test, the correlation coefficients with the qmax values are high collectively. The relationship between the mean value of the sensory test and the qmax values was plotted by dividing the test samples into spring and summer ladies’ knitted fabrics and autumn and winter ladies’ knitted fabrics. The qmax values of the spring and summer ladies’ knitted fabrics were distributed more widely than the qmax values of the autumn and winter ladies’ knitted fabrics, and a correlation with the value of the sensory test is recognized. The correlation coefficient between the qmax values of autumn and winter ladies’ knitted fabrics and the mean value of the sensory test is low. The reason for this is thought to be the fact that the distribution of qmax values is narrow. The relationship between the thickness of the knitted fabrics and the standard deviation of the sensory test was plotted, but the effect of fabric thickness is not seen. The initial maximum values qmax are shown in Figure 3. The initial maximum values of heat flux ranked highest with the spring and summer ladies’ knitted fabrics,
Figure 1. Distribution of qmax and weight
IJCST 22,1
6.0
10
Mean values of sensory tests
5.0
4.0
3.0
2.0
1.0
Figure 2. Correlation between the value of warm/cool evaluation and qmax
0
0
1.0
2.0
3.0
qmax (kW/m2) Notes: : First sensory test; : second sensory test; deviation
: ±standard
qmax (kW/m2)
2
1 9 8 7 6 14
16
18
20
22
24
26
28
Weight (g/cm2)
Figure 3. Relation between qmax and fabric weight
Notes: The values of the mean and standard deviation of each group of fabrics are plotted in this figure; : tailored-type spring and summer fabrics (n = 134); : spring and summer knitted fabrics (n = 70); : autumn and winter knitted fabrics (n = 65); : ±standard deviation
followed by the spring and summer tailored-type jacket fabrics, and then the autumn and winter ladies’ knitted fabrics. It was clear that spring and summer ladies’ knitted fabrics provide the coolest feeling of the fabrics tested. The relationship between the weight per unit of fabric and air resistance AR is shown in Figure 4. The air resistance of the spring and summer ladies’ knitted fabrics
Quality of ladies’ knitted fabrics AR (kPa·s/m)
0.6
0.4
11
0.2
0 9 10
2
3
Figure 4. Air resistance of ladies’ knitted fabrics
Weight (g/cm2) Notes: : Spring and summer knitted fabrics (n = 70); : autumn and winter knitted fabrics (n = 65)
is low and the weight per unit of fabric is low. The air resistance and the weight per unit of fabric of autumn and winter ladies’ knitted fabrics are distributed widely, and the standard deviation is high. It is conceivable that this is related to the fact that there are many types of fabric designs, such as tailored jackets, over-alls, trousers, skirts, etc. The accuracy of the equations for the objective evaluation of the quality of ladies’ tailored-type jacket fabrics derived from six blocks of the mechanical properties was lower for spring and summer ladies’ tailored-type jacket fabrics than for autumn and winter ladies’ tailored-type jacket fabrics (Inoue and Niwa, 2009). Therefore, we used a thermo labo to measure the initial maximum values qmax of heat properties as heat transport properties. To analyze the subjective evaluation of spring and summer tailored-type jacket fabrics, a regression analysis was conducted using stepwise block regression applied to the expert judges’ judgment value THV (Inoue and Niwa, 2009), using six blocks (tensile, bending, shearing, surface, compression, and construction) of the 19 mechanical properties and one block of the initial maximum values qmax of heat properties of the fabrics, as the seven blocks of 20 fabric properties, including the secondary term of each property. For this analysis, the effect of the initial maximum values qmax was added to the six blocks of mechanical properties for spring and summer tailored-type jacket fabrics for a total of seven blocks. The regression formula is as follows: THV ¼ C 0 þ
20 X i¼1
X i 2 M i1 X 2 2 M i2 C i1 þ C i2 i si1 si2
where: C0, Ci1, Ci2 ¼ constant coefficients of the ith variable terms. Xi
¼ mechanical property of the ith variable term.
Mi1, si1
¼ the population mean and standard deviation.
Mi2, si2
¼ the square mean and standard deviation.
!
IJCST 22,1
12
The results are shown in Table I. The effects of qmax were lowest with the seventh block, and the accuracy of the regression was at the same level as the accuracy of the regression using the six blocks of the mechanical properties. The correlation between the initial maximum value qmax of heat flux and the THV was also low, at 0.017, with multiple regression analysis using the seven blocks of fabric properties with THV. We found no effects of the initial maximum values qmax of heat flux on the subjective evaluation of the quality of spring and summer tailored-type jacket fabrics. To analyze the subjective evaluation of 70 types for spring and summer ladies’ knitted fabrics and 65 types for autumn and winter ladies’ knitted fabrics, a regression analysis was conducted for the knitted fabrics of each season, using stepwise block regression applied to the subjective evaluation values of the 25 “consuming public” judges’ judgment values, and using three blocks (the initial maximum value qmax of heat flux, air resistance, and weight per unit of fabric). The results are shown in Table II. Based on the accuracy of the regression of knitted fabrics, an effect of the initial maximum values qmax of heat flux on the subjective evaluation values is recognized. Importance order Xi 1
2
3
4
5
6
Tensile LT logEM1 logEM2 RT Compression LC logWC RC Shear logG log2HG log2HG5 Surface MIU logMMD logSMD Bending logB1 LogB2 log2HB1 log2HB2 Construction logT logW
R
RMS
Importance order Xi 1
0.499 0.930
2 0.618 0.846 3 0.666 0.805 4 0.705 0.765 5 0.731 0.738
6 0.741 0.725 7
Table I. Accuracy of the regression to THV
Tensile LT logEM1 logEM2 RT Compression LC logWC RC Shear logG Log2HG Log2HG5 Surface MIU logMMD logSMD Bending logB1 logB2 log2HB1 log2HB2 Construction logT logW Heat absorption logqmax
R
RMS
0.499 0.930
0.618 0.846
0.666 0.805
0.705 0.765
0.731 0.738
0.741 0.725 0.741 0.724
Notes: Suffix 1, wrap direction; suffix 2, weft direction; Sn, number of subjects; R, accuracy of the regression; RMS, root mean square of regression error; regression based on significant judgments (Sn ¼ 6) (Inoue and Niwa, 2009) for tailored-type fabrics for spring and summer
The accuracy of the regression of the initial maximum value qmax of heat flux, air resistance and weight per unit of fabric for spring and summer ladies’ knitted fabrics was 0.655 and these three properties related more closely to the subjective evaluation values of the consumers than did the same values for autumn and winter ladies’ knitted fabrics. This is thought to be due to the fact that knitted fabrics are worn in more direct contact with the skin than spring and summer tailored-type jacket fabrics. It can be clarified that the initial maximum values qmax of heat flux affected the subjective evaluation of the quality of ladies’ knitted fabrics. The contributions of these three properties to THV were investigated for each spring and summer fabric and each autumn and winter knitted fabric The contributions of each property to THV are shown in Figure 5. For spring and summer knitted fabrics, when the initial maximum value qmax of heat flux is high, THV is high, and when the weight per unit of fabric is low, THV is recognized as being high. For autumn and winter knitted fabrics, the contribution of air resistance was recognized. The tendency of THV to be low when air resistance is high is recognized. However, the whole accuracy is based on three variables, a small number, and for the subjective evaluation of the quality of knitted fabrics, relationships with factors other than the initial maximum value qmax of heat flux and air resistance are suggested. The judges of the quality of knitted fabrics were consumers in this research, but it is also necessary to further examine the judgments of textile experts.
Quality of ladies’ knitted fabrics
13
Conclusions . The initial maximum values qmax of heat flux are recognized as having a tendency to decrease, especially when autumn and winter ladies’ knitted fabrics form two layers rather than when there is only a single layer fabric. . The reproducibility of the warm/cool feeling of ladies’ knitted fabric is recognized in the summer and autumn evaluations of the sensory test. . The spring and summer ladies’ knitted fabrics provide the coolest feeling of the fabrics tested from the initial maximum values of heat flux of the spring and summer ladies’ knitted fabrics, the spring, and summer tailored-type jacket fabrics, and the autumn and winter ladies’ knitted fabrics. . The regression analysis was conducted using stepwise block regression applied to the expert judges’ judgment value THV, using six blocks of the mechanical properties and one block of the initial maximum values qmax of the heat flux of heat properties of the fabrics, as the seven blocks of fabric properties, including S/S knitted fabrics (n ¼ 70) Importance order Xi R
RMS
1 2 3
0.489 0.449 0.437
logqmax logW logAR
0.529 0.629 0.655
A/W knitted fabrics (n ¼ 65) Importance order Xi R 1 2 3
logAR logqmax logW
0.307 0.363 0.374
RMS 0.544 0.533 0.530
Notes: Sn, Number of subjects; N, number of samples; S/S, spring and summer; A/W, autumn and winter; W, weight; AR, air resistance; R, accuracy of the regression; RMS, root mean square of regression error; regression based on general consumer judgments (Sn ¼ 25) for knitted fabrics
Table II. Accuracy of regression to THV
IJCST 22,1
Spring and summer knitted fabrics
Contribution to THV
4
14
3 2
logqmax logW
1 0 logAR
–1 –2 –4
–2
2
0
4
Xi – Mi1 s i1 Autumn and winter knitted fabrics
Contribution to THV
4 3 2 logqmax 1 0 –1 –2 –4
Figure 5. The contribution of each property to the THV of knitted fabrics
logAR
logW –2
0
2
4
Xi – Mi1 s i1 Notes: Number of samples = 75 (spring and summer), 60 (autumn and winter); number of subjects = 25
.
.
the secondary term of each property. The results of the regression analysis show that the initial maximum values qmax of the heat flux of the heat properties of the fabrics do not affect the subjective evaluation of the quality of spring and summer tailored-type jacket fabrics. The results of the regression analysis of ladies’ knitted fabric properties for 70 types of spring and summer fabrics and 65 types of autumn and winter fabrics applied to the subjective evaluation value have confirmed that the initial maximum values qmax of heat flux affect the subjective evaluation of the quality of ladies’ knitted fabrics. For the subjective evaluation of the quality of knitted fabrics, relationships with factors other than the initial maximum value qmax of heat flux and air resistance are suggested.
References Imai, J., Yoneda, M. and Niwa, M. (1987), “Sensory tests for objective evaluation of fabric warm/cool touch”, Jpn. Res. Assn Text. End-Uses, Vol. 28, pp. 414-22.
Inoue, T. and Niwa, M. (2009), “Objective evaluation of the quality of fabrics for ladies’ tailored-type jackets for spring and summer”, J. Text. Eng., Vol. 55, pp. 1-11. Matsudaira, M., Kawabata, S. and Niwa, M. (1984), “Measurements of mechanical properties of thin dress fabrics for hand evaluation”, J. Text. Mach. Soc. Japan (predecessor Journal of J. Text. Eng.), Vol. 37, pp. T49-T57.
15
Appendix
Mechanical properties
Symbols Characteristic value
Unit
Measuring conditions High sensitivity (Matsudaira et al., 1984)
Tensile
EM
%
Strip biaxial deformation
LT
Bending Surface
WT RT B 2HB MIU MMD SMD
Shearing
G 2HG
Compression
Quality of ladies’ knitted fabrics
2HG5 LC WC
RC Thickness and T weight W
Tensile strain at maximum load Linearity Tensile energy Resilience Bending rigidity Hysteresis
– 2
Upper limit tensile force (maximum load): 50 gf/cm
gf cm/cm % gf cm2/cm gf cm/cm
Pure bending Maximum curvature, K ¼ ^2.5/cm Coefficient of friction – Contactor for friction measurement: ten parallel steel piano wires with 0.5 mm dia. and 5 mm length Mean deviation of – Simulating finger skin geometry. MIU Contact force: 50 gf Geometrical mm Contactor for geometrical roughness roughness: a steel piano wire, with 0.5 mm dia. and 5 mm length. Contact force: 10 gf Shear stiffness gf/cm degree Shear deformation under constant tension of 10 gf/cm Hysteresis at gf/cm Maximum shear angle, f ¼ ^88 f ¼ 0.58 Hysteresis at f ¼ 58 gf/cm Linearity – Upper limit pressure: 10 gf/cm2 2 Compressional gf cm/cm energy Resilience % Thickness at mm Thickness at 0.5 gf/cm2 pressure 0.5 gf/cm2 Weight of specimen per unit area Weight per unit area mg/cm2
Corresponding author Takako Inoue can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
Table AI. Characteristic values of basic mechanical properties and measuring conditions for KESF measurements
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
IJCST 22,1
Computerized pattern making focus on fitting to 3D human body shapes
16
Young Sook Cho Faculty of Home Economics, Tokyo Kasei University, Tokyo, Japan
Received 6 November 2008 Accepted 9 June 2009
Keiichi Tsuchiya Graduate School of Shinshu University, Nagano-ken, Japan
Masayuki Takatera and Shigeru Inui Faculty of Textile Science and Technology, Shinshu University, Nagano-ken, Japan
Hyejun Park Department of Clothing and Textiles, College of Human Ecology, Chungham National University, Daejon, South Korea, and
Yoshio Shimizu Faculty of Textile Science and Technology, Shinshu University, Nagano-ken, Japan Abstract Purpose – This paper aims to describe the development of a method of constructing three-dimensional (3D) human body shapes that include a degree of ease for purpose of computerized pattern making. Design/methodology/approach – The body shape could be made with ease allowance to an individual’s unique body shape using sweep method and a convex method. And then generates tight skirt patterns for the reconstructed virtual body shape using a computerized pattern making system. Findings – This paper obtains individual patterns using individually reconstructed 3D body shapes by computerized pattern development. In these patterns, complex curved lines such as waist lines and dart lines are created automatically using the developed method. The method is successfully used to make variations of a tight skirt to fit different size women. The author also used the method to make other skirts of various designs. Originality/value – The method described in this paper is useful for making patterns and then garments, without the need for the garments to be later adjusted for the subject. Keywords Modelling, Computer applications, Textile technology, Human anatomy Paper type Research paper International Journal of Clothing Science and Technology Vol. 22 No. 1, 2010 pp. 16-24 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011008776
This research was supported by the Ministry of Education, Science, Sports and Culture, Grant-in-Aid for twenty-first century COE Program and Scientific Research (C), 16500124, 2005.
Introduction Traditional pattern making systems use two-dimensional perimeter element information, which is not enough to correspond to individual variations in the shape of the human body. By using three-dimensional (3D) data measured by 3D scanning devices, it is possible to express individual unique characteristic of body shape for pattern making. There are number of current research efforts in the field of 3D body modeling and computerized pattern making systems using 3D data (Cho et al., 2005; Bigliani et al., 2000; Watanabe, 1999). However, almost all of them use scanned 3D dummy shapes, not 3D human body shapes (Heisey et al., 1990a, b; Kim and Park, 2007; Miyoshi and Hirokawa, 2001). When making patterns using traditional techniques, pattern makers used dummies (Miyoshi, 2002). These dummies are shaped to include a degree of ease. Ease allowance (looseness in certain areas) in pattern making allows for body movement and unrestricted fit. Too much ease can be sloppy and unattractive. When making individually personalized clothing patterns, it is difficult to map the pattern to actual 3D representations of the body, because ease allowance must be given. Therefore, pattern makers (dummy used) using traditional techniques need not allow any additional ease. However, it is necessary to confirm that they actually fit well on real people body shape. Adjustments to garments are often required. Since we are using scanned 3D real human body data as opposed to pattern makers’ dummies, our system must generate patterns that include a degree of ease. Our method first constructs, a virtual body shape, which includes ease, and then generates tight skirt patterns for the constructed virtual body shape using a computerized pattern making system. By using 3D real human body shape directly, it is not necessary to confirm actually fit well on real people body shape. This system has the potential to be much more efficient than traditional pattern making techniques. It also generates patterns which are much more suitable for each unique human body. Methods The lower body, waist, stomach, and hip shape express unique characteristics of an individuals’ shape. Thus, these factors need to be considered when reconstructing 3D body shapes with ease allowance during the pattern making process. In our research, we develop a method of reconstructing 3D individual body shapes which retain individuals’ unique characteristics (Figure 1). Method of reconstruction of real human body shape In the reconstruction process, it is necessary to divide the body in three parts, from waist to stomach, from stomach to hip, and from hip to hemline. The following steps describe our method: (1) First, we construct a line model using 3D scanned body data. (2) We extract waist line (WL) having maximum Z-value of back shape line in the side view of 3D body model. Stomach line (SL) is extracted by line having maximum Z-value in front shape line under waist. Hip line (HL) is extracted by line having minimum Z-value in back shape line under stomach. Figure 2 shows WL, SL, and HL on lower body so that it is possible to divide into three parts, from WL to SL (I), from SL to HL (II) and from HL to hemline (III).
3D human body shapes
17
IJCST 22,1
18 Figure 1. Individual shape of stomach and hips
Front shape line
Back shape line
Y WL I SL
II
HL
Figure 2. 3D body shape and line model extracted WL, SL, and HL
III Z
(3) Extracted SL is arranged and copied to HL at regular intervals using sweep method. As a result, there are existing two lines on the same Y coordinates in the II area (Figure 3). These two lines are connected using convex hull method for gaining the besieging lines at each position in II area. The convex hull of a set of points is the smallest convex set that includes the points. For a two-dimensional finite set the convex hull is a convex polygon. When creating the besieging lines, it is possible to give ease allowance retaining an individuals’ shape in areas such as stomach and HLs shapes (Figure 4). (4) Using sweep method, the besieging line is copied from HL to hem line as III area (Figure 5). (5) Uneven lines from waist to stomach back shape line on I area are smoothed using a convex method. As shown in Figure 6, we can make the body shape with ease allowance for making patterns unique to an individual’s unique body shape.
3D human body shapes SL
19
II HL
Figure 3. The result of II area using sweep method
HL
HL
SL
SL
Figure 4. The convex method for creating the besieging lines of SL and HL (a) SL and HL
(b) Connected SL and HL
II
(c) Besieging line
HL HL
III
Figure 5. The result of sweeped HL on III area
Experiments 3D measurement of human body shape We used scanned ten subjects’ body data to examine the effectiveness of our reconstruction method. Tables I and II show size information of the ten subjects based on JIS size indication. For example, number three subject had difficulty choosing skirts because an M size, which is suitable for her waist is too loose around her hips. Six other subjects in the Tables I and II had similar problems with size choice. We try to reconstruct their body shape for individual pattern making and make unique skirts for each subject. We then examine how well they fit.
IJCST 22,1
I
I
20 Figure 6. The results of smoothed waist and stomach back lines on I area
Table I. JIS size indication (cm)
Waist size Hip size
Subject number
Table II. JIS size information of subjects (cm)
1 2 3 4 5 6 7 8 9 10
S
M
L
LL
58-64 82-90
64-70 87-95
69-77 92-100
77-85 97-105
Waist size
Hip size
M L M LL M L LL L L L
S L S L S L L L L M
Individual pattern development process Our development method is presented simply here (Cho et al., 2006). First, we make the 3D reconstructed body surface of the clothes using triangular patches and sets grainlines for weft and warp on 3D body surface. We arrange 12 cross-sectional grainlines at 158 intervals to make 14 sections. After we set grainlines accurately for weft and warp, we fit the fabric lattice to the contour surface. We form a fabric lattice with a mesh structure in weft and warp direction. We cut 3D surfaces using plane. We create patterns by making angle of fabric lattices at right angles from three dimensionally contoured panels into two dimensions. Panels can then be described using curved lines in this process, 3D cutting line is flattened on the two-dimensional pattern. Finally, we can achieve the pattern (Figure 7).
WL
WL
HL
HL
Sewing line
3D human body shapes
21
Figure 7. Individual pattern development process
Results Results of reconstructed 3D human body shapes Figure 8 show the results of reconstructions for subjects 5 and 8. As shown, our reconstructed body shapes include ease allowance. These reconstructed body shapes, which include ease allowance are used as body shapes for pattern development. Results of individual pattern development We obtained individual patterns using individually reconstructed 3D body shapes by computerized pattern development. Figure 9 shows the completed pattern of four subjects with four dart lines. In individual patterns, complex curved lines such as waistlines and dart lines are created automatically using our developed method. Even though subjects 3 and 5 have same size as shown Table II, their shape, amounts and length of waist and dart lines are different depending on individual body shapes. These lines are one of individual characters in pattern making. Result of making skirts using personalized patterns We made tight skirts using patterns created using our method and then examined fitness on each individual subject’s body. Figure 10 shows tight skirts for four subjects made using our pattern making system. Subjects indicated that the fit of our skirt. It shows how our method works for making patterns for different body sizes and shapes.
Figure 8. The reconstructions for Subjects 5 and 8
IJCST 22,1
22
Figure 9. The results of individual pattern
Figure 10. The result of making skirts using unique created patterns
Notes: Subjects of 3, 5, 7, 8
Notes: Subjects of 3, 5, 7, 8
Application for various designs There are various designs based on a basic skirt pattern. In this research, we tried to apply our method to make various design skirts with patterns unique to a given subject. Two subjects used our method to create various design. Figure 11 shows the results for the two subjects.
3D human body shapes
23 Front view
Front view
Side view
Side view
Notes: Subjects of 9,10
Conclusions We developed a method of reconstructing 3D individual body shapes, which retain an individual’s unique characteristics. It is part of a computerized pattern making system, which we developed. We could make the body shape with ease allowance for making patterns unique to an individual’s unique body shape using sweep method and a convex method. We obtained individual patterns using individually reconstructed 3D body shapes by computerized pattern development. In our patterns, complex curved lines such as WL and dart lines are created automatically using the developed method. We successfully used our method to make variations of a tight skirt to fit different size women. We also used our method to make other skirts of various designs. Our method is useful for making patterns and then garments, without the need for the garments to be later adjusted for the subject. References Bigliani, R., Eischen, J.W., House, D.H. and Breen, D.E. (2000), “Collision detection in cloth modeling”, Cloth Modeling and Animation, A.K. Peters, Natick, MA, pp. 199-205. Cho, Y.S., Komatsu, T., Park, H.J., Inui, S., Takatera, M. and Shimizu, Y. (2006), “Individual pattern making using computerized draping method for clothing”, Textile Research Journal, Vol. 76 No. 8, pp. 646-54. Cho, Y.S., Okada, N., Park, H.J., Takatera, M., Inui, S. and Shimizu, Y. (2005), “An interactive body model for individual pattern making”, International Journal of Clothing Science & Technology, Vol. 17 No. 2, pp. 91-9. Heisey, F., Brown, P. and Johnson, R.F. (1990a), “Three-dimensional pattern drafting – part 1: projection”, Textile Research Journal, Vol. 60 No. 11, pp. 690-6. Heisey, F., Brown, P. and Johnson, R.F. (1990b), “Three-dimensional pattern drafting – part 2: garment modeling”, Textile Research Journal, Vol. 60 No. 12, pp. 731-7.
Figure 11. The results of application for various designs
IJCST 22,1
24
Kim, S. and Park, C.K. (2007), “Basic garment pattern generation using geometric modeling method”, International Journal of Clothing Science & Technology, Vol. 19 No. 1, pp. 7-17. Miyoshi, M. (2002), The Dress Making, Bunka Women’s University, Tokyo. Miyoshi, M. and Hirokawa, T. (2001), “Study on the method of measuring a vacant space distance in a worn jacket for clothing pattern design”, Journal of the Japan Research Association for Textile End-Uses, Vol. 42 No. 4, pp. 233-42. Watanabe, Y. (1999), “Ordering your cloth to fit yourself”, The Journal of the Institute of Electronics, Information and Communication Engineers, Vol. 82 No. 4, pp. 404-11. Corresponding author Masayuki Takatera can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Lipase treatment to improve hydrophilicity of polyester fabrics Hye Rim Kim and Wha Soon Song Department of Clothing and Textiles, Sookmyung Women’s University, Seoul, South Korea
Hydrophilicity of polyester fabrics 25 Received 17 February 2009 Accepted 12 June 2009
Abstract Purpose – The purpose of this paper is to investigate the conditions of the treatment using commercial lipase to improve the hydrophilicity of the polyethylene terephthalate (PET) fabrics. Design/methodology/approach – The lipase treatment conditions, such as the pH, temperature, treatment time, and concentration, are controlled by measuring the hydrolytic activity, moisture regain, and wettability of the treated fabrics. The effects of calcium ions on the moisture regain and wettability of the treated fabrics are also evaluated. Findings – The lipase treatment conditions for PET fabrics are controlled at a pH of 7.5, a temperature of 308C, a treatment time of 60 min, and a lipase concentration of 50 percent (owf). The moisture regain of the PET fabrics that are treated with lipase improved 3.3 times that of the untreated PET fabric. Calcium chloride did not affect the moisture regain of the treated fabrics but affected their wettability. The surface of the PET fabrics that are treated under optimum conditions and in the presence of calcium chloride showed many cracks and voids, unlike the surface of the untreated PET fabrics. Research limitations/implications – The lipase treatment did not affect the handle of the PET fabrics in the present paper because the weight loss is very small. Originality/value – In this paper, the control conditions for the improvement of the hydrophilicity of PET fabrics using the low-cost commercial lipase are determined. The results of the study could further the environment-friendly finishing of PET fabrics. Keywords Fabric production processes, Moisture measurement, Textile testing Paper type Research paper
1. Introduction Concerns regarding health, energy, and the environment drive the improvement of enzyme technology in the textile industry (Bielen and Li, 2002). Enzymatic processing has been developed for natural fibers in wide-ranging operations, from cleaning preparations to finishing (Cavaco-Paulo and Gu¨bitz, 2003; Kirk et al., 2002). In addition to natural fibers, enzymatic hydrolysis on synthetic fibers has been explored to enhance their hydrophilicity (Cavaco-Paulo and Gu¨bitz, 2003; Guebitz and Cavaco-Paulo, 2007), using lipases, polyesterases, and cutinases (Alisch-Mark et al., 2006; Chaya and Kitano, 1999; Chaudhary et al., 1998; Fischer-Colbrie et al., 2004; Guebitz and Hsieh and Cram, 1998; Kim and Song, 2006, 2008b; Vertommen et al., 2005; Walter et al., 1995; Yoon et al., 2002). Among these enzymes, lipases have been reported as hydrolyzing ester linkages in polyethylene terephthalate (PET), thus producing polar hydroxyl and carboxylic groups (Guebitz and Cavaco-Paulo, 2007; Hsieh and Cram, 1998; Kim and Song, 2006, 2008a; Yoon et al., 2002;). Lipases have been obtained from bacterial, fungal, and animal pancreases and are used as crude mixtures with
International Journal of Clothing Science and Technology Vol. 22 No. 1, 2010 pp. 25-34 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011008785
IJCST 22,1
26
other hydrolases or in purified form (Hassan et al., 2006). The industrial applications of lipases in synthetic fiber processing have been limited, however, due to their high-production cost and the limited number of lipases available in industrial amounts (Houde et al., 2004). The enzyme hydrolysis of PET has used novel lipases made for laboratory use (Alisch et al., 2004; Chaya and Kitano, 1999; Yoon et al., 2002) or high-cost commercial lipases (Kim and Song, 2006). Owing to these problems, PET hydrolysis using lipase has been limited to broaden its industrial application. The use of low-cost commercial lipases for PET hydrolysis is a possible way to further environmentally friendly finishing of PET fabrics. The majority of this research is directed towards improving the hydrophilicity of PET fabrics using the commercial lipase, Lipolase 100L from Novozymes, and relatively few studies have examined the use of commercial lipases to hydrolyze PET fabrics. Commercial lipases have low-hydrolytic activity on PET. This low activity can be improved, however, by adding activators during the processing. Lipases from bacteria and fungi are known to demonstrate improved activity in the presence of inorganic salts such as calcium ions (Jung, 2003). Several studies on the effects of calcium ions on lipase activity have been limited to substrates such as olive oil or trybutyrin (Decker, 1997; Sharma et al., 2001). Few studies have investigated the effect of calcium ions on PET fabrics during enzymatic processing (Kim and Song, 2006, 2008a). This study investigates the conditions of the treatment of PET fabrics with commercial lipase to improve the hydrophilicity of the PET fabrics. The hydrolytic activity of lipase is evaluated via the number of carboxylic groups, using the titration method. Each treatment condition, such as the pH, temperature, treatment time, and concentration, is controlled by measuring the hydrolytic activity, moisture regain, and wettability of the treated fabrics. The effects of calcium ions on the moisture regain and wettability of the treated fabrics are evaluated. 2. Experimental 2.1 Materials For the experiment, 100 percent PET fabric (test fabric from KS K 0905) was used (Table I). The PET fabric consisted of filament fibers and had plain weave structures. Aspergillus oryzae lipase abbreviated as AOL (EC 3.1.1.3, Lipolase 100L, Novozymes) was used without further purification. The activity of lipase is 100KLU/g at pH 7.0, 308C using tribytyrin as a substrate. Tris (hydroxymethyl) amino methane abbreviated as Tris (pKa 8.3 at 208C, Sigma Chemical Co.) was used as a buffer. Tris buffer solution was used as the basis for all applications. The pH of the buffer solution was adjusted with 1 M HCl (Duksan Pure Chemicals, Korea) and 0.1 M NaOH (Junsei Chemicals, Japan). Thymophthalein (TPH) from Aldrich Chemical Co. was used as an indicator. A total of 95 percent ethanol (Duksan Pure Chemicals, Korea) was used to inactivate the enzymes in the test solution during the titration. All the chemicals were used without further purification.
Fiber Table I. Characteristics of fabric
100 percent polyester
Yarn count (denier)
Fabric count (yarns/in.)
Fabric weight (g/m2)
Thickness (mm)
70
113 £ 95
70 ^ 5
0.094 ^ 0.02
2.2 Hydrolytic activity on PET fabric The hydrolytic activity of lipase was measured by the titration method (Kim and Song, 2008b; Sigmaaldrich.com, 2006; Japanese Standard JIS K 0601, 1995; Walter et al., 1995). Each fabric sample was cut into 4 £ 4 cm pieces that weighed approximately 0.1 g. Each sample was placed into a 20 ml vial bottle that contained 8 ml of the 50 mM Tris buffer. The PET fabrics were then treated with different temperature, pH, treatment time, and concentration. All treatments were performed at 150 rpm using a shaking water bath. After the treatments, the test solution was transferred to a 50 ml Erlenmeyer flask, and 20 ml ethanol was added to the test solution to inactivate its enzymes. Four drops of 0.9 percent TPH indicator was added to the test solution. The test solution was titrated with 0.1 M NaOH into a light blue color. Then, the volume of the 0.1 M NaOH used in the sample test was recorded. Each test was carried out ten times. The blank test performed in the same condition with the sample test, but enzymes were not included. Each test was carried out five times. The hydrolytic activity of lipase was calculated using the following equation (Kim and Song, 2008b; Sigmaaldrich.com, 2006; Japanese Standard JIS K 0601, 1995; Walter et al., 1995): Hydrolytic activity ðml=gÞ ¼
Vs 2 Vb PET
where Vs: volume of 0.1 M NaOH for the sample test, Vb: volume of 0.1 M NaOH for the blank test, and PET: weight of the PET sample (g). 2.3 Enzymatic treatment Each fabric sample was cut into specific dimensions and weighed approximately 1 g. Depending on pH, temperature, concentration, and treatment time, the PET fabrics were treated with lipase in Tris buffer solution, using a liquor ratio 80:1. All the lipase treatments were performed at 150 rpm using a shaking water bath (BS-21, Jeio Tech., Korea). The enzyme inactivation was performed at 808C for 10 min. The treated fabrics were thoroughly washed with water and dried at room temperature. Then, weight loss, moisture regain, wettability, and scanning electron microscope (SEM) micrographs of treated fabrics were measured. The weight loss was evaluated by the dry weight of the fabrics before and after the treatment. The moisture regain was evaluated according to ASTM D629-99. It was measured via the dry weight and the moisture conditioning weight of the treated PET fabrics. The samples were dried at 1058C for 1 h using a drying oven. After cooled, they were weighed in a desiccator for 30 min. The moisture conditioning was carried out at 208C with a 65 percent relative humidity for 24 h. The moisture regain of the PET fabrics was calculated using the following equation: Moisture regain ð%Þ ¼
Wm 2 Wd £ 100 Wd
where Wm: weight of the fabrics in a moisture equilibrium at 208C at 65 percent relative humidity and Wd: weight of the fabrics dried at 1058C for 1 h. The wettability of the treated fabrics was evaluated via water absorbency and water contact angle (WCA). The water absorbency of the PET fabrics was evaluated
Hydrophilicity of polyester fabrics 27
IJCST 22,1
according to the AATCC test method 79-1992. The WCA of the PET fabrics was measured using the contact angle measurement system (KRUSS DSA100, KRU¨SS, Inc., Germany). Each test was carried out ten times. Changes in the surface of the treated fabrics were analyzed with a SEM (JSM-5410, Japan) after the samples were plated with gold.
28
3. Results and discussion 3.1 Effects of the pH and temperature Figure 1 shows the hydrolytic activity of lipase on PET fabrics. PET fabrics were treated with 50 percent (owf) lipase for 60 min, using different pH levels from 7 to 8.5 and different temperatures from 25 to 608C. The highest value was achieved at a pH of 7.5 and a temperature of 308C. The hydrolytic activity of the treatment at 308C showed approximately a twofold increase over that of the treatment at 258C, when the pH was controlled from 7.0 to 7.5. When the pH was controlled from 8.0 to 8.5, the hydrolytic activity of lipase remained low because the lipase was inhibited and denaturalized above the critical pH. The hydrolytic activity showed the highest values at a temperature of 308C and a pH of 7.5. Since the lipase actively hydrolyzed the ester linkages in the PET fabrics in this critical condition, the number of carboxylic groups increased. Also, the decrease in the hydrolytic activity at temperatures above 308C and a pH of 7.5 was related to the sensitive nature of the biocatalyst to the temperature and the pH. Enzymatic treatment above the critical temperature and pH would denaturalize the enzymes (Cavaco-Paulo and Gu¨bitz, 2003; Chaudhary et al., 1998; Kim and Song, 2008b). Figure 2 shows the effects of temperature on the moisture regain and wettability of PET fabrics. PET fabrics were treated with 50 percent (owf) lipase for 60 min at different temperatures, from 25 to 508C. The pH level was adjusted to 7.5. The moisture regain showed the highest value (1.579 percent ^ 0.06) at 308C, and improved 3.3 times
Hydrolytic activity (ml/g)
2.0
7.0 (pH) 7.5 8.0 8.5
1.5
1.0
0.5
Figure 1. Effects of the pH and temperature on the hydrolytic activity of the lipase-treated PET fabrics
0.0
25
30
35
40 45 50 55 60 Temperature (°C) Notes: Treatment conditions: 50 percent (owf) lipase; treatment time, 60 min
that of the untreated PET fabric. The WCA and the water absorption time of the lipase-treated PET fabrics decreased when the temperature was increased to up to 308C. The WCA and the water absorption time of the treated fabrics at 308C decreased 1.42 and 2.03 times those of the untreated PET fabrics, respectively. The highest degree of wettability of the PET fabrics was thus achieved at a temperature of 308C. The lipase hydrolytic activity affected the improvement of the moisture regain and wettability of the treated PET fabrics. The number of carboxylic groups increased because of the lipase-hydrolyzed ester linkages in the PET fabrics under the controlled treatment conditions, which improved the fabrics’ moisture regain and wettability. The highest levels of moisture regain and wettability of the PET fabrics were achieved at 308C and a pH of 7.5.
Hydrophilicity of polyester fabrics 29
25
30 35 40 45 Temperature (°C)
105 100 95 90 85 80 75 70 65 0.0
50
110 Water absorbency (sec.)
1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0
WCA (degree)
Moisture regain (%)
3.2 Effects of the treatment time Figure 3 shows the effects of the treatment time on the hydrolytic activity. Over different time periods from 10 to 360 min, the fabrics were treated with 50 percent (owf)
25
30 35 40 45 Temperature (°C)
50
100 90 80 70 60 50 40 0.0
25
30 35 40 45 Temperature (°C)
Notes: Treatment conditions: 50 percent (owf) lipase; pH, 7.5; treatment time; 60 min; , lipase-treated PET
50
,untreated PET;
Figure 2. Effects of temperature on the moisture regain and wettability of the lipase-treated PET fabrics at varying temperatures
Hydrolytic activity (ml/g)
1.8 1.6 1.4 1.2 1.0 0.8
0.00
0
50
100 150 200 250 300 350 400 Treatment time (minutes)
Notes: Treatment conditions: 50 percent (owf) lipase; pH, 7.5; temperature, 30°C
Figure 3. Effects of the treatment time on the hydrolytic activity of the lipase-treated PET fabrics
IJCST 22,1
30
lipase at a pH of 7.5 and a temperature of 308C. After the 60 min treatment, the hydrolytic activity improved 1.5 times that after the 10 min treatment. Above 60 min, the hydrolytic activity gradually decreased because a longer treatment time could cause denaturalization of enzymes or aggregation between enzyme molecules (Cavaco-Paulo and Gu¨bitz, 2003). Figure 4 shows the effects of the treatment time on the moisture regain of the PET fabrics. The fabrics’ moisture regain and wettability improved when the time was increased to up to 60 min, and showed the highest values at 60 min. The moisture regains of the PET fabrics that were treated for 60 min improved 3.3 times those of the untreated PET fabrics. The WCA and the water absorption time decreased rapidly as the treatment time was increased to 60 min, at which time the lowest value emerged. From the measurement of the hydrolytic activity and the wettability, the treatment time was controlled at 60 min.
Figure 4. The moisture regain and wettability at various treatment times
0
30 60 90 120 150 180 Time (minutes)
110
105 100 95 90 85 80 75 70 65 0.0
Water absorbency (sec.)
1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0
WCA (degree)
Moisture regain (%)
3.3 Effects of the enzyme concentration Figure 5 shows the effects of the enzyme concentration on the hydrolytic activity. The PET fabrics were treated with lipase at different concentration levels that ranged from 10 to 100 percent (owf). At the 50 percent concentration, the hydrolytic activity improved almost 1.7 times that at the 30 percent concentration (owf). At concentrations above 50 percent (owf), however, the value of the hydrolytic activity stayed within the acceptable error range. Since the cleavage site of the PET fabrics was limited, the excess enzyme supplies could not help improve the hydrolytic activity of lipase (Cavaco-Paulo and Gu¨bitz, 2003; Kim and Song, 2008b). In addition, the excess enzyme supplies could have caused aggregation between the enzyme molecules. The enzymes would have been partly inactivated depending on the type of aggregation formed (Cavaco-Paulo and Gu¨bitz, 2003; Kim and Song, 2008b). Figure 6 shows the effects of the enzyme concentration on the moisture regain and wettability of the PET fabrics. Enzymatic treatment was carried out at a temperature of 308C and a pH of 7.5 for 60 min. The enzyme concentration was controlled from 10 to 100 percent (owf). The moisture regain of the treated fabrics improved when the lipase concentration was increased from 10 to 70 percent; and at lipase concentrations above 70 percent, the moisture regain decreased slightly. The WCA and the water absorption time decreased rapidly as the enzyme concentration increased to 50 percent (owf). The lowest values of the WCA and the
0
30 60 90 120 150 180 Treatment time (minutes)
100 90 80 70 60 50 40 0.0
0
30 60 90 120 150 180 Treatment time (minutes)
Notes: Treatment conditions: 50 percent (owf) lipase; pH, 7.5; temperature, 30°C; , lipase-treated PET
,untreated PET;
Hydrophilicity of polyester fabrics
2.5
Hydrolytic activity (ml/g)
2.0
31 1.5
1.0
0.00
Figure 5. Effects of the enzyme concentration on the hydrolytic activity of the lipase-treated PET fabrics
0
10 20 30 40 50 60 70 80 90 100 Concentration (%, owf) Notes: Treatment conditions: pH, 7.5; temperature, 30°C; treatment time, 60 min
2.0 110
1.8
WCA (degree)
Moisture regain (%)
1.4 1.2 1.0 0.8 0.6
Water absorbency (SEC.)
100
1.6
90 80 70
0.4
100 90 80 70 60 50 40
0.2 0.0
0.0 0 10 20 30 40 50 60 70 80 90 100
0 10 20 30 40 50 60 70 80 90 100
Concentration (%, owf)
Concentration (%, owf)
Notes: Treatment conditions: pH, 7.5; temperature, 30°C; treatment time, 60 min;
0.0
0 10 20 30 40 50 60 70 80 90 100 Concetration (%, owf)
, untreated PET;
, lipase-treated PET
water absorption time were obtained at a 50 percent lipase concentration. At above 50 percent lipase concentrations, the WCA and the water absorption time increased. The lipase concentration was controlled at 50 percent (owf) via the measurement of the hydrolytic activity, moisture regain, and wettability of the PET fabrics. The lipase treatment condition was controlled at a pH of 7.5, a temperature of 308C, a treatment time of 60 min, and a lipase concentration of 50 percent (owf). 3.4 Effects of calcium chloride Figure 7 shows the effects of the calcium chloride on the moisture regain and wettability of the PET fabrics. The PET fabrics were treated in the presence of different amounts of calcium chloride, from 1 to 50 mM. The moisture regain did not significantly differ when there was calcium chloride and when there was none. The WCA and the water absorption time, however, decreased considerably where there was calcium chloride. The WCA and the water absorption time
Figure 6. The moisture regain and wettability at various enzyme concentrations
IJCST 22,1
32
of the PET fabrics that were treated in the presence of 3 mM of calcium chloride decreased almost 1.2 and 1.35 times their levels, respectively, when there was no calcium chloride. The improved wettability of the PET fabrics was probably due to the increase in the number of voids and cracks on the surface of the PET fabrics in the presence of calcium chloride. The increase in the number of voids and cracks was confirmed as shown in Figure 8. Also, these surface changes caused the weight loss of the treated fabrics. The weight losses of the treated fabrics in the presence of calcium chloride and in the absence of it were 0.317 percent ^ 0.075 and 0.152 percent ^ 0.067, respectively. In the presence of calcium chloride, the weight loss of the treated fabrics improved because the lipase retained its activity in the presence of calcium ions (Sharma et al., 2001). In the authors’ previous study on porcine pancreas lipase (Kim and Song, 2008b), the wettability of the treated fabrics improved in the presence of 30 mM calcium chloride. Unlike the previous study, the small number of calcium ions in this study helped improve the hydrolytic activity of AOL because the calcium ions helped activate more lipases from bacteria and fungi than from animal pancreases (Jung, 2003). The weight loss of the treated fabrics with calcium ions was too small, however, to affect the handle of the fabrics. 3.5 SEM micrographs Figure 8 shows the SEM micrographs of the PET fabrics during their lipase treatment. The surface of the PET fabrics that were treated under optimum conditions and in the presence of calcium chloride had many cracks and voids, unlike the surface of the untreated PET fabrics. Even though the voids and cracks were largely responsible for the improved wettability of the treated fabrics, they did not affect the handle of the fabrics because the weight loss was too small (Kim and Song, 2006; Kim and Song, 2008b). 2.0
90
1.8
80
80
1.6 1.4
Water absorbency (sec.)
WCA (degree)
Figure 7. The moisture regain and wettability of the lipase-treated PET fabrics in the presence of calcium chloride
Moisture regain(%)
70
70
60
1.2 50
60 50 40 30 20 10
1.0 0.0
0.0
0
0 5 10 15 20 25 30 35 40 45 50
0 5 10 15 20 25 30 35 40 45 50
0 5 10 15 20 25 30 35 40 45 50
CaCl2 (mM)
CaCl2 (mM)
CaCl2 (mM)
Notes: Treatment conditions: 50 percent (owf) lipase; pH, 7.5; temperature, 30°C; treatment time, 60 min;
Figure 8. SEM micrographs of the lipase-treated PET fabrics
Untreated PET
a Lipase-treated PET (pH 7.5, 30, 50% (owf), 60min.)
b Lipase-treated PET in the presence of 3mM CaCl2
4. Conclusion This study investigated the conditions of commercial lipase treatment of PET fabrics to improve the fabrics’ hydrophilicity. Each treatment condition, such as the pH, temperature, treatment time, and concentration, was controlled by measuring the hydrolytic activity, moisture regain, and wettability of the treated fabrics. The effects of calcium ions on the moisture regain and wettability of the treated fabrics were evaluated. The lipase treatment conditions were controlled at a pH of 7.5, a temperature of 308C, a treatment time of 60 min, and a lipase concentration of 50 percent (owf). The moisture regain of the PET fabrics that were treated with lipase improved 3.3 times that of the untreated PET fabric. Calcium chloride did not affect the moisture regain but affected the wettability of the treated fabrics. The surface of the PET fabrics that were treated under optimum conditions and in the presence of calcium chloride showed many cracks and voids, unlike the surface of the untreated PET fabrics. Even though the voids and cracks were largely responsible for the improved wettability of the treated fabrics, they did not affect the handle because the weight loss was very small. This study determined the control conditions for the improvement of the hydrophilicity of PET fabrics using commercial lipase. Using a large amount of lipase could limit lipase application in PET processing, though. This problem can probably be solved if enhanced lipases for PET processing are developed or if activators and auxiliaries for the improvement of the hydrolytic activity of lipases are studied.
References Alisch, M., Feuerhack, A., Mu¨ller, H., Mensak, B., Andreaus, J. and Wimmermann, W. (2004), “Biocatalytic modification of polyethylene terephthalate fibers by esterases from actionomycete isolates”, Biocatalysis and Biotrasformation, Vol. 22 Nos 5/6, pp. 347-51. Alisch-Mark, M., Herrmann, A. and Zimmermann, W. (2006), “Increase of the hydrophilicity of polyethylene terephthalate fibers by hydrolases from Thermomonospora fusca and Fusarium solani f. sp. pisi”, Biotechnol. Lett, Vol. 28, pp. 681-5. Beilen, J.B.V. and Li, Z. (2002), “Enzyme technology: an overview”, Current Opinion in Biotechnology, Vol. 13 No. 4, pp. 338-44. Cavaco-Paulo, A. and Gu¨bitz, G.M. (2003), Textile Processing with Enzymes, The Textile Institute, New York, NY, pp. 96-191. Chaudhary, A.K., Beckman, E.J. and Russell, A.J. (1998), “Enzymes for polyester synthesis”, ACS Sym. Ser., Vol. 684, American Chemical Society, Washington, DC, pp. 18-57. Chaya, E. and Kitano, M. (1999), “Possibility of modifying polyester fibers using lipases”, Sen-I Gakkaishi, Vol. 55 No. 5, pp. 150-4. Decker, L.A. (1997), Worthington Enzyme Manual: Enzymes, Worthington Biochemical, Lakewood, NJ, pp. 112-29. Fischer-Colbrie, G., Heumann, S., Liebminger, S., Almansa, E., Cavaco-Paulo, A. and Guebitz, G.M. (2004), “New enzymes with potential for PET surface modicition”, Biocatal. Biotransfor., Vol. 22 Nos 5/6, pp. 341-6. Guebitz, G.M. and Cavaco-Paulo, A. (2007), “Enymes go big: surface hydrolysis and functionalisation of synthetic polymers”, Trends Biotechnol., Vol. 26 No. 1, pp. 32-8. Hasan, F., Shah, A.A. and Hameed, A. (2006), “Industrial applications of microbial lipases”, Enzyme Microb. Tech., Vol. 39 No. 2, pp. 235-51.
Hydrophilicity of polyester fabrics 33
IJCST 22,1
34
Houde, A., Kademi, A. and Leblanc, D. (2004), “Lipases and their industrial applications”, Appl. Biochem. Biotech., Vol. 118, pp. 155-70. Hsieh, Y.L. and Cram, L.A. (1998), “Enzymatic hydrolysis to improve wetting and absorbency of polyester fabrics”, Text. Res. J., Vol. 68 No. 5, pp. 311-19. Japanese Standard JIS K 0601 (1995), Determination of lipolytic activity of lipase for industrial use. Jung, D.H. (2003), Enzymology, Dae Kwang Seo Rim, Seoul, pp. 177-86. Kim, H.R. and Song, W.S. (2006), “Lipase treatment of polyester fabrics”, Fiber. Polym., Vol. 7 No. 4, pp. 339-43. Kim, H.R. and Song, W.S. (2008a), “Effects of Triton X-100 and calcium chloride on the porcine pancreas lipase treatment of PET fabrics”, J. Korean Society of Clothing and Textiles, Vol. 32 No. 6, pp. 911-17. Kim, H.R. and Song, W.S. (2008b), “Optimization of enzymatic treatment of polyester fabrics by lipase from porcine pancreas”, Fiber. Polym., Vol. 9 No. 4, pp. 423-30. Kirk, O., Borchert, T.V. and Fuglsang, C.C. (2002), “Industrial enzymes applications”, Curr. Opin. Biotech., Vol. 13 No. 4, pp. 345-51. Sharma, R., Chisti, Y. and Banerjee, U.C. (2001), “Production, purification, characterization, and applications of lipases”, Biotechnology Advances, Vol. 19, pp. 627-62. Sigmaaldrich.com (2006), “Sigma quality control test procedure”, available at: www.sigmaaldrich. com/sigma/enzyme%20assay/l3126enz.pdf (accessed 11 November 2006). Vertommen, M.A.M.E., Nierstrasz, V.A., van der Veer, M. and Warmoeskerken, M.M.C.G.J (2005), “Enzymatic surface modification of poly(ethylene terephthalate)”, J. Biotechnol., Vol. 120, pp. 376-86. Walter, T., Augusta, J., Muller, R.J., Widdecke, H. and Klein, J. (1995), “Enzymatic degradation of a model polyester by lipase from Rhizodus delemar”, Enzyme Microb. Tech., Vol. 17, pp. 218-24. Yoon, M.Y., Kellis, J. and Poulose, A.J. (2002), “Enzymatic modification of polyester”, AATCC Review, Vol. 2, pp. 33-6. Corresponding author Wha Soon Song can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Geometric disposition of threads in single-layer woven structures
Geometric disposition of threads
Elena Chepelyuk Department of Design, Kherson National Technical University, Kherson, Ukraine
Valeriy Choogin Department of Mechanical Technology of Fibre Materials, Kherson National Technical University, Kherson, Ukraine, and
35 Received 20 March 2009 Revised 1 June 2009 Accepted 1 June 2009
Jenny Cousens and Michael Hann The School of Design, University of Leeds, Leeds, UK Abstract Purpose – The purpose of this paper is to analyse the advantages of a new interpretation of the geometric disposition of threads within woven fabric structures, and to develop a method of determining the parameters of threads, with reference to each order of their disposition. Design/methodology/approach – Based on the analysis of the geometrical models proposed by Barker and Midgely, by Pierce and by Novikov, the substantiation of the advantages of a stricter model, offered by the authors, for determining the geometric disposition of threads within single layer woven fabric structures with the help of the tangent function is given. This model allows the substantial expansion of the actual bounds of the interval of the order of the geometric disposition of threads in woven fabric structures to 0.2-9.8. Findings – The tangent function can approximate the crimp height ratio of the warp threads within the woven fabric structure with accuracy within the limits of geometric disposition angle change from 18 to 898. Research limitations/implications – The work has applications in the industrial production of woven fabrics. Practical implications – This research will allow the design of a woven fabric with practically any ratio of crimp height for the warp and weft threads to effectively achieve the required performance characteristics of the cloth. Originality/value – This paper extends the knowledge of the geometrical characteristics of woven fabric structure, and proposes intelligent methods of determining the parameters of thread cross-sections in accordance with the orders of the geometric disposition of threads in woven fabric structure. Keywords Fabric testing, Thread, Modelling, Geometry Paper type Research paper
1. General introduction This paper presents theoretical developments of the geometrical models of woven fabric structure proposed previously by Barker and Midgley (1914), Peirce (1937) and Novikov (1946a, b, c). These established geometrical models conform, to varying degrees to the actual arrangement of threads in a woven fabric, on which the quality and speed of the development of new woven fabric structures to a large extent depends. 1.1 Literature considerations The study of woven fabric geometry was pioneered by Barker and Midgley (1914) and Peirce (1937), and many of the later investigations have been based on their models.
International Journal of Clothing Science and Technology Vol. 22 No. 1, 2010 pp. 35-48 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011008794
IJCST 22,1
36
Figure 1 shows the two geometrical models of ordinary woven fabric with a plain weave interlacing proposed by Barker and Midgley. Figure 1(a) shows a section of the woven fabric with the warp threads at an extreme position in relation to the central line of the cloth (A0 ), the weft thread bending without straight sections, equal diameters of warp and weft threads, and with identical crimp height hwp ¼ hwft ¼ a. In Figure 1(b), the warp threads are twice the diameter of the weft threads. The model is again of a plain weave, but woven as a weft rib. There is greater space between the warp ends and they do not deviate from the centre line (hwp ¼ 0). The thinner weft threads are bent significantly and have the maximum crimp height – hwft ¼ Max. These two extreme variants of thread crimp in a woven fabric are the basis for subsequent research by various authors. A simpler geometrical model, proposed by Peirce is shown in Figure 2. This geometrical model assumes that the threads have circular cross-sections and are Y
Z
Weft
Warp
B
B hwp/2
a
c
A′
A hwft/2 C
C′
b Y′
Z′ (a) Repeat Weft
Warp A
A
Figure 1. Models of Barker and Midgley
(b)
Weft
Warp D C θ1
d2 θ1
Figure 2. Geometrical model of Peirce
d1
A
θ1
h1/2
O h2/2
B P2
incompressible. The geometrical parameters are determined by the following conventional signs: †
ABCD
– the length of the warp thread in these circumstances consists of circular arcs (AB and CD) and straight lines (BC).
†
u1
– the inclination angle of warp in relation to the plane of the woven fabric for the two warp and weft systems.
†
d1 , d 2
– diameters of the warp and weft threads.
†
h1 , h 2
– maximum displacements of the warp and weft threads axes, normal to the plane of the woven fabric.
†
p2
– distance between the weft threads.
This model identifies a property of significant interest for this research. For a single-layer fabric, Peirce showed that the sum of the diameters of the warp and weft threads is equal to the sum of the displacement heights of the crimp in the structure of a single-layer woven fabric: d 1 þ d 2 ¼ h 1 þ h2
ð1Þ
The Peirce model is of limited practical use unless supported by computer calculation. Further research shows an updating of the geometrical model of the unit cell of plain weave by Peirce, in particular, the replacement of the circular cross-section of the warp and weft with an elliptical one (Peirce, 1937). In an attempt to reflect the actual behaviour of threads within woven fabrics many authors have offered other geometrical models which take into account the deformation of the threads within the structure (Zvorikina, 1946; Surnina, 1973). Based on the full range of thread arrangements in woven fabric, from the equal crimping of warp and weft threads to the absence of crimp in warp or weft, identified by Barker and Midgley (1914), and the development of the theory by Peirce (1937) and Novikov (1946a, b, c) developed a theory of woven fabric structure phases, which allows the specification of the range of probable variants of geometrical thread disposition within a woven fabric structure. The term “phase” is indicative of a particular form of thread disposition, between the two extremes of equal crimping and the absence of crimping in either warp or weft. The object of this paper is the substantiation of the advantages of the author’s interpretation of these phases of geometrical thread disposition within single-layered woven fabrics, based on the fundamental work by Novikov (1946a, b, c). The substantive postulates of Novikov’s theory can be considered as the initial position for this research. 2. Geometrical analysis of the disposition of threads within single-layered woven structures by Novikov According to the theory of geometric thread disposition by Novikov (1946a, b, c), the arrangement of threads, with a circular cross-section, in a single-layer fabric of plain weave can be identified by one of nine structure “phases” (Figure 3). The first phase relates to the extreme case where the warp remains straight within the woven fabric, and only the weft bends. At the ninth phase, the contrary is true: the weft remains straight, and the warp bends to the maximum degree. Between the first
Geometric disposition of threads 37
IJCST 22,1
dwp = dwft Weft
I
Warp hwp = 0
hwft = 4r
Warp
Weft
38
II hwft = 3.5
hwp = 0.5r
III hwp = 1r
hwft = 3r
hwp = 1.5r
hwft = 2.5
hwp = 2r
hwft = 2r
IV
V
VI hwp = 2.5r
hwft = 1.5
VII hwp = 3r
hwft = 1r
hwp = 3.5r
hwft = 0.5
VIII
Figure 3. Phases of geometrical disposition of threads in single-layer woven structures by Novikov
Warp hwp = 4r
Weft
IX
Weft
hwft = 0 Warp
and ninth phases the warp and weft both bend to varying degrees. The fifth phase corresponds to the model of Barker and Midgely, shown in Figure 1(a), where there is equal crimp height of warp hwp and weft hwft (designations are ours hereinafter). Novikov suggested determining the difference between each phase by a rate of half of the thread radius (0.5 r). An important assumption is made. In transition to the following phase the decrease in the crimp height of the warp thread hwp is equal to increase in the crimp height of the
weft thread hwft. For example, with equal diameters dwp ¼ dwft of warp and weft threads, the crimp height of the warp thread hwp increases by 0.5 r, and the crimp height of the weft thread hwft decreases by the same value. Thus, it is observed that dwp þ dwft ¼ hwp þ hwft ¼ 4r. As a basic characteristic of woven fabric structure Novikov suggested using the crimp height ratio of warp hwp and weft hwft. Surnina (1973) suggested that the crimp height ratio of the warp and weft threads be expressed by the coefficient K h ¼ hwp =hwft (designated from this point as K Nh ). Table I presents the value of the coefficient K Nh ¼ hwp =hwft for all nine phases identified by Novikov. It is necessary to note that for the first phase K Nh ¼ 0=8 ¼ 0; and for the ninth, K Nh ¼ 8=0 ¼ 1. At the first phase, the warp threads can theoretically occupy a position one above the other and similarly at the ninth with the weft threads. It is noted, however, that this is theoretical conjecture, and that it is practically impossible to obtain these positions. Another essential disadvantage is the integer designation of the phases. In practice, we are compelled to use imprecise expressions (designation), for example: “the given woven fabric has a structure between the third and fourth phase, closer to the third”. In his work Novikov also considered a variation of his phase model for woven structures with unequal thread diameters, using the example dwp ¼ 2dwft. Considering the phenomenon whereby threads deform or flatten, Novikov provided an opportunity to interpret the cross-section of threads as elliptical, as shown by the work of Zvorikina (1946). In order to eliminate the disadvantages, while preserving the valuable essence of Novikov’s theory, a regularity of change in the value of the crimp height ratios of threads in a woven fabric, in the process of transition from a considered phase to the subsequent, was established:
Geometric disposition of threads 39
K Nh ¼ 0:143ðN F 2 1Þ þ 0:0235ðN F 2 1ÞðN F 2 2Þ þ 0:005ðN F 2 1ÞðN F 2 2ÞðN F 2 3Þ þ 0:00108ðN F 2 1ÞðN F 2 2ÞðN F 2 3ÞðN F 2 4Þ þ 0:000425ðN F 2 1ÞðN F 2 2ÞðN F 2 3ÞðN F 2 4ÞðN F 2 5Þ þ 0:0001958ðN F 2 1ÞðN F 2 2ÞðN F 2 3ÞðN F 2 4ÞðN F 2 5ÞðN F 2 6Þ þ 0:0001972ðN F 2 1ÞðN F 2 2ÞðN F 2 3ÞðN F 2 4ÞðN F 2 5ÞðN F 2 6ÞðN F 2 7Þ
Thread disposition phase I II III IV V VI VII VIII IX
Crimp height of threads Warp hwp Weft hwft 0 0.5 r 1.0 r 1.5 r 2.0 r 2.5 r 3.0 r 3.5 r 4.0 r
4.0 r 3.5 r 3.0 r 2.5 r 2.0 r 1.5 r 1.0 r 0.5 r 0
ð2Þ
Crimp height ratio of threads K Nh ¼ hwp =hwft 0 0.143 0.333 0.6 1.0 1.666 3.0 7.0 1
Table I. Value of coefficient K Nh for the crimp height ratio of the threads in a woven fabric proposed by Novikov
IJCST 22,1
40
Analysis of the cumbersome formula (2) shows an essential limitation of its use. Geometrical approximation of function (2) is shown on Figure 4 in the form of dotted curve K Nh which, as an example, allows the search for a simpler law Kh ¼ f(NF). 3. Tangential law of distribution of the phases of geometric disposition of threads in single-layer woven structures In Figure 4 point M(Kh ¼ 1, NF ¼ 5), corresponding to the average (fifth) phase of geometric thread disposition in woven fabric structures on curve K Nh ¼ f ðN F Þ also coincides with curve K Nh on the chart of the tangent function. It is equal to the tangent 458, corresponding to the middle of the range of change of argument NF. On Figure 4, the continuous curve K tg h ¼ f ðN F Þ shows the law of change Kh according to function tg aF in the interval of actual values of thread disposition angle aF ¼ 18-898. This interval was not casually chosen. An attempt to keep the existing representation about phases of mutual bend of threads in the woven fabric to a maximum led to the following initial position of searches of the range of argument variation. It is known, from Novikov, that at the fifth phase of geometrical thread disposition, the warp and weft threads have identical crimp height values: hwp ¼ hwft, corresponding to coefficient Kh ¼ 1.0. The tangent of angle aF ¼ 458 is also equal to 1.0. Therefore, on Figure 4 point M with coordinates (5; 1) and (458, 1) is common for curves K Nh and K tg h . With the exception of the values unsuitable for practical use tg aF ¼ 0 at aF ¼ 08 and tg aF ¼ 1 at aF ¼ 908, order NF(1) of the first phase of thread disposition (corresponding to the first phase by Novikov) is approximated by thread 11
Ratio between values of warp and weft crimp Kh = hwp/hwft
10 9
Novikov KhN
8 7 6
The tangent function Khtg
5
∆ϕM
4 3 ϕN (M) ϕtg (M)
2 M
1
Figure 4. Comparison of the phases of thread disposition by Novikov and the tangent function
0 0.2
1
2
1°
5°
15° 25° 35° 45° 55° 65° 75° 85° 89°
3
4
5
6
7
8
9
9.8
Phase number NF Phase angle
αF
disposition angle aF1 ¼ 58, and the order of the ninth phase – aF9 ¼ 858. The interval of change between each subsequent order NF(iþ 1) from previous NF(i ) corresponds to a change of thread disposition angle aFi of 108. Thus, the range of possible variations in the crimp height ratios of threads in woven structures, under the tangent law K tg h ¼ hwp =hwft in limits NF ¼ 1-9, changes for the tangent function from 0.087 (by Novikov Kh(1) ¼ 0.0) to 11.430 (by Novikov Kh(9) ¼ 1). Of practical interest is the increased possibility of using the tangent law for a wider range of variation of argument – thread disposition angle aFi: from 18 to 898 corresponding to final numerical values of coefficient of orders K tg h ¼ tg aF ¼ 0:0175 and 57:29. Thus, using the tangent function essentially enables us to increase the quantity of thread disposition phases (from NF ¼ 0.2 up to NF ¼ 9.8 inclusive). It allows the calculation of practically any set arrangement of threads in a single-layer fabric for implementation on a weaving machine. Here, it is appropriate to pay attention to the necessity of coordinating the thread disposition order with the physical parameters of threads and structures of the woven fabric, in order to exclude the possibility of designing unworkable structures or an unreasonable overestimation of woven fabric density. Table II shows numerical values for the coefficient of thread disposition order K tg h under the tangent law in the specified limits of variation aF, and also K Nh by Novikov, for convenience of comparison (Chepelyuk and Choogin, 2007). With the purpose of essentially increasing the information value, the following universal designation of the coefficient of the crimp ratio Kh of the given thread disposition order NF of the structure of single-layer fabrics is offered for practical use: 3;2 3;2 K NhaFF . For example, the notation K h27 ¼ tg a27 should be read as “Kh is calculated for a tangent of thread disposition angle aF ¼ 278 and corresponds to the thread disposition order NF ¼ 3.2”. The fractional part 0.2 is determined in view of the interval between the integer values of the orders of 108 from the ratio: (278 2 258)/108. Designations for the thread disposition orders for the above-mentioned values aF ¼ 18 and 898 will 0:2 9:8 accordingly have the appearance: K h1 and K h89 . Here, fractional parts of the orders are determined as follows:
Thread disposition orders of woven fabric structure, NF 0 0.2 1 2 3 4 5 6 7 8 9 9.8 10
Geometric disposition of threads 41
Coefficients of crimp height ratios of threads in a woven fabric Thread disposition Under N.G. Novikov Using the tangent N angle, grad aF K h ¼ hwp =hwft function K tg h ¼ tg aF 0 1 5 15 25 35 45 55 65 75 85 89 90
– – 0:8 ¼ 0.000 1:7 ¼ 0.143 1:3 ¼ 0.333 3:5 ¼ 0.600 1:1 ¼ 1.0 5:3 ¼ 1.666 3:1 ¼ 3.00 7:1 ¼ 7.000 8:0 ¼ 1 – –
0 0.0175 0.087489 0.267949 0.466308 0.700207 1.0 1.428148 2.144507 3.732051 11.430052 57,29 1
Table II. Conformity of coefficients of the crimp height ratio of threads in a woven fabric to the thread disposition orders under Novikov’s theory and with use of the tangent function
IJCST 22,1
42
½ð18 2 08Þ ¼ 18 ¼ 0:2; 58
½ð898 2 858Þ ¼ 48 ¼ 0:8 58
4. Thread parameters in single-layer woven structures The theory offered for the geometric disposition of threads in two-dimensional woven fabric structures, based upon the tangential law of thread disposition order variation, gives the designer the possibility of predicting more precisely the properties of an elaborate woven fabric, ensuring fitness for purpose. 4.1 Parameters of the crimped thread During weaving, threads are in a stressed state and undergo deformations of stretching, bending and compression. It results in a condensing of the elementary fibres, and a reduction in the primary area of thread cross-section. At the fell of the cloth, on the weaving machine, the effects of condensing and crimping are caused by the physical properties of weft and warp threads to resist the bending and stretching, and also by the action of the driven elements of the weaving machine. The cross-section of the threads, due to mutual pressure, cannot remain in their initial form of a circle (a special case of the ellipse). The cross-section of threads is naturally transformed into the ellipse. The major (awft for weft and awp for warp) and the minor (bwft for weft and bwp for a warp) semi-axis of ellipses have to be determined with consideration for coefficients of crimp hawft , hbwft for weft and hawp , hbwp for warp:
hawp hbwp hawft hb ; awp ¼ d wp · ; bwft ¼ dwft · wft ; bwp ¼ dwp · ; ð3Þ 2 2 2 2 where dwft and dwp – initial conditional diameters of weft and the warp threads before their crimp, in mm. awft ¼ dwft ·
4.2 Transformation of thread parameters when changing the thread disposition order of single-layer woven fabric structures With the purpose of minimising the tensely-deformed conditions of threads by producing a defined thread disposition order of woven fabric structure during weaving, and the essential simplification of process of forming a woven fabric structure upon the release of external loading, an obvious model of weft arrangements, with changeable parameters in a unit cell of the woven fabric, with a limiting density at all 11 phases: from 0.2 up to 9.8 (Figure 5) is offered below. Using only the physical properties of the warp and weft threads, without special measures, it is possible to obtain any desired geometric thread disposition on a weaving machine. For this purpose, it is necessary to identify certain ratios of conditional diameters of weft and warp, with reference to the given thread disposition of the structure. For example, if the average fifth thread disposition order is obtained simply enough with an equality of conditional diameters of weft and warp threads, then for achievement of the extreme thread disposition orders 0.2 and 9.8 it is necessary to precisely understand the following two key rules: (1) For reducing the thread disposition order from the fifth down to 0.2 it is necessary to considerably reduce the cross-section of the weft threads
Warp (0.2)
hwft (5)
Weft (1)
Weft (5)
Lwft (0.2)
Area of crumple
hwft (0.2)
Weft (9)
Ο1
Ο2 (0.2)
βwft (1)
Lwft (5)
Ο2 (1)
Transformation of warp
Weft (0.2)
Weft (9.8)
Ο2 (2)
Ο2 (3)
βwft (5) Ο2 (4)
βwft (9)
Ο2 (5)
Ο2(6)
Weft (5)
Ο2 (7)
Ο2 (8)
Ο2(9)
Ο2 (9.8)
Transformation of weft
hwft (9.8)
Weft (9.8)
Warp (9.8)
Geometric disposition of threads 43
Figure 5. The transformation of threads under alteration of the thread disposition order number
IJCST 22,1
44
whilst simultaneously increasing the corresponding cross-section of the warp threads. (2) For increasing the thread disposition order from the fifth up to 9.8 it is necessary to considerably increase the cross-section of weft threads whilst simultaneously reducing the cross-section of the warp threads. As a basis for the search for and analysis and choice of the most effective woven fabric structure it is recommended that the designer carry out calculations of the limiting density of thread arrangement at the first cycle of the development of the woven fabric structure. For further calculations we shall make use of the property of the single-layer fabric, proposed by Peirce (1937) that the sum of crimp heights of the warp and the weft is equal to the sum of their diameters ðhwp þ hwft Þ ¼ ðd wft þ dwp Þ. Accepting the replacement of round thread cross-section with the elliptical, we can write the equality for the fifth thread disposition order as: ðhwp þ hwft Þ ¼ ð2bwft þ 2bwp Þ ¼ T wf ;
ð4Þ
followed by the important property: to keep the set thickness of the woven fabric Twf it is necessary to reduce the thickness of the warp threads 2bwp accordingly when increasing the thickness of the weft threads 2bwft (and vice versa). Further, it is necessary to pay attention to a significant feature: the value of Khi thread disposition order is expressed in relative units. It allows the acceptance of the crimp height of the weft for one, i.e. hwft ¼ 1, whilst the crimp height of the warp is considered as a part of hwft: hwp ¼ hwft £ K h
or
hwp ¼ T wf 2 hwft
ð5Þ
Then taking into account of crimp of threads it is possible to write: hwft þ hwp ¼ 2bwft þ 2bwp ¼ ð1 þ K h Þhwft
and
hwft ¼
2bwft þ 2bwp T wf ¼ ð6Þ 1 þ Kh 1 þ Kh
Then you should serially take the values of the thread disposition order coefficient under the tangent law K tg h , further – Kh(i ), and carry out calculations of the weft crimp height in the woven fabric hwp(i ) under the formula (6) and of the warp crimp height hwp(i ) under the formula (5) for every (i ) order. For further calculations, it is necessary to know the values of the semi-axes of the elliptical cross-sections of the weft and warp threads awft(i ) and bwft(i ) for each thread disposition order of the woven fabric structure. It is known that, by researching the 1-4th thread disposition orders, the thickness of the woven fabric is determined by means of the equation: T h ¼ hwft þ 2bwft ;
ð7Þ
and by researching the 6-9th thread disposition orders the thickness of the woven fabric is created by the sum: T h ¼ hwp þ 2bwp
ð8Þ
Upon the calculation of the value of the weft crimp height for each order hwft(i ) with the equation (6) and of the warp crimp heights hwp(i ) with equation (5) it is possible to
proceed to the calculation of the thickness of the threads in the direction of small semi-axis of the ellipse: For orders 0:2 2 4:0 : 2bwftði Þ ¼ T h 2 hwftði Þ and
Geometric disposition of threads
2bwpði Þ ¼ ð1 þ K hði Þ Þhwft 2 2bwftði Þ ð9Þ For orders 5:0 2 9:8 : 2bwpði Þ ¼ T h 2 hwpði Þ and 2bwftði Þ ¼ ð1 þ K hði Þ Þhwpði Þ 2 2bwpði Þ Further, it is necessary to determine the size of threads along the major semi-axis of the ellipse for each thread disposition order. It is known, that: awft hawft ¼ bwft hbwft
and
awp hawp ¼ ; bwp hbwp
ð10Þ
whence: awftði Þ ¼
bwftði Þ £ hawftði Þ
hbwftði Þ
and
awpði Þ ¼
bwpði Þ £ hawpði Þ
hbwpði Þ
ð11Þ
Table III demonstrates the effect a change in the geometrical parameters of warp and weft has on the various elements of geometrical thread disposition within in the woven fabric structure: the thread disposition order, the angle of thread disposition and the coefficient K tg h using the tangent law. The examples assume the use of warp and weft threads of equal linear density and the same fibre composition and construction for the middle equation (5) order of thread disposition. A designer can clearly and quickly identify the parameters of warp and weft threads, and therefore the ideal order of geometrical thread disposition in the woven fabric structure to achieve the required aesthetic properties. Here, it is necessary to emphasize a very important circumstance: any single variant of geometrical parameter combinations of warp and weft threads guarantee the minimum of process intensity in forming the fabric on the weaving machine. 4.3 Practical experimental research of two-dimensional woven fabric structures Figure 6 shows examples of variations in crimp of the warp thread. The warp was R29.2/2 tex yarn in a single-layer woven structure of plain weave interlacing with different weft yarn thicknesses: . R7.5/2 tex; . R29.2/2 tex; and . R50/2 tex. The warp yarn is coloured blue, with the weft threads woven in white or cream. To photograph the cross-section, the fabric is cut along the length of a warp. Figure 6(a) shows the finest thickness of weft (R7.5/2 tex). The warp threads show minimal bending. The fabric exhibits the thread disposition order < 1. Figure 6(b) shows equal
45
18 28080 58 158 258 358 458 558 658 758 858 878540 898
0.2 0.4266 1 2 3 4 5 6 7 8 9 9.58 9.8
0.017455 0.037245 0.087489 0.267949 0.466308 0.700207 1.0 1.428148 2.144507 3.732051 11.430052 27.271486 57.289962
K tg h
Crimp height ratio of warp and weft threads
Notes: Centi-milli-metre (cmm) ¼ 0.01 mm ¼ 102 5 m
aGDT grade
NGDT
Table III. Effect of warp and weft parameters on the variations in geometrical disposition of threads in woven fabric structures
Thread disposition angle
16.2 18.46 24.59 47.52 71.71 97.15 124.51 155.26 191.79 237.88 301.29 324.79 334.6
Weft Twft tex
Parameters of elliptical cross-section
334.6 324.79 301.29 237.88 191.79 155.26 124.51 97.15 71.71 47.52 24.59 18.46 16.2
70.76 69.42 66.2 56.78 49.1 42.35 36.0 29.65 22.9 15.22 5.79 2.55 1.24
1.24 2.58 5.79 15.22 22.9 29.65 36.0 42.35 49.1 56.78 66.2 69.45 70.76
6.49 6.93 8.0 11.12 13.66 15.9 18.0 20.1 22.34 24.88 28.0 29.07 29.51
29.51 29.07 28.0 24.88 22.34 20.1 18.0 15.9 13.66 11.12 8.0 6.93 6.49
9.74 10.4 12.0 16.68 20.49 23.85 27.0 30.15 33.51 37.32 42.0 43.61 44.26
44.26 43.6 42.0 37.32 33.51 30.15 27.0 23.85 20.49 16.68 12.0 10.39 9.74
Warp Weft Warp Weft Warp Weft Warp Twp hwft hwp bwft bwp awft awp tex (0.01 mm) (0.01 mm) (0.01 mm) (0.01 mm) (0.01 mm) (0.01 mm)
Crimp height
46
Threads disposition order
Number of filling threads per metre
IJCST 22,1
Geometric disposition of threads 47 Weft
Warp
(a)
Warp
Weft
(b)
Warp
Weft
(c)
bending of warp and weft threads, with the use of yarn of equal thickness (R29.2/2 tex in warp and weft). The fabric exhibits the thread disposition order < 5. Figure 6(c) shows the greatest thickness of weft (R50/2 tex), therefore the warp thread shows significant bending. The fabric exhibits order < 9.
Figure 6. Variation in the disposition of warp threads (R29.2/2 tex) in a plain weave woven structure interlacing with weft of (a) R7.5/2 tex; (b) R29.2/2 tex; (c) R50/2 tex
IJCST 22,1
48
The designer has individual control of the thread disposition within the woven fabric structure using this technology of experimental fabric specimen research. 5. Conclusion The approximation of the possible variations in the ratios of crimp height of warp and weft in woven fabric structure hwp/hwft with the precise tangent law essentially expands the limits of the actual use of the crimp height ratio coefficient of threads in the woven fabric structure K tg hi ¼ tg aF from 0.2 up to 9.8 (instead of 2-8 by Novikov (1946a, b, c)), and raises the efficiency and speed of calculations. Using the basic property of a single-layer fabric proposed by Peirce (1937), about the equality of the sum of diameters of warp and weft threads to the sum of their crimp heights, it is proposed that it is necessary to take into account an important property of the woven fabric: to keep the set thickness of the woven fabric it is necessary to reduce the thickness of the warp threads accordingly when increasing the thickness of the weft threads (and vice versa). For the practical calculation of the parameters of the elliptical section of threads in the warp and the weft, the use of the method of ratio of thickness of weft and warp threads with reference to each order of geometrical disposition of threads in the woven fabric structure is proposed. It will allow, without applying special measures and without excessive stress (overstrain), using only the physical properties of the warp and weft threads, the minimisation of the tensely deformed thread conditions and the production of practically any desired woven fabric structure (or phase) on the weaving machine. References Barker, A.F. and Midgley, E. (1914), Analysis of Woven Fabric, Scott, Greenwood & Son, London, pp. 72-9. Chepelyuk, E.V. and Choogin, V.V. (2007), “Tangent law of the distribution of phases order of one-layer woven fabric structures”, The Bulletin of Kiev National Univ. of Technol. and Design, No. 6, pp. 111-7 (in Russian). Novikov, N.G. (1946a), “On the woven fabric structure and its designing applying the geometrical method”, J. Technol. of Text. Ind., No. 2, pp. 9-17 (in Russian). Novikov, N.G. (1946b), “On the woven fabric structure and its designing applying the geometrical method”, J. Technol. of Text. Ind., No. 6, pp. 24-8 (in Russian). Novikov, N.G. (1946c), “On the woven fabric structure and its designing applying the geometrical method”, J. Technol. of Text. Ind., Nos 11/12, pp. 17-25 (in Russian). Peirce, F.T. (1937), “Geometry of cloth structure”, J. Text. Inst., Vol. 28, pp. T45-T96. Surnina, H.F. (1973), The Design of Woven Fabric on Set Parameters, Light Industry, Moscow, pp. 27-48 (in Russian). Zvorikina, E.K. (1946), “Investigation of the phenomenon of weft contraction on weaving”, dissertation, Textile Industry, Moscow (in Russian).
To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
A new methodology for the development of sizing systems for the mass customization of garments Maria L. Mpampa, Philip N. Azariadis and Nickolas S. Sapidis Department of Product and Systems Design Engineering, University of the Aegean, Ermoupolis, Greece
Mass customization of garments 49 Received 27 June 2008 Revised 12 March 2009 Accepted 12 March 2009
Abstract Purpose – The purpose of this paper is to derive a new method for developing sizing systems for the mass customization of garments. Design/methodology/approach – A range of recently published works has been studied. A new method is derived by following a basic statistical analysis on anthropometric data which are supported by an iterative mass customization model and introduced “satisfaction performance” indices. The derived method is applied successfully to an anthropometric data consisting of 12,810 Greek men. Findings – With the proposed method, it is possible to control the degree of mass customization and the corresponding number of garment sizes. Under this way, a balance between the number of sizes (in other words: production cost) and the percentage satisfaction of consumers can be achieved. The proposed method consists of six subsequent tasks which are applied to the target population data for the development of mass customization models for male shirts, coats and trousers. Research limitations/implications – Future work could be focused on the development of methods for the automatic garments grading with respect to the proposed mass customization models and practise. Originality/value – The methodology presented in this paper can be applied to the development of mass customization models for other categories of garments and target population. Keywords Garment industry, Human anatomy, Mass customization Paper type Research paper
1. Introduction Garments are manufactured massively using predefined size charts which allow for the reduction of production cost. It is, therefore, practically impossible to obtain a perfect fit between a piece of cloth and an individual buyer (Kotha, 1995; Pine, 1993). Owing to their low cost, preˆt-a-porter garments are dominating the modern markets, while partial individualization is achieved using sizing systems with normalized dimensions. Under this way, absolute individualization is sacrificed to the benefit of production economy (Fralix, 2000; Walter, 2006). The concept of mass customization is devised to serve the individualized needs of consumers and increase their satisfaction percentage (Anderson et al., 1999). This term implies a strategy for producing customized garments with maximum differentiation through a low-cost production process (Davis, 1987). Nowadays, this manufacturing model is enabled by modern information technologies, computer-aided design and manufacture systems, three-dimensional (3D) body scanners, interactive web-based
International Journal of Clothing Science and Technology Vol. 22 No. 1, 2010 pp. 49-68 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011008802
IJCST 22,1
50
applications, etc. (Ashdown and Delong, 1995; Salusso-Deonier, 1989; Gazzuolo et al., 1992). From the manufacturers’ point of view, there is an apparent trend on producing individualized garments with the greatest possible differentiation without affecting the production cost (DeLong et al., 1993). However, the relation between the size charts and body dimensions is not constant because of the changes that occur in the human population. Recent body surveys in Germany proved that a garment sizing system for a certain body type does not cover more than the 25 per cent of the population in which it is addressed (Lanenegger and van Osch, 2002; Walter, 2002). Hence, body measurements should be updated regularly in order to provide current information on essential sizes and their geographical distribution in the population (Istook et al., 2003). Consequently, for a successful garment mass customization model, the development and maintenance of up-to-date anthropometric databases of the target market population is essential (Salusso-Deonier, 1982; LaBat and Delong, 1990; Workman, 1991; Goldsberry et al., 1996; Ashdown, 1998). Such an approach necessitates the existence of a proper methodology for producing sizing systems (which are not proportionally graded) with respect to a target population and the corresponding garments type(s). These sizing systems should satisfy the majority of the target population and at the same time should imply a cost-effective and affordable production process by the garment manufacturer (Fralix, 2000; Walter, 2006). In the present work, a new methodology for the development of effective sizing systems is proposed, aiming at the introduction of mass customization in the manufacturing process of a garment company. With the proposed methodology one is able to derive a sizing system appropriate for a particular group of people (i.e. children, elderly, solders, etc.). The effectiveness of the proposed methodology is illustrated by applying it to the development of size charts for a target population consisting of 12,180 Greek men between 20 and 30 years old. With the obtained charts, we were able to produce garments that satisfy up to 92.4 per cent of the population. In the next section, we introduce the main methodology for producing sizing systems with respect to a specific target population and garment type, and with a variable degree of mass customization. Sections 3 and 4 present an application of the new methodology for producing certain garments for a Greek population consisting of 12,810 men. Finally, Section 5 concludes the paper giving also some remarks for future work. 2. The proposed methodology for the development of sizing systems 2.1 Review of existing works and main definitions One early empirical sizing system, the CS 215-58 Standard, was developed in the USA in 1958 based on the manufacturers’ experience. In 1970, another empirical sizing system, the PS 42-70 Standard of the USA, was developed using military anthropometrical data and a “trial-and-error” approach. Both sizing systems were based on out-of-date measurements taken at 1941 (Ashdown, 1998). Later, the development of new anthropometric databases demanded the classification of human bodies under various body types. This classification was achieved by Salusso-Deonier (1982) through a method called as “Principal component sizing system (PCSS)”. Soon after, Tryfos (1986) proposed another method based on the “integer programming” approach in order to minimize the number of the different sizes in a size chart. This method attempts to optimise an objective function (or indicator) named as the
“aggregate loss of fit” which measures the difference between real body dimensions and the produced size charts. In 1994, the American Society for Testing and Materials (ASTM) developed the ASTM D5585-94 Standard of the USA utilizing the experience of garment designers and market information. This sizing system was validated using US army and navy anthropometrical data. Ashdown (1998) and McCulloch et al. (1998) focused on the development of sizing systems more satisfactory than the ASTM D5585-94 using the “aggregate loss of fit” indicator in combination with the PCSS method. Gupta and Gangadhar (2004) proposed a “statistical” method for the minimisation of the “aggregate loss of fit” indicator with respect to the population’s body-type distribution. These models, however, were not developed with respect to the mass-customisation concept but to support the mass production of garments. Loker et al. (2005) describe a variety of size-specific statistical and visual analysis methods than can be applied to market body scan data to improve the apparel fit of a sizing system. They apply their method to modify an existing sizing system of a garments company. Finally, Hsu and Wang (2005) use a decision tree-based data-mining approach to establish a sizing system for the manufacture of garments, which allows for a wider coverage of body shapes with a fewer number of sizes and generates regular sizing patterns and rules. Their method is specialised for men’s pants. In this paper, we adopt two definitions originated in Lanenegger and van Osch (2002) for the characterization of garment dimensions. These definitions will be referred later by using the terms constraints A and B: . Definition 1 (constraint A). A dimension is called as primary when it plays an essential role in assessing when a garment is wearable by a consumer or not. In other worlds, a primary dimension tells when a piece of garment is able to cover the entire body of a consumer or not. . Definition 2 (constraint B). A dimension is called as secondary when it plays an essential role in assessing a garment’s fitting. In other words, a secondary dimension measures how a piece of garment (that fulfils constraint A) fits in a consumer’s body. Both constraints A and B are mentioned in the present paper as manufacturing constraints and vary with respect to the garment’s type (i.e. upper/lower body clothing, etc.). The purpose of the introduced method is to provide a feasible mass customization model for a particular target group and a garment type, which will be also called as related garment. The chest girth, waist girth and height are commonly used for the manufacture of the upper body garments. The neck girth and sleeve/arm length are commonly used for the manufacture of shirts. The waist girth and inside leg length are commonly used for the manufacture of lower body garments. In the methodology proposed in this paper, a set of body measurements (linear dimensions of height, inside leg length, sleeve/arm length, and girth dimensions of chest, neck and waist) of a target population is analyzed producing primary and secondary sizes that correspond to the primary and secondary dimensions of the related garment(s), respectively. The input body measurements should comply with the manufacturing constraints of the related garment(s). Then a sizing system is derived with the minimum number of different
Mass customization of garments 51
IJCST 22,1
52
garment sizes according to a chosen mass customization model. Consumers’ satisfaction is assessed with a new tool that is proposed in this paper. 2.2 The proposed mass customization methodology The overall methodology consists of six subsequent stages, which are shown in Figure 1 and are detailed in the following paragraphs. Input: body measurements
Stage 1: Statistical analysis
Stage 2: Linear regression analysis
Stage 3: Classification of the target population in body types
Stage 4: Determination of primary sizes
Stage 5: Determination of secondary sizes for every primary size
Stage 6: Output mass customization model
Figure 1. The flow chart of the proposed mass customization methodology
Mass customization model 1
Mass customization model 2
Mass customization model 3
Case A
Mass customization model 4
Case B
Stage 1: statistical analysis of body data. A statistical analysis of body measurements takes place in order to determine the range of the various body dimensions. This analysis includes the calculation of the minimum (Min), maximum (Max) and mean value (Mean) as well as the standard deviation (SD) for every distinct set of body measurements. Stage 2: linear regression analysis. Using the least squares linear regression analysis technique, the correlation between different body dimensions is determined. In this way, one is able to distinguish the primary from the secondary dimensions of the sizing system; essential information for the manufacture of garment patterns (Robertson and Minter, 1996; Jarosz, 1999; Barroso et al., 2005). Generally, in this method, the selection of the primary and secondary dimensions is performed with respect to Constraints A and B in order to ensure that the produced sizing system will comply with the manufacturing constraints of the related garment and with the specific body measurements of the target population. This selection is facilitated using the correlation coefficient R. According to the BS 7231 Standard (BS 7231, 1990) two dimensions are related according to the following rules: if the correlation coefficient R , 0.5, then no relationship exists; if 0.5 , R , 0.75, then there is a mild relationship; if R . 0.76, then a strong relationship exists. Using R, it is possible to reduce the number of independent measurements by removing those which exhibit a strong correlation along the primary and secondary dimensions. Stage 3: classification of the target population in body types. The purpose of this stage is to classify the target population in body shapes according to the body measurements of each subject. The classification is performed utilizing: . the height dimension; and . the “drop value” DV ¼ [(Chest girth) 2 (Waist girth)] (Cooklin, 1999; Gupta and Gangadhar, 2004).
Mass customization of garments 53
The “drop value” parameter is used to classify the different body shapes of the target population by determining distinct relationships between key dimensions (Cooklin, 1999; prEN 13402-3, 2004). Based on drop values, the population is classified into categories which correspond to generally perceived body shapes. In this stage, we separate the target population in six body types according to their height (Table I) and in seven body types with respect to their drop value (Table II). This partition will allow of a sizing system covering a large range of different body shapes. Stage 4: determination of primary sizes. The determination of the primary dimension depends on the garment type and Constraint A. Therefore, using the results Body type Very short , Mean 2 2SD Mean 2 2SD , Short , Mean 2 SD Mean 2 SD , Normal , Mean Mean , Tall , Mean þ SD Mean þ SD , Very tall , Mean þ 2SD Too much tall . Mean þ 2SD
Height (cm) , 164 164-171 171-178 178-185 185-192 192 .
Table I. Classification of the population vs height
IJCST 22,1
54
of stage 2 analysis, we are able to associate a body dimension with the primary dimension that is used for the manufacture of the related garment. In this way, the target population is classified in primary sizes derived along the selected primary dimension. In Gupta and Gangadhar (2004), the primary size ranges between [(Mean 2 SD), (Mean þ SD)]. In this paper, we chose a larger range between [(Mean 2 2SD), (Mean þ 2SD)], in order to take into account a greater part of the population (. 95 per cent). This choice, however, does not necessarily affect the final number of the produced garment sizes, since in the proposed method one is able to select the degree of mass customization. A threshold d is determined afterwards in order to define the difference between two successive primary sizes xk, xkþ 1. In this way, the obtained sizing system satisfies Constraint A as long as the corresponding part of the population lies within the interval [Mean 2 2SD, Mean þ 2SD]. The resulted number of primary sizes is: n¼
ðMean þ 2SDÞ 2 ðMean 2 2SDÞ 4SD ¼ d d
ð1Þ
Although the development of a sizing system can be based either on a constant or a variable d (McCulloch et al., 1998), in the proposed method, a constant d is selected for compatibility with the prEN 13402-3 (2004) standard. The effectiveness of this particular selection is verified by the application of the proposed method in recent anthropometric data. Stage 5: determination of secondary sizes for every primary size. For every primary size, a set of secondary sizes is developed according to the type of garment and Constraint B. Similarly to the primary dimension, the secondary dimension lies also within [Mean 2 2SD, Mean þ 2SD] and the number of obtained secondary sizes n is calculated through equation (1). All secondary sizes are grouped together into two-dimensional tables with respect to the selected primary dimension and are utilized for the development of the required sizing system. Stage 6: selection of a mass customization model. Each mass customization model defines a sizing system that starts with the “medium body shape” (according to Stage 3) and proceeds to covering “neighboring body shapes” according to this model. Each model depends on the desired mass customization degree which is iteratively improved in sequential steps. For each obtained mass customization model, a sizing system is developed according to the following scheme: (1) Step 0: mass production. Development of medium sizes (primary sizes) only, which are traditionally used for the mass production of preˆt-a-porter garments. Body type
Table II. Classification of population according to the “drop value”
Very small Small Medium Full Large Extra large Very extra large
Difference of chest – waist girth (cm) .18 14-18 10-14 6-10 2-6 2 2 to 2 , 22
(2) Step 1: mass customization model 1. Development of two secondary sizes for every primary size according to the first secondary dimension. (3) Step 2: mass customization model 2. Development of eight secondary sizes for each primary size according to the first and second secondary dimensions. This model is valid for garments with two secondary dimensions. (4) Step 3: mass customization model 3: . Case A (only one secondary dimension exists). Development of four secondary sizes for every primary size according to the secondary dimension. . Case B (at least two secondary dimensions exist). Development of up to 25 sizes along the two secondary dimensions for every primary size. These sizes are produced for the two body categories which are “before” and “after” the medium body shape. (5) Step 4: mass customization model 4. Development of all possible sizes for every primary and secondary dimension. This model corresponds to a sizing system with the maximum population satisfaction and the larger number of different garment sizes. This is the best possible degree of mass customization for a particular target population. Our experiments show that the resulted garments fit to the 99.9 per cent of the target population. The selection of a mass customization model is facilitated using a new assessment tool called as “Total satisfaction percentage.” The new index is based on the value of “satisfaction percentage” according to the following definitions. Definition 3 (“satisfaction percentage”). For each size k of the sizing system of a given mass customization model j, and for each garment type i, the satisfaction percentage aijk of the target population is equal to the percentage of the target population that its three characteristic (one primary and two secondary) body dimensions (xi, yi, zi) fall within the corresponding garment sizes (xik, yik, zik) up to a maximum tolerance value (Dxi,max, Dyi,max, Dzi,max), i.e. jxi 2 xik j # Dxi;max , jyi 2 yik j # Dyi;max , jzi 2 zik j # Dzi;max . Here: 1 ðDxi;max ; Dyi;max ; Dzi;max Þ ¼ ðdp ; ds1 ; ds2 Þ; 2 where dp, ds1, ds2 correspond to the constant difference between two successive values of the primary dimension, and the first and second secondary dimensions, respectively. Definition 4 (“total satisfaction percentage”). The “Total satisfaction percentage” aij of a target population with respect to a given garment type i and a mass customization model j is calculated by: n X aijk ð2Þ a ij ¼ k¼1
Contrarily to existing approaches (see, e.g. the “aggregate loss” of fit (Tryfos, 1986; McCulloch et al., 1998; Gupta and Gangadhar, 2004)) the proposed indicator is able to measure the degree (or percentage) to which a certain sizing system “satisfies” the target population. On the other hand, the “aggregate loss” index simply expresses the average Euclidean distance between the dimensions of individuals and their allocated garment size. Therefore, the proposed index is able to assist a garment manufacturer to
Mass customization of garments 55
IJCST 22,1
56
select the mass customization model that is most appropriate for its production process and market pursues. 3. Application of the proposed methodology for the development of mass customization models for a Greek-men population The proposed methodology is applied to anthropometric data that correspond to 12,180 Greek men from the entire country aging between 20 and 30 years old. Six basic body dimensions were measured the time period between 2003 and 2004 by the traditional technique. These include the linear dimensions height, inside leg length, sleeve/arm length and the girth dimensions of chest, neck and waist. The purpose of this application is to derive mass customization models for the production of men: . shirts; . garments of the upper body (excluding shirts and underwear); and . garments for the lower body. In order to avoid redundant analysis, this section presents the first five stages of the application of the proposed method, while Section 4 summarizes the different mass customization models according to stage 6. 3.1 Stage 1: statistical analysis of body data The results of the statistical analysis of the six body dimensions of the 12,180 individuals are listed in Table III. The mean values (Mean) express the body dimensions of the “average” 20-30 years old Greek man. 3.2 Stage 2: linear regression analysis For each pair of body dimensions, the linear regression correlation coefficient R is calculated in order to determine the body dimensions that will be utilized for satisfying constraints A and B with respect to the related garment type. Table IV summarizes the results of this analysis. According to Table IV, there is a strong relationship (R ¼ 0.9948) between the chest and waist girth as it is also shown in Figure 2. For clarity reasons, the diagram points (rhombs) depict the mean value of the waist girths that correspond to a certain chest-girth value. Similarly, the chest girth vs the neck girth has very strong relationship, while there is a mild relationship between the chest girth and the height. There is no strong relationship between the waist girth and linear dimensions. The height has strong relationship with the linear dimensions sleeve and inside leg length. Dimension
Table III. The results of the statistical analysis of the six body dimensions of the target population
Mean
SD
Min Max Mean 2 SD Mean þ SD Mean 2 2SD Mean þ 2SD
Height 178 7 154 Chest girth 96.7 8.7 82 Waist girth 88.3 10.8 70 Neck girth 38.8 2.5 34 Inside leg length 80 5 65 Sleeve length 61.3 3 54
208 137 135 51 117 71
171 88 77 36 75 58
185 106 99 41 85 64
164 80 67 34 69 56
192 114 110 44 90 67
Summarizing, there is a strong linear correlation within length and girth dimensions, but there is no linear correlation between length and girth dimensions. This is an important observation since most empirical size diagrams are based on a linear increment of the sizing systems. In other words, it is mistakenly supposed that when the girth body size is increased, the height size is increased, respectively. Because of this issue, existing sizing systems result to clothes which are suitable to a limited percentage of the target population (Ashdown, 1998; Lanenegger and van Osch, 2002). Based on the above, the chest girth is selected as the primary dimension for upper body garments, while the neck girth is chosen as the primary dimension for the case of shirts. Also, the waist girth is considered as the primary dimension for the development of lower body garments. The rest of the dimensions like the height, sleeve or leg length will be selected as secondary dimensions with respect to one of the related garment types.
Mass customization of garments 57
3.3 Stage 3: classification of the population in body types 3.3.1 Population vs height. The resulted classification of the target population is depicted in Table V. The population is divided into six categories with respect to the height dimension.
Dimension
Height
Chest girth
Waist girth
Neck girth
Height Chest girth Waist girth Neck girth Inside leg length Sleeve length
1.0000 0.6354 0.6740 0.3886 0.9802 0.9927
0.7875 1.0000 0.9948 0.9866 0.0387 0.8531
0.7024 0.9928 1.0000 0.9712 0.2131 0.7664
0.7543 0.9634 0.9700 1.0000 0.4704 0.7671
Table IV. Correlation coefficient R of the body dimensions
140 130
Mean waist girth
120 110 100 90 y = 1.0445x – 12.121 R = 0.9948
80 70 60 80
90
100
110 Chest girth
120
130
140
Figure 2. Correlation of chest girth vs mean waist girth
IJCST 22,1
3.3.2 Population vs “drop value”. The resulted classification of the target population according to the drop value is listed in Table VI. Based on this analysis, it results that a large part of Greek male population has greater waist girth compared to the “medium” body type and it is categorized under the “full” body type. Thus, the difference between chest and waist girth for the major percentage of the population equals to 8 cm.
58
3.4 Stage 4: determination of primary sizes In this stage, the target population is classified in sizes according to the chosen primary dimension. More specifically: . the chest girth is selected for the manufacture of the upper body garments; . the neck girth is selected for the manufacture of shirts; and . the waist girth is selected for the manufacture of lower body garments. The chest girth dimension is subdivided setting by d ¼ 4 cm between two successive sizes according to the EN Standard (prEN 13402-3, 2004). This results to eight discrete sizes (n ¼ 8) in the range between 82 and 114 cm. Similarly, the neck girth dimension is subdivided according to d ¼ 1 cm resulting to ten discrete sizes (n ¼ 10) in the range between 34 and 44 cm. Finally, the waist girth is divided into sizes according to d ¼ 4 cm, which results to ten discrete sizes (n ¼ 10) in the range between 70 and 110 cm. 3.5 Stage 5: determination of secondary sizes for every primary size 3.5.1 Upper body garments (excluding shirts and underwear). For upper body garments (excluding shirts and underwear), the primary dimension is the chest girth and the secondary dimensions are the waist girth and the height. Both secondary dimensions lie within Mean 2 2SD and Mean þ 2SD. For example, for chest girth size 90-94, five waist girth sizes are developed (d ¼ 4) which correspond to individuals belonging to Body type
Table V. Classification of the population vs height
Very short , Mean 2 2SD Mean 2 2SD , Short , Mean 2 SD Mean 2 SD , Normal , Mean Mean , Tall , Mean þ SD Mean þ SD , Very tall , Mean þ 2SD Too much tall . Mean þ 2SD
Body type
Table VI. Classification of the target population according to the drop value
Very small Small Medium Full Large Extra large Very extra large
Height
Population
Percentage
, 164 164-171 171-178 178-185 185-192 192 .
240 2,040 4,568 3,978 1,143 211
2.0 16.7 37.5 32.7 9.4 1.7
Difference of chest – waist girth (cm)
Population
%
. 18 14-18 10-14 6-10 2-6 2 2 to 2 , 22
550 1,302 2,707 3,243 2,313 1,333 732
5 11 22 27 19 11 6
the Medium body type. This distribution is depicted with the 3D graph shown in Figure 3. Introducing the height as the second secondary dimension four more sizes are developed for every primary and first secondary size as it is shown in Table VII for chest girth size between 90 and 94 cm. These four height sizes correspond to the short, normal, tall and very tall body types. 3.5.2 Male shirts. The primary dimension for male shirts is neck girth and the secondary dimension is sleeve length. The secondary dimension lies within Mean 2 2SD and Mean þ 2SD. For example, for neck size 38, six sleeve length sizes are developed (d ¼ 2) which correspond to individuals belonging to the Medium body type. This distribution is depicted with the 3D graph shown in Figure 4. 3.5.3 Lower body male garments. For lower body male garments like the trousers, the primary dimension is the waist girth and the secondary dimension is the inside leg length. The secondary dimension lies within Mean 2 2SD and Mean þ 2SD. For example, for waist girth size 74-78, six inside leg length sizes are developed (d ¼ 3) which correspond to individuals belonging to the Medium body type. This distribution is depicted with the 3D graph shown in Figure 5.
Mass customization of garments 59
6 82-86
5
86-90 4
90-94
% 3
94-98
102-106
0
106-110
90-94
Total
106-110
98-102
90-94
82-86
102-106
94-98
Waist girth
Chest girth
Figure 3. 3D representation of the sizes produced according to the chest (primary dimension) and the waist girth (secondary dimension)
110-114 86-90
78-82
98-102
1
70-74
2
Chest girth
Waist girth
164-171
171-178
74-78 78-82 82-86 86-90 90-94
0.60 0.90 1.00 0.50 0.40 3.40
1.20 2.00 2.20 1.30 0.70 7.40
Height (per cent) 178-185 0.70 1.50 1.50 1.00 0.70 5.40
185-192
Total
0.10 0.30 0.40 0.30 0.10 1.20
2.60 4.70 5.10 3.10 1.90 17.40
Table VII. Sizes for upper body men garments (excluding shirts and underwear) and corresponding population percentages for chest girth size between 90 and 94 cm
IJCST 22,1
60 %
4.5
35
4.0
36
3.5
37
3.0
38
2.5
39
2.0
40 41
1.5
42
1.0
43
0.5
44
Sleeve length
39 37
4.0
70-74
3.5
74-78 78-82
3.0
82-86
2.5
86-90
% 2.0
90-94 94-98
1.0
98-102
0.5
102-106 94 98
86-90
78-82
70-74
Inside leg length
88-91
82-85
76-79
0.0
102-106
1.5
70-73
Figure 5. 3D representation of the sizes derived according to the waist girth (primary dimension) and the inside leg length (secondary dimension)
43
Neck girth
35
66-68
62-64
58-60
41
0.0 54-56
Figure 4. 3D representation of the sizes derived according to the neck (primary dimension) and sleeve length (secondary dimension)
106-110
Waist girth
4. Mass customization models for the Greek men population: application of stage 6 of the new methodology The mass customization models presented in section 2 are applied to the following categories of male garments: . shirts; . coats; and . trousers.
Mass production for preˆt-a`-porter garments correspond to conventional sizing systems. The rest sizing systems presented below correspond to the mass customization models 1-4. 4.1 Sizing systems for male shirts Ten medium sizes are proposed according to the mass production model of preˆt-a-porter shirts. The medium sizes correspond to sleeve length 60-62 cm (Table VIII): . According to mass customization model 1, 30 sizes are developed based on the sleeve length: ten for sleeve length 58-60 cm and ten for sleeve length 62-64 cm. . Mass customization model 2 is not applicable in this case because there is not a need to use a second secondary dimension. . Mass customization model 3 implies the development of 50 sizes: four more sizes based on the secondary dimension are developed for each medium size. Hence, ten more sizes are produced for sleeve length 56-58 cm, and ten for sleeve length 64-66 cm. . Mass customization model 4 implies the development of 59 sizes, which provides a comprehensive degree of mass customization for male shirts.
Mass customization of garments 61
4.2 Sizing systems for male coats Utilizing the chest and waist girth as well as height data of the target population, three mass customization models are developed for male coats. However, the obtained sizing systems can be regarded as general sizing systems for producing male upper garments excluding shirts and underwear. All results are summarized in Table IX, where for spacing reasons, we have omitted the sizes which correspond to chest girths between 94 and 110 cm: . According to mass production model, eight medium sizes based on the chest girth are produced. These regular sizes correspond to the waist girth and height of body type B (Medium). . According to mass customization model 1, 23 sizes are developed using the waist girth as secondary dimension. Seven sizes for body type A (small), eight for body type B (medium) and eight for body type C (full).
Sleeve length A B C D E F G H
55 57 59 61 63 65 67 69
54-56 56-58 58-60 60-62 62-64 64-66 66-68 68-70
0 35
1 36
2 37
0A 0B 0C 0D 0E 0F
1A 1B 1C 1D 1E 1F
2A 2B 2C 2D 2E 2F
3 38
Neck girth 4 5 39 40
6 41
7 42
8 43
9 44
3B 3C 3D 3E 3F 3G
4B 4C 4D 4E 4F 4G
6B 6C 6D 6E 6F 6G
7B 7C 7D 7E 7F 7G
8B 8C 8D 8E 8F 8G
9B 9C 9D 9E 9F
5B 5C 5D 5E 5F 5G
Table VIII. Sizing system for male shirts
IJCST 22,1 Chest girth
Body type
1 (82-86)
A-Small B-Medium C-Full D-Large E-Extra large A-Small B-Medium C-Full D-Large E-Extra large A-Small B-Medium C-Full D-Large E-Extra large
62 2 (86-90)
3 (90-94)
Table IX. Sizing system for upper body men garments excluding shirts and underwear
.. . 8 (110-114)
.
.
A-Small B-Medium C-Full D-Large E-Extra large
Height 2 (Medium) 3 (Tall) 171-178 178-185
Waist girth
1 (Short) 164-171
4 (Very tall) 185-192
70-74 74-78 78-82 82-86 70-74 74-78 78-82 82-86 86-90 74-78 78-82 82-86 86-90 90-94
1B1 1C1 1D1 1E1 2A1 2B1 2C1 2D1 2E1 3A1 3B1 3C1 3D1 3E1
1B2 1C2 1D2 1E2 2A2 2B2 2C2 2D2 2E2 3A2 3B2 3C2 3D2 3E2
1B3 1C3 1D3 1E3 2A3 2B3 2C3 2D3 2E3 3A3 3B3 3C3 3D3 3E3
1B4 1C4 1D4 1E4 2A4 2B4 2C4 2D4 2E4 3A4 3B4 3C4 3D4 3E4
94-98 98-102 102-106 106-110
8A1 8B1 8C1 8D1
8A2 8B2 8C2 8D2
8A3 8B3 8C3 8D3
8A4 8B4 8C4 8D4
According to mass customization model 2, 69 sizes are developed using the waist girth and height as secondary dimensions. Eight more sizes are produced for every primary size compared to the mass production model. According to mass customization models 3 and 4, 152 sizes are developed using the waist girth and height as secondary dimensions. About 24 more sizes are produced for every primary size compared with the mass production model. In this case, mass customization model 4 results to the same sizing system with the third one.
4.3 Sizing systems for male trousers Utilizing the waist girth and inside leg length, four mass customization models are developed for male trousers. All results are summarized in Table X: . According to mass production model, ten medium sizes based on the waist girth are produced. These medium sizes correspond to the inside leg length 79-82 cm. . According to mass customization model 1, 30 sizes are developed using the inside leg length as secondary dimension. About 20 more sizes than the mass production model are developed: ten for inside leg length 76-79 cm and ten for inside leg length 82-85 cm. . Mass customization model 2 is not applicable in this case since there is not a need to use a second secondary dimension. . According to mass customization model 3, 50 sizes are developed using the inside leg length. About 20 more sizes are developed compared to mass
A B C D E F G H
Inside leg length
72 75 78 81 83 86 89 92
70-73 73-76 76-79 79-82 82-85 85-88 88-91 91-93
0A 0B 0C 0D 0E 0F
0 72 70-74 1A 1B 1C 1D 1E 1F
1 76 74-78 2A 2B 2C 2D 2E 2F
2 80 78-82 3A 3B 3C 3D 3E 3F 3G
3 84 82-86 4A 4B 4C 4D 4E 4F 4G 4H
5A 5B 5C 5D 5E 5F 5G 5H
Waist girth 4 5 88 92 86-90 90-94 6A 6B 6C 6D 6E 6F 6G 6H
6 96 94-98 7A 7B 7C 7D 7E 7F 7G 7H
7 100 98-102
9 108 106-110 9A 9B 9C 9D 9E 9F 9G 9H
8 104 102-106 8A 8B 8C 8D 8E 8F 8G 8H
Mass customization of garments 63
Table X. Sizing system for lower body men garments
IJCST 22,1 .
64
customization model 2: ten for inside leg length 73-76 cm, and ten for inside leg length 85-88 cm. According to mass customization model 4, 73 sizes are developed. The produced sizing system satisfies 99.9 per cent of the target population.
4.4 Assessment of the developed sizing systems Table XI shows the results of applying the proposed satisfaction percentage index for the assessment of the developed mass customization models for the three aforementioned garment types. In the case of mass production (model 0), the total satisfaction percentage values are unacceptably low, particularly when the garment manufacturing constrains involve three distinct dimensions (e.g. coats). The satisfaction percentage is considerably increased with mass customization model 1, which expresses a first step towards mass customization for the case of shirts and trousers. Mass customization 2 is required for cloths like coats where three dimensions are utilized by the corresponding manufacturing constraints. This model provides a significant improvement to the cloths fitting compared to the previous one. Mass customization models 3 and 4 provide a considerable level of mass customization. The total satisfaction percentages of both models in the target population are very high, with the cost of producing a relatively large number of garment sizes. All results are shown in the graph of Figure 6. 5. Implementation Mass production strategies have driven apparel production for decades with a negative impact in design and fit of clothing. These strategies have categorized whole populations by a relatively small number of sizing systems and made it virtually impossible to meet the needs of those individuals who have special fitting requirements (Istook, 2002). The proposed methodology for the development of sizing systems combined with computer-aided and information technologies can enable the creation of garments, customized for fit, in a very quick and accurate manner. These customized garments can be inserted into normal production lines as an additional “size” and produced like every other garment of the same style. This means that successful companies with huge libraries of garment styles would be able to implement the proposed mass customization strategy with relatively little effort. Potential increase in production cost that would occur due to cutting a few garments at a time, rather than hundreds, could be offset with increases in sales and customer loyalty. The proposed method for mass customization can be implemented by using either manual measurements or automatic scanning as well as. In this paper, we applied the proposed method in sizing data taken by conventional manual methods which are usually subject to noise and inaccuracies. These shortcomings are not expected to appear in automatic scanning data which are high-accurate and filtered for noise removal. Finally, implementing a field research for the collection of anthropometric data are usually an expensive and time-consuming task which prerequisites the accomplishment of several criteria (Ujevic et al., 2006). Discussing on this task is, however, out of the scope of the present research. 6. Conclusions A new methodology for the mass customization of garments has been proposed in this paper. With the proposed method, it is possible to control the mass customization
0 1 2 3 4
10 30 – 50 59
24.9 67.4 – 90.2 92.4
0 1 2 3 4
8 23 69 152 152
9.6 24.0 53.2 83.3 83.3
0 1 2 3 4
10 30 – 50 73
23.4 55.9 – 80.3 93.9
Upper body garments Lower body garments Shirts Coats Trousers Model Number of sizes n Total satisfaction (%) Model Number of sizes n Satisfaction (%) Model Number of sizes n Total satisfaction (%)
Mass customization of garments 65
Table XI. The total satisfaction percentage of the target population according to developed mass customization models
66 Figure 6. The total satisfaction percentage of the target population with respect to the number of different garment sizes
100% Total satisfaction percentage aij
IJCST 22,1
80%
60%
40%
Shirts Coats
20%
Trousers
0% 0
50
100
150
200
Number of sizes n
degree according to four different models, which affect the corresponding sizing system of a specific garment type. In this way, a garments producer can choose the mass customization model that is closer to his manufacturing line and marketing pursues. This selection is facilitated through an assessment tool called “Total satisfaction percentage.” This statistical tool is in place to examine how much a sizing system “satisfies” the target population and it is harmonized with the European Standards of sizing systems development, which are going to be applied in the near future. The proposed methodology has been successfully applied for the development of mass customization models for male shirts, coats and trousers with respect to Greek men between 20 and 30 years old. This application indicates the limitations of conventional sizing systems and the advantages of using a “scalable” mass customization model. The methodology presented herein, can be applied to the development of mass customization models for other categories of garments and target population. Our future work includes the development of methods for the automatic garments grading with respect to the aforementioned mass customization models and practise. References Anderson, L.J., Brannon, E.L., Ulrich, P.V., Marshall, T., Staples, N., Grasso, M., Butenhoff, P. and Beninati, M. (1999), Discovering the Process of Mass Customization: A Paradigm Shift for Competitive Manufacturing, American Apparel Manufacturers Association, Auburn, AL, available at: www.auburn.edu/mcrp.html Ashdown, S.P. (1998), “An investigation of the structure of sizing systems: a comparison of three multidimensional optimized sizing systems generated from anthropometric data with the ASTM standard D5585-94”, International Journal of Garments Science and Technology, Vol. 10 No. 5, pp. 324-41. Ashdown, S.P. and Delong, M. (1995), “Perception testing of apparel ease variation”, Applied Ergonomics, Vol. 26 No. 1, pp. 47-54.
Barroso, M.P., Arezes, P.M., da Costa, L.G. and Miguel, A.S. (2005), “Anthropometric study of Portuguese workers”, International Journal of Industrial Ergonomics, Vol. 35, pp. 401-10. BS 7231 (1990), Part 1: Body Measurements of Boys and Girls from Birth to 16.9 Years, British Standards Institution, London. Cooklin, G. (1999), Pattern Grading for Women’s Clothes: The Technology of Sizing, Blackwell Science, Oxford, pp. 3-18. Davis, S.M. (1987), Future Perfect, Addison-Wesley, Reading, MA. DeLong, M., Ashdown, S., Butterfield, L. and Turnbladh, K.F. (1993), “Data specifications needed for apparel production using computers”, Garments Textile Research Journal, Vol. 11 No. 4, pp. 1-7. Fralix, M.T. (2000), “Mass customization using the internet”, Proceedings of the 80th World Conference of the Textile Institute, Manchester, 16-19 April. Gazzuolo, E., DeLong, M., Lohr, S., LaBat, K. and Bye, E. (1992), “Predicting garment pattern dimensions from photographic and anthropometric data”, Applied Ergonomics, Vol. 23, pp. 161-71. Goldsberry, E., Shim, S. and Reich, N. (1996), “Women 55 years and older: overall satisfaction and dissatisfaction with the fit of ready-to-wear: Part II”, Garments and Textile Research Journal, Vol. 14 No. 2, pp. 121-31. Gupta, D. and Gangadhar, B.R. (2004), “A statistical model for developing body size charts for garments”, International Journal of Garments Science and Technology, Vol. 16 No. 5, pp. 458-69. Hsu, C.-H. and Wang, M.-J.J. (2005), “Using decision tree-based data mining to establish a sizing system for the manufacture of garments”, International Journal of Advanced Manufacturing Technology, Vol. 26, pp. 669-74. Istook, C. (2002), “Enabling mass customization: computer-driven alteration methods”, International Journal of Clothing Science & Technology, Vol. 14 No. 1, pp. 61-76. Istook, C., Little, T., Hong, H. and Plumlee, T. (2003), “Automated garment development from body scan data”, NTC Project S00-NS15, National Textile Center Annual Report, November (formerly I00-S15). Jarosz, E. (1999), “Anthropometry of elderly women in Poland: dimensions for design”, International Journal of Industrial Ergonomics, Vol. 25, pp. 203-13. Kotha, S. (1995), “Mass customization: implementing the emerging paradigm for competitive advantage”, Strategic Management International, Vol. 16, pp. 21-42. LaBat, K.L. and Delong, M.R. (1990), “Body cathexis and satisfaction with fit of apparel”, Garments and Textile Research Journal, Vol. 8 No. 2, pp. 42-8. Lanenegger, R. and van Osch, R. (2002), “One size for Europe: a garment size system for Europe”, Avantex Frankfurt, 13 May. Loker, S., Ashdown, S. and Schoenfelder, K. (2005), “Size-specific analysis of body scan data to improve apparel fit”, Journal of Textile and Apparel, Technology and Management, Vol. 4 No. 3, pp. 1-15. McCulloch, C.E., Paal, B. and Ashdown, S.A. (1998), “An optimization approach to apparel sizing”, Journal of the Operational Research Society, Vol. 49, pp. 492-9. Pine, B.J. (1993), Mass Customization: The New Frontier in Business Competition, Harvard Business School Press, Boston, MA. prEN 13402-3 (2004), “Size designation of clothes – Part 3: measurements and intervals”, Final Draft.
Mass customization of garments 67
IJCST 22,1
68
Robertson, S.A. and Minter, A. (1996), “A study of some anthropometric characteristics of motorcycle riders”, Applied Ergonomics, Vol. 27 No. 4, pp. 223-9. Salusso-Deonier, C.J. (1982), “A method for classifying adult female body form variation in relation to the US standard for apparel sizing”, doctoral dissertation, University of Minnesota, Minneapolis, MN, available at: www.wsu.edu:8080/, salusso/BODY/s.html Salusso-Deonier, C.J. (1989), “Gaining a competitive edge with top quality sizing”, Quality Congress Transactions, Vol. 43, American Society of Quality Control, Toronto, pp. 371-6. Tryfos, P. (1986), “An integer programming approach to the apparel sizing problem”, Journal of the Operational Research Society, Vol. 37 No. 10, pp. 1001-6. Ujevic´, D., Rogale, D., Drenovac, M., Pezelj, D., Hrastinski, M., Narancˇic´, S.N., Mimica, Z. and Hrzˇenjak, R. (2006), “Croatian anthropometric system meeting the European Union”, International Journal of Clothing Science & Technology, Vol. 18 No. 3, pp. 200-18. Walter, L. (2002), “Will the “e-Tailor” become reality?”, Mass-Customization, Industrial Customisation, Industrial MtM and Personalised On-Line Shopping in the European Fashion Business – Project Results & Future Perspectives, Euratex, The EU Apparel Business Goes High-Tech, Brussels, 15 October. Walter, L. (2006), “The textile & clothing industry in Europe”, Textile Asia, Vol. 37 No. 4, pp. 40-6. Workman, J.E. (1991), “Body measurement specifications for fit models as a factor in garments size variation”, Garments and Textile Research Journal, Vol. 10 No. 1, pp. 31-6. Corresponding author Philip N. Azariadis can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Recognition and re-visualization of woven fabric structures
Woven fabric structures
Pranut Potiyaraj Department of Materials Science, Faculty of Science, Center of Excellence in Textiles, Chulalongkorn University, Bangkok, Thailand and National Center of Excellence for Petroleum, Petrochemicals, and Advanced Materials, Chulalongkorn University, Bangkok, Thailand, and
79 Received 4 May 2009 Accepted 4 September 2009
Chutipak Subhakalin, Benchaphon Sawangharsub and Werasak Udomkichdecha Department of Materials Science, Faculty of Science, Center of Excellence in Textiles, Chulalongkorn University, Bangkok, Thailand Abstract Purpose – The purpose of this paper is to develop a computerized program that can recognize woven fabric structures and simultaneously use the obtained data to 3D re-visualize the corresponding woven fabric structures. Design/methodology/approach – A 2D bitmap image of woven fabric was initially acquired using an ordinary desktop flatbed scanner. Through several image-processing and analysis techniques as well as recognition algorithms, the weave pattern was then identified and stored in a digital format. The weave pattern data were then used to construct warp and weft yarn paths based on Peirce’s geometrical model. Findings – By combining relevant weave parameters, including yarn sizes, warp and weft densities, yarn colours as well as cross-sectional shapes, a 3D image of yarns assembled together as a woven fabric structure is produced and shown on a screen through the virtual reality modelling language browser. Originality/value – Woven fabric structures can now be recognised and simultaneously use the obtained data to 3D re-visualize the corresponding woven fabric structures. Keywords Pattern recognition, Image processing, Computer software, Fabric production processes Paper type Research paper
Introduction Recent advances in computer technology offer several automation processes in the weaving industry including structure recognition and characterization as well as 3D visualization of woven fabrics. Woven fabric structures greatly affect optical and mechanical properties of woven fabric. Accurate identification of the structural characteristics is required Sincere gratitude must be given to the Thailand Research Fund for funding this project. Also, this work would not have been possible without the support of National Center of Excellence for Petroleum, Petrochemicals and Advanced Materials, Chulalongkorn University.
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 79-87 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018577
IJCST 22,2/3
80
for quality control of the existing products, and is useful for developing new products. Traditionally, these characteristics, especially weave pattern, are manually determined by visual inspection. This manual operation is usually tedious, time consuming, and depends on the skills of the inspector. Some computerized methods for recognition of weave patterns from fabrics have been proposed, mostly based on image analysis techniques. Ohta et al. (1986) used a drum scanner to capture reflected images of woven fabrics. An image analysis was utilized to obtain the central lines of warps and wefts, and the fabric density was consequently determined. As described by Kinoshita et al. (1989) reflected and transmitted images of woven fabrics were captured using a charge-coupled device (CCD), and the obtained data were used to calculate the power spectra which were different for each type of weave. By acquiring reflected images from an image scanner, a similar technique utilizing fast Fourier transform to obtain power spectra was used to detect the appearance of reed marks (Sakaguchi et al., 2000). Using decision rules for identifying warp and weft floats based on geometric features of yarn distribution, the digitalized images captured from a microscope were analyzed to identify basic weave patterns. It was also reported that the determination of fabric counts using this technique show good agreement with manual measurement (Huang et al., 2000). Kang et al. (1999) used a CCD camera to capture fabric reflected and transmitted images perpendicular to the focal plane with halogen light mounted on the top of the fabrics. Both reflected and transmitted images were subjected to several image processing techniques. The results were combined in order to detect weave patterns. Visualization of woven fabrics has been extensively studied. The main challenge for applying 3D computer graphics to the simulation of woven fabric structures is to understand sufficiently the woven fabric structures (Hearle, 1994). A number of mathematical models have been developed in order to represent the thread paths in a woven fabric. One of the most important models is that of Peirce (1937) which assumes that yarn is incompressible, flexible and of circular cross-section. This geometrical model has been widely adopted and modified by many researchers (Lin, 1996). None of these models was believed to be the actual path, but all were reasonable approximations. The paths in real fabrics are complicated and depend upon yarn properties and forces, which cause the yarns to arrange themselves in positions of minimum energy, subject to external constraints and frictional effects (Lin and Newton, 1999). The structures can be represented in 2D fashion, taking the weave and some simple fabric properties such as warp and weft density into account. However, the 2D visualization cannot be totally satisfactory for today’s textile applications. Later, several programs were produced which resulted in graphics showing the thread path of fabrics in three dimensions, particularly that of Xu (1992). Her work made progress on the representations of varieties of single-layer weaves as well as a limited range of multi-layer fabrics. The employed method followed sine curves and straight lines to represent the yarn path within a fabric based on Peirce’s geometrical model for the plain weave and Love’s Peirce-like parameters based on Peirce’s geometrical model for non-plain weaves. Since then, other researchers have published papers that demonstrate solid-modelling techniques, for example, Keefe et al. (1992) and Keefe (1994a, b). Keefe’s work provides new possibilities for the simulation of woven fabrics, that is, the 3D models generated can be used further for mechanical analysis, and for engineering design and analysis of 3D woven fabrics. Lin (1996) also developed a program to generate and display various woven fabric structures in three-dimensions by computer graphics. The thread
paths displayed are generated by using cubic B-splines, the mathematics of which are well-developed, mainly for other areas of engineering design, especially in aerospace engineering. This work emphasized the true 3D solid model of limited typical weave structures. The program requires high-performance computers to process the simulation. Chen and his colleagues have developed a computerized simulation system for 3D woven fabrics based on their mathematical models (Chen et al., 1992). The structure has been parameterized and modelled mathematically (Chen et al., 1993a, b). The virtual reality modelling language (VRML) was first used for visualization of some woven structures based on the extrusion method in which the yarn cross-sectional image was swept along the predetermined yarn path (Chen and Potiyaraj, 1999). The structures were limited to angle-interlock and orthogonal structures. In their work, most of the weave parameters were also not taken into account. VRML is a computer language used for describing objects in a scene. VRML codes are simple ASCII text that can be parsed by a VRML interpreter (Ames et al., 1996). These interpreter programs are often called VRML Browser and are freely available as an add-on for several internet browsers. Lomov et al. (2007) successfully used VRML to represent both woven and knitted fabrics taking into account also mechanical parameters of the structures. In this research, image analysis techniques and algorithms for recognition of weave structures from reflected images of woven fabric were proposed. The weave patterns and parameters which were stored in digital format were, in turn, used for automatically generating 3D images of woven structures based on Peirce’s geometrical models. A Windows-based software system was developed to handle the seamless combination processes of weave pattern recognition and re-visualization of woven fabrics. Facilitating further development of fabric products, the system also allows various weave parameters to be adjusted in order to alter the appearance of the re-visualized 3D images, including yarn sizes, warp and weft densities, yarn colours as well as yarn cross-sectional shapes. Preparation and processing of fabric reflected images Two major types of woven fabrics were used as samples in this research, namely, plain and twill fabrics. They were scanned using an HP scanjet 5470c desktop flatbed scanner at the resolution of 600 dots per inch (dpi). It must be noted that the fabrics must be held properly so that warp and weft yarns were straight and perpendicular. Each square-inch scanned image was kept in the bitmap format. The scanned images of fabrics were converted into grayscale images. In order to amplify the distinction between free spaces and yarn-occupied spaces, the gray level at each pixel was reassigned by equalization the histogram. Equalization produces a flattened histogram with a more uniform gray level distribution so as to minimize the uneven distribution of gray levels of pixels caused by local illumination (Kang et al., 1999). An example of fabric before and after equalization is shown in Figure 1. Image analysis When scanning through a selected line in a prepared and processed fabric image, one encounters alternate bright and dark areas. A bright area is found when the yarn float is present, while a dark area is found when the yarn going down interlaces with the corresponding yarn or at the space not covered by warp or weft. Luminance at each pixel along a selected line in a fabric image can be calculated from RGB data according to
Woven fabric structures
81
IJCST 22,2/3
82
Figure 1. Scanned images
(a)
(b)
Notes: (a) A plain fabric; (b) a plain fabric after equalization
Grassmann’s law. A plot of luminance at each pixel along a selected line from a fabric image is shown in Figure 2. It can be seen that peaks are present in various widths and heights. Thus, the interpretation from this plot was not accurate enough. In order to smooth these peaks, the autocorrelation method was adopted using the following equations: C x;0 ¼
M X N X i
C 0;y ¼
Gi;j Gi2x;j
ð1Þ
Gi;j Gi;j2y
ð2Þ
j
M X N X i
j
where:
Figure 2. A luminance curve of a plain fabric image
Gi,j
is the luminance at coordinate.
M
is the maximum scanning point in the warp direction.
N
is the maximum scanning point in the weft direction.
Cx,0
is the autocorrelation value at points along the warp.
C0,y
is the autocorrelation value at points along the weft.
x and y
are the coordinates of pixel in the warp and weft direction, respectively.
400
400
350
350
300
300
250
250
200
200
150
150
100
100
50
50
0 R1
0
It was indicated that the autocorrelation technique was successfully adopted to determine structural repeat units in the fabric weave from the transmitted images of woven fabrics (Kang et al., 1999). Allowing a simple flatbed scanner to be used as an image capturing tool the same technique was applied to the reflected image, and a new algorithm to determine the weave structures was elaborated. When the luminance data were recalculated, a more uniform plot was obtained as shown in Figure 3. The width of each peak can be calculated by detecting the lowest point at each wave. This lowest point is the point where the slope of curve changes. This can be determined using the following equation: C x;0 ¼
C ðxþ1Þ;0 2 C ðx;0Þ 2x xþ1
ð3Þ
C 0; y ¼
C 0;ð yþ1Þ 2 C ð0; yÞ 2y yþ1
ð4Þ
Woven fabric structures
83
Scanning throughout the fabric image, the width of each peak in warp and weft directions was then combined and interpreted into a weave pattern. The spaces of yarns and the averages of the warp and weft yarns were also evaluated by calculating the position of each peak. Subsequently, warp and weft densities were determined by taking the reciprocal of the averaged spaces of warp and weft yarns (Huang et al., 2000). In order to represent a weave mathematically, a 2D binary matrix was used to keep the data of the weave pattern. An element value of the binary matrix is either 0 or 1. A value of 1 means the same as a mark on design paper indicating a warp-over-weft cross-over, and a value of 0, corresponds to a blank on design paper, meaning a weft-over-warp cross-over. The position of each element in the matrix is located by the co-ordinate (i, j), indicating the ith column from the left and the jth row from the bottom. Such a matrix is, hereafter, called a weave matrix. The determined weave patterns were digitally kept in this format. Simulation of 3D fabric structures There are several possible approaches which can be used for modelling the shape of yarns. As the most common method, yarn surfaces are approximately modelled from short cylinders or truncated cones with their axis along the direction of the yarn axis tangent. Lomov et al. (2007) created the yarn/ply surface using small simple plane elements, typically triangles. It was also discussed that a complicated model can be a 9,000
9,000
8,000
8,000
7,000
7,000
6,000
6,000
5,000
5,000
4,000
4,000
3,000
3,000
2,000
2,000
1,000
1,000 0
0 R1
Figure 3. An autocorrelated luminance curve of a plain fabric image
IJCST 22,2/3
84
tube of varying cross-sections which are created by moving the pre-defined cross-sectional shape along the yarn axis. The advantage of this method is that it can model a variety of dimensions of cross-sections along the yarn. This work employed this last-mentioned technique. Although VRML lacks the necessary primitives for cross-sectional shapes, several yarn cross-sectional shapes were employed in this work, namely, circular, ellipse, racetrack and lenticular. Peirce’s model was based on the assumption that a circular cross-sectional yarn in consideration is incompressible. In the actual situation, yarns are deformed so that two diameters, which correspond to yarn width and yarn thickness in the structures, must be considered. The extent which a yarn will be deformed is the flattening coefficient which is always less than 1. Peirce’s model dealt with circular cross-sectional yarn for which warp and weft yarn cross-sectional coordinates were calculated based on the equation of a circle. In case of an ellipse cross-section, E(xi,yi) and PX(xi,yi) were acquired using the equation of an ellipse. A racetrack shape is the combination of a half-circle with a rectangle in the middle. Thus, the co-ordinates were calculated based on the equation of circle. In the case of a lenticular cross-section image which is formed by two incomplete circles, the equation of a circle was adopted with only selected co-ordinates. Mathematical modelling of yarn paths in a single layer woven fabric structure was developed based on Peirce’s model. The yarn path can be divided into three sections, namely, the overfloat section where the warp is floated over the weft, the underfloat section where the weft is floated over the warp and the linking section where the former two sections are joined. Warp yarn path coordinates, Eðxi ; yi Þ; were determined according to equations (5) and (6). Weft yarn path coordinates, Pðxi ; yi Þ, were calculated similarly: 1 ð5Þ xi ¼ xi21 þ ppc 8 < D4 jwi; j ¼ 1 ð6Þ yi ¼ 2D : 4 jwi; j – 1 where: ppc
is the weft density (picks per centimetre).
D
is the summation of warp thickness (d1) and weft thickness (d2).
In the 3D simulation technique, an object can be created using the extrusion method. This method involves sweeping a 2D cross-section along a line, which is called a spline, resulting in 3D images. In this work, when a yarn path was used as a spline, an image of this yarn was generated by sweeping the specified yarn cross-section along the spline. The integration of these yarns at appropriate positions in a 3D space gave the simulated 3D image of the structure. Programming implementation According to the techniques and models explained earlier, a Windows-based software system was developed using Visual Basic programming. The interface of the program is shown in Figure 4. When a scanned fabric image was supplied, image processing and
Woven fabric structures
85
Figure 4. The program interface
analysis were then performed. The obtained data were further analyzed in order to identify a weave pattern as well as warp and weft densities. The acquired information was then combined with other default parameters in order to re-visualize the fabric structures. The yarn cross-sectional image was then swept through the yarn path images using VRML extrusion node. The positions of these yarn cross-sectional images in a 3D space were also automatically calculated. The VRML file was then generated accordingly and the 3D images were displayed as shown in Figure 5. It was possible to adjust some parameters, e.g. warp and weft densities, yarn colour and yarn cross-sectional shape. These altered 3D images were visualized and this feature is useful for product development. The experimental results indicated that the program was able to recognize most of plain fabrics correctly. However, in the case of twill and satin fabrics, the correct scanning was around 90 percent since, in some cases, yarns in fabrics tended to offset from the straight and perpendicular lines. The colour information was discarded during the image-processing step. In addition, coloured patterns affect the accuracy of the weave identification. 3D images of woven structures were correctly simulated according to the data obtained from the weave-recognition step. The advantage of displaying 3D fabric images using VRML is that the fabric is able to be studied at a distance or in much closer detail. Thus, internal geometry of yarns in the structures can be visually studied. Although, in this research, mechanical deformation data were not taken into account, with the proposed modelling technique, if the yarn mechanical information is available, it is possible to visualize corresponding fabric structures.
IJCST 22,2/3
86
Figure 5. 3D re-visualization of a twill weave
Conclusions A novel computer program has been developed in order to perform two tasks simultaneously: recognition and re-visualization of woven structures. An image processing and image analysis technique, as well as recognition algorithm were elaborated in order to identify woven fabric structures. Woven fabric images in digital format were obtained by an ordinary desktop flatbed scanner. The fabric images were processed into greyscale and then equalized. Autocorrelation was performed at every pixel in the equalized images, and autocorrelation curves were then plotted. The distances between peaks of autocorrelation curves as well as the peak width were used to identify the weave pattern as well as warp and weft densities. It was found that the program can identify approximately 90 percent of fabrics, in particular the plain, twill and satin fabrics without coloured patterns. Subsequently, 3D simulation of the woven fabric structure was done by generating each warp and weft yarn images and assembling them into fabric images. Various yarn cross-section shapes, namely, circular, racetrack and lenticular, were employed. Yarn paths were computed according to several parameters based on Peirce’s geometrical model. The yarn cross-sectional co-ordinates were then swept through the yarn path co-ordinates using VRML extrusion node resulting in a 3D image of the woven fabric structure. References Ames, A.L., Nadeau, D.R. and Moreland, J.L. (1996), VRML 2.0 Source Book, 2nd ed., Wiley, Toronto. Chen, X. and Potiyaraj, P. (1999), “CAD/CAM for orthogonal and angle-interlock woven structures for industrial applications”, Textile Research Journal, Vol. 69 No. 9, pp. 648-55.
Chen, X., Knox, R.T., McKenna, D.F. and Mather, R.R. (1992), “A parametric modelling and simulation system for multi-layer fabric reinforcements”, Proceedings to the 5th International Conference on Fibre Reinforced Composites, Newcastle Upon Tyne, March 24-26, The Plastic and Rubber Institute, London, pp. 26/1-26/11. Chen, X., Knox, R.T., McKenna, D.F. and Mather, R.R. (1993a), “Composite reinforcements: de-idialised solid modelling and computer integrated manufacturing”, Proceedings of the Conference on Managing Integrated Manufacturing: Organization Strategy & Technology, Keele, Staffordshire, Keele University, Keele, September 22-24, Vol. 2, pp. 365-77. Chen, X., Knox, R.T., McKenna, D.F. and Mather, R.R. (1993b), “Solid modelling and integrated manufacturing of textile interlinking structures”, Proceedings of International Conference: Design to Manufacture in Modern Industry, Bled, Slovenia, June 7-9, 1993, Part 2, pp. 682-8. Hearle, J.W.S. (1994), “Textile for composites”, Textile Horizons, Vol. 14 No. 6, pp. 12-15. Huang, C.C., Lui, S.C. and Yu, W.H. (2000), “Woven fabric analysis by image processing Part I: identification of weave patterns”, Textile Research Journal, Vol. 70 No. 6, pp. 481-5. Kang, T.J., Kim, C.H. and Oh, K.W. (1999), “Automatic recognition of fabric weave patterns by digital image analysis”, Textile Research Journal, Vol. 69 No. 2, pp. 77-83. Keefe, M. (1994a), “Solid modelling applied to fibrous assemblies – Part I: twisted yarns”, Journal of Textile Institute, Vol. 85 No. 3, pp. 338-49. Keefe, M. (1994b), “Solid modelling applied to fibrous assemblies – Part II: woven fabric”, Journal of Textile Institute, Vol. 85 No. 3, pp. 350-8. Keefe, M., Edwards, D.C. and Yang, J. (1992), “Solid modelling of yarn and fiber assemblies”, Journal of Textile Institute, Vol. 83 No. 2, pp. 185-96. Kinoshita, M., Hashimoto, Y., Akiyama, R. and Uchiyama, S. (1989), “Determination of weave type in woven fabric by digital image processing”, Journal of the Textile Machinery Society of Japan, Vol. 35 No. 2, pp. 1-4. Lin, H.Y. (1996), “The simulation of woven fabric structure by 3D computer graphics”, PhD thesis, University of Manchester Institute of Science and Technology, Manchester. Lin, H.Y. and Newton, A. (1999), “Computer representation of woven fabrics by using B-splines”, Journal of Textile Institute, Vol. 90 No. 1, Part 1, pp. 59-72. Lomov, S.V., Mikolanda, T., Kosek, M. and Verpoest, I. (2007), “Model of internal geometry of textile fabrics: data structure and virtual reality implementation”, Journal of the Textile Institute, Vol. 98 No. 1, pp. 1-13. Ohta, K., Sakaue, K. and Tamura, H. (1986), “Pattern recognition of fabrics surfaces”, Journal of the Textile Machinery Society of Japan, Vol. 32 No. 1, pp. 7-10. Peirce, F.T. (1937), “The geometry of cloth structure”, Journal of Textile Institute Transaction, p. T45. Sakaguchi, A., Kim, H., Matsumoto, Y. and Toriumi, K. (2000), “Woven fabric quality evaluation using image analysis”, Textile Research Journal, Vol. 70 No. 11, pp. 950-6. Xu, Y.H. (1992), “Computer representation of woven fabric structure”, MSc thesis, University of Manchester Institute of Science and Technology, Manchester. Corresponding author Pranut Potiyaraj can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
Woven fabric structures
87
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
IJCST 22,2/3
Fit analysis using live and 3D scan models Elizabeth Bye University of Minnesota, St Paul, Minnesota, USA, and
88
Ellen McKinney Art Institute of Dallas, Plano, Texas, USA
Received 19 July 2007 Accepted 4 August 2009
Abstract Purpose – The purpose of this paper is to develop a “good fit” for garments for customer satisfaction, comfort, and functionality as well as a manufacturer’s success and reputation. Design/methodology/approach – This paper reviews and evaluates garments on a live fit model and makes recommendations for the acceptance or modification of the garment for production. As more manufacturing, product development, and designing responsibilities continue to take place globally, alternatives to the traditional fit analysis are under consideration. Findings – Fit analysis using live and three-dimensional scan models as an alternative to the traditional fit analysis are under consideration. Originality/value – This paper evaluates garments on a live fit model and makes recommendations for the acceptance or modification of the garment for production. Keywords Garment industry, Image scanners, Clothing, Design, Modelling Paper type Research paper
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 88-100 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018586
Review of literature Faster, secure communication technology, and options for imaging support alternatives to traditional fit analysis. The three-dimensional (3D) scanner captures an accurate representation of the body/garment relationship that minimizes visual distractions including color and texture (Ashdown et al., 2004). This interactive image of the scanned model can be rotated or enlarged and thus results in visual fit information that provides more critical detail than photographs (Douty, 1968) or videotapes (Kohn and Ashdown, 1998). Fit researchers have used scanned images and expert judges to evaluate the fit of pants (Ashdown et al., 2004) and cooling vests (Nam et al., 2005). Using one pose, a three-point scale was used to analyze fit at 13 body points in the pant fit analysis. In the vest study, a five-point scale was used to evaluate fit at 36 body points from three different poses. Ashdown et al. (2004) concluded that 3D scan models have the potential to substitute for live fit models when: . recording one single instance of fit; . creating a database of various fit models in a single size; . recording multiple poses; and . conducting a virtual fit analysis at any time or location. Nam et al. (2005) found using 3D scan images for fit analysis convenient and accurate because there were no constraints due to time, model availability, or fatigue. Specific issues with fit that were influenced by the thickness and irregularity of the surface
were difficult to evaluate. Broader issues with expert training and developing a reliable instrument remain. Kohn and Ashdown (1998) found a positive correlation between fit analysis using a video image and traditional fit analysis using a live model. Both models resulted in a reliable analysis of the garment/body relationship. Nam et al. (2005) found that some criteria were difficult for judges to evaluate from a 3D scan. Ashdown et al. (2004) noted that some dimensions of a live fit analysis cannot be addressed from a 3D scan. The goal of this study was to compare use of a live model versus a 3D scan model on judges’ ability to evaluate fit criteria and the reliability of fit analysis scores. Research design The results of using a live model were compared to the results of using a 3D scan of the same model during a fit analysis. An expert panel of six judges completed both live and 3D scan fit analyses of a dress and a pant slopers on the model. These garments were selected for the fit analysis because they are the foundations for many clothing styles. A total of 17 criteria for pant fit and 24 criteria for dress fit were rated on a five-point scale. Judges’ ability to assess a fit score for each garment and the fit score on each criterion were compared between the live and 3D scan models. Scores from each criterion for both dress and pant were compared to evaluate differences related to individual garments or interaction between a garment and the model variation. Judges were randomly assigned, so differences in scoring were evenly distributed between judges. Therefore, no analysis for observer effect was planned. Research hypotheses This study was designed to compare the results between live and 3D scan fit analysis. Null hypotheses were formulated to test ability to score and reliability of the scores. Criterion were tested individually and results reported by group. Ability to score To compare ability to score between the two model types, the following hypothesis was postulated: H1. There is no difference in ability to assess fit analysis scores between using a live model and 3D scan model; H0. mlive ¼ m3D-scan. Each garment was constructed and modeled by a different individual. To look for scoring differences or interactions between individual garments, two additional hypotheses were developed: H1a.
There is no difference in ability to assess fit analysis scores between the 19 garments that were evaluated; H0. mGarment1 ¼ mGarment2 · · · ¼ mGarment19 .
H1b.
H0. There is no interaction between the fit analysis model type and the garment evaluated.
Reliability of scores If a live fit analysis and a 3D scan fit analysis are equivalent, each method should result in the same scores for each criterion. To test this, the following hypothesis was postulated:
Fit analysis using live and 3D scan models 89
IJCST 22,2/3
H2. There is no difference in fit analysis scores for each criterion between using a live model and 3D scan model; H0. mlive ¼ m3D-scan. Fit analysis scores for each criterion were also compared among individual garments and for interaction of garments with model type: H2a.
There is no difference in fit analysis scores between the 19 garments that were evaluated; H0. mGarment1 ¼ mGarment2 · · · ¼ mGarment19 .
H2b.
H0. There is no interaction between model type and garment evaluated.
90
Data-collection procedures Internal Review Board approval was received before data-collection procedures were initiated. Garments A total of 19 students of a sophomore level clothing design class were paired with a partner to develop a custom fit dress sloper and pant sloper. Basic dress and pant patterns were drafted according to Armstrong (2006). The dress sloper (Figure 1) was composed of a basic bodice with front and back waist darts, bust darts, back shoulder darts, one-dart sleeves, and a skirt with two front and two back darts. A center-back zipper opening was used. The pant sloper (Figure 2) was a trouser style with two front darts, two back darts, a center back zipper, and waistband. All students used the same muslin to cut and sew the test garments.
Figure 1. Dress sloper
Fit analysis using live and 3D scan models 91
Figure 2. Pant sloper
Fit analysis Expert judges participated in both a live fit analysis and a 3D scan fit analysis. In this incomplete block design, four (of six possible) randomly selected judges analyzed the fit of each garment. Two live and two 3D scan analyses were completed for each of 19 dress and 19 pant slopers, resulting in four analyses of each garment. A total of 152 garment fit analyses were completed. For the live analysis, each student modeled her garments while two judges evaluated the fit. For the 3D scan analysis, each model was scanned in her garments (Figures 3 and 4) with a VITUS/Smart 3D Body Scanner from Human Solutions. Scans were saved as rotating movie files. Judges could play, pause, and replay the video as needed during the 3D scan fit analysis.
IJCST 22,2/3
92
Figure 3. Scan of a model in dress sloper
Figure 4. Scan of a model in pant sloper
A unique fit analysis sheet was developed for the dress and pant slopers based on published fit analysis instruments. Criteria groups for both the dress and pant slopers were: . Overall, alignment. “Is the sloper correctly aligned on the body?” . Dart placement. “Are darts pointing in the right direction and ending in the correct place?” . Looseness/tightness (L/T). “Check the following areas for looseness (gapping or bagginess) or tightness (strain). Comment on needed corrections”. Under each criterion group, individual criteria were listed for analysis. See Tables I and II for individual criteria by group. The following scale, with 1 indicating the lowest score and 5 indicating the highest score was used:
Model type Criteria OA Center front Center back Left-side seam Right-side seam Waist Shoulder seam Sleeve grain Hem Dart placement Front bust darts Front skirt waist darts Back shoulder darts Back bodice waist darts Back skirt waist darts Sleeve dart L/T Front neckline Back neckline Bust Bodice back Front waist Back waist Front hip Back hip Armscye Sleeve
Live
Scan
P
38 36 38 38 38 38 38 35
30 37 14 13 23 13 22 38
0.007 0.567 0.000 0.000 0.000 0.000 0.000 0.091
38 38 38 38 37 38
14 17 5 15 16 5
0.000 0.000 0.000 0.000 0.000 0.000
38 37 38 35 38 37 38 38 38 38
38 33 38 37 37 36 37 38 26 36
N/A 0.053 N/A 0.324 0.324 0.567 0.324 N/A 0.000 0.165
Note: Figures in italics indicate significance at the *0.05 level
(1) (2) (3) (4) (5)
unacceptable fit; poor fit; acceptable fit; good fit; and excellent fit.
Space for comments was provided for each criterion. Data analysis and results Data preparation Fit analysis sheets were reviewed for completeness. Responses with no score or with a comment such as, “unable to see details,” “can’t see,” or “can’t tell” were re-coded to indicate missing data. Coding non-response was important to account for the missing data when conducting a Type II comparison of means. Score data for each criterion was entered into SPSS 13.0 for Windows. A second variable was created for each criterion to indicate ability to score.
Fit analysis using live and 3D scan models 93
Table I. Difference in number of dress slopers scored without problem
IJCST 22,2/3
94
Table II. Difference in number of pant slopers scored without problem
Model type Criteria OA Inseams Left-side seam Right-side seam Front crotch seam Back crotch seam Dart placement Front darts Back darts L/T Waistband Stomach Front crotch Front full hip Front thigh Back crotch Back upper hip Back full hip Back thigh
Live
Scan
P
38 37 36 38 38
16 25 28 28 27
0.000 0.000 0.006 0.001 0.001
34 36
18 22
0.000 0.000
37 37 38 38 38 38 38 38 38
36 37 38 38 38 38 38 38 38
0.567 N/A N/A N/A N/A N/A N/A N/A N/A
Note: Figures in italics indicate significance at the *0.05 level
H1: ability to score between model types Descriptive statistics were run to analyze the number of criteria scored for each model type. A two-way ANOVA was conducted for each criterion to test H1: “There is no difference in ability to assess fit analysis scores between using a live model and 3D scan model”. Dress slopers. Significant differences in ability to score between model types were seen for 54 percent of criteria (Table I). For overall alignment (OA) criteria, significant differences were seen in 75 percent of criteria, except for center back and hem. The null hypothesis was rejected in favor of the alternative: mlive – m3D-scan. For dart placement criteria, 100 percent of criteria had significant differences; therefore, the null hypothesis was rejected. For dress L/T criteria, there was a significant difference for 10 percent of the criteria: the armscye. The null hypothesis was accepted for dress L/T criteria. Pant slopers. There were significant differences in ability to score by model type for 44 percent of pant sloper criteria. Significant differences were seen with 100 percent of the OA and 100 percent of the dart placement criteria (Table II). For these groups, the null hypothesis was rejected in favor of the alternative hypothesis: mlive – m3D-scan. For the L/T group, no significant differences were seen. H0: mlive ¼ m3D-scan was accepted. H1a: difference in ability to score between garments A two-way ANOVA was conducted to test H1a: “There is no difference in ability to assess fit analysis scores between the 19 garments that were evaluated.” There was no significant difference for any criteria in ability to assess fit analysis scores between the 19 garments for dresses or pants. Therefore, H0: mGarment1 ¼ mGarment2 · · · ¼ mGarment19 was accepted.
H1b: interaction between model types and garment A two-way ANOVA was conducted to test H1b: “There is no interaction between fit analysis model type and garment evaluated.” There was no significant difference in ability to score for 96 percent of dress criteria, except L/T of back neckline ( p ¼ 0.049) and 100 percent of pants criteria caused by interaction between model type and garment. Therefore, the H0: “There is no interaction between fit analysis model and garment evaluated” was accepted for both dresses and pants.
Fit analysis using live and 3D scan models 95
H2: difference in scores between model types Mean scores were calculated for each criterion by model type. A two-way ANOVA was conducted to test H2: “There is no difference in fit analysis scores for each criterion between using a live model and 3D scan model.” Mean scores for each criterion were compared for significant differences at the 0.05 level. Dress slopers. There was a significant difference between models for 29 percent of criteria (Table III). Significantly different scores were seen at hem alignment, back bodice waist dart and skirt dart placement, and back neckline, bust, front hip, and back hip L/T criteria. The null hypothesis was accepted for 71 percent of dress criteria. Model type Criteria OA Center front Center back Left-side seam Right-side seam Waist Shoulder seam Sleeve grain Hem Dart placement Front bust darts Front skirt waist darts Back shoulder darts Back bodice waist darts Back skirt waist darts Sleeve dart L/T Front neckline Back neckline Bust Bodice back Front waist Back waist Front hip Back hip Armscye Sleeve
Live
Scan
P
3.60 3.33 2.80 3.00 2.87 3.03 3.40 3.30
4.50 4.50 4.00 4.00 4.00 4.00 4.00 2.5
0.331 0.513 0.449 0.780 0.178 0.835 0.326 0.000
2.77 3.40 3.37 3.07 3.33 2.90
4.00 3.50 4.00 4.00 3.50 4.00
0.053 0.057 0.651 0.029 0.025 0.274
2.67 3.03 2.70 2.77 2.97 2.90 3.30 3.20 2.73 2.70
2.50 3.50 3.00 4.00 4.00 4.00 2.50 2.00 4.00 4.00
0.058 0.029 0.024 0.331 0.289 0.537 0.002 0.001 0.891 0.104
Note: Figures in italics indicate significance at the *0.05 level
Table III. Difference in mean score of dress sloper criteria
IJCST 22,2/3
Pant slopers. Significant differences in fit analysis scores between using a live model and 3D scan model were seen in 44 percent of criteria. The H0: mlive ¼ m3D-scan was accepted for 80 percent of OA criteria and 100 percent of dart placement criteria. A significant difference was seen in 67 percent of the L/T criteria between live model and 3D scan model (stomach, front crotch, front full hip, front thigh, back full hip, and back thigh). The null hypothesis was rejected for L/T criteria (Table IV).
96 H2a: difference in scores between garments Means were calculated for each criterion by model type. To test H2a: “There is no difference in fit analysis scores between the 19 garments that were evaluated”; a two-way ANOVA was conducted to compare means at the 0.05 level. Dress slopers. No significant differences in scores were seen for 58 percent of dress criteria. Significant score differences were seen between garments for 38 percent of OA criteria (right-side seam p ¼ 0.012, shoulder seam p ¼ 0.005, and hem p ¼ 0.000); 33 percent of dart placement criteria (back bodice waist darts p ¼ 0.013 and sleeve dart p ¼ 0.000); and 50 percent of L/T criteria (back neckline p ¼ 0.014, bust p ¼ 0.022, bodice back p ¼ 0.017, back waist p ¼ 0.024, and sleeve p ¼ 0.002). The null hypothesis was accepted. Pant slopers. No significant differences were seen for 69 percent of criteria; therefore, the H0: mGarment1 ¼ mGarment2· · · ¼ mGarment19 was accepted. Significant differences were seen for 20 percent of OA criteria (front crotch seam alignment p ¼ 0.005) and 44 percent of L/T criteria (front crotch p ¼ 0.015, front full hip p ¼ 0.001, back crotch p ¼ 0.010, and back full hip p ¼ 0.017).
Model type Criteria
Table IV. Difference in mean score of pant sloper criteria
OA Inseams Left-side seam Right-side seam Front crotch seam Back crotch seam Dart placement Front darts Back darts L/T Waistband Stomach Front crotch Front full hip Front thigh Back crotch Back upper hip Back full hip Back thigh
Live
Scan
P
3.57 3.13 3.09 3.48 3.04
3.07 3.29 2.86 3.36 3.36
0.069 0.224 0.068 0.042 0.806
3.52 3.09
3.29 3.21
0.148 0.582
3.61 3.35 3.35 3.35 3.61 2.96 2.57 3.22 3.35
3.57 3.07 2.71 2.57 3.21 2.86 2.64 2.57 2.5
0.810 0.015 0.000 0.000 0.001 0.063 0.517 0.001 0.000
Note: Figures in italics indicate significance at the *0.05 level
H2b: interaction between model type and garment Means were calculated for each criterion for each garment by model type. To test H2b: “There is no interaction between model type and garment evaluated”; a two-way ANOVA was conducted to compare means at the 0.05 level. Dress slopers. No significant differences were seen for 96 percent of dress sloper criteria, The null hypothesis was accepted. Interaction between fit analysis model and garment evaluated was seen at hem alignment ( p ¼ 0.011). Pant slopers. No significant differences were seen for 81 percent of pant sloper criteria. The null hypothesis was accepted. Differences in scores were seen by interaction between garment and model type for front crotch seam alignment ( p ¼ 0.046); and front crotch ( p ¼ 0.026) and back crotch ( p ¼ 0.035) L/T.
Fit analysis using live and 3D scan models 97
Summary of data analysis Results of data analysis for all hypotheses are summarized in Tables V and VI. Discussion Ability to assess fit with live or 3D scan model There were missing data on many garment fit analysis sheets completed with 3D scan models. Problems were seen mainly in the OA and dart placement criteria groups. For dress slopers, 98 percent OA and 100 percent dart placement were scored with live models and only 63 percent OA and 32 percent dart placement were scored with 3D scan models. For pant slopers, results were similar, with 98 percent of OA and 92 percent of dart placement criteria scored with live models, but only 65 and 54 percent, respectively,
Criteria group All criteria OA Dart placement L/T
H1. Difference between live and scan model (%) Dresses Pants 54 75 100 10
44 100 100 0
H1a. Difference between garments (%) Dresses Pants 0 0 0 0
0 0 0 0
H1b. Interaction between model and garment (%) Dresses Pants 4 0 0 11
0 0 0 0
Note: Percentage indicates the percentage of fit criteria in that criteria group with significant differences at the *0.05 level
Criteria group All criteria OA Dart placement L/T
H2. Difference between live and scan model (%) Dresses Pants 29 13 33 40
44 20 0 67
H2a. Difference between garments (%) Dresses Pants 42 38 33 50
31 20 0 44
Table V. Summary of significant differences in ability to assess fit analysis scores
H2b. Interaction between model and garment (%) Dresses Pants 4 13 0 0
19 20 0 23
Note: Percentage indicates the percentage of fit criteria in that criteria group with significant differences at the *0.05 level
Table VI. Summary of significant differences in fit analysis scores
IJCST 22,2/3
with 3D scan models. Significant differences were seen in ability to score between models on six out of eight OA criteria (75 percent), except center back or hem. These criteria may be more visible on a scan because a ridge is caused by the zipper at center back and the hem is a strong edge. Scoring of criteria that required identification of seam lines and body landmarks was more difficult on a scan. These results are in agreement with the Ashdown et al. (2004) findings that seam lines are difficult to identify in 3D scans.
98 Differences in fit score evaluated with live or 3D scan model Overall, differences in fit scores were not significant between live and scan models. On the dress, hem alignment had a significantly different score and the mean live score was higher. This may be because a tilted hemline is more easily observed as the 3D scan is rotated to view all sides, resulting in a lower fit score. With the pants, front crotch seam alignment was significantly different in mean score between model type, between garments, and garment/model interaction. Significant differences were seen in the mean criteria score, between mean garment scores, and by garment model interaction in the L/T thigh, crotch, and full hip criteria (Table VII). Two out of three pant sloper L/T scores were higher with the live model than with the 3D scan model. This could be because pulls and wrinkles are more noticeable in a 3D scan than with a live model, resulting in a lower fit score. With a live model, surface color can be used positively to create visual illusions of a better fit, while the 3D scan image emphasizes the physical structure of the fit. These results indicate a need for awareness that fit scores may be lower when using 3D scan. Interactions between ability to score and fit score evaluated In the OA and dart placement criteria groups, where ability to evaluate with 3D scan models was poor, the scores were significantly different between model types. In the L/T criteria group where ability to evaluate with 3D scan models was good, significant differences were seen in scores between model types on 67 percent of pant sloper criteria and 40 percent of dress sloper criteria. The judges felt they could evaluate looseness and tightness on a scan, but actually scored many criteria differently. These results can guide training for evaluation of 3D scans. All judges had experience with live fit analysis; however, using the 3D scan model was a new or unfamiliar experience. Judges approached both live and scan models positively and technical help was available during the scan evaluation. There was no time limit to judging with either method.
Table VII. P-values for pant slopers OA and L/T criteria with significant differences
Full hip Front Back Crotch Front Back Thigh Front Back
Higher live score OA L/T
Differences between garments OA L/T
0.000 0.001
0.001 0.017
0.042
0.000 0.001 0.000
0.005
0.015 0.010
Garment model interaction OA L/T
0.046
0.026 0.035
Conclusions and implications While 3D scans offer the convenience of evaluating garment fit from any time or location, there are some concerns about ability to score and reliability of scores for specific fit criteria. All seam line alignment scores were significantly lower with 3D scan models than live models. Judges had significantly lower ability to score dart placement with 3D scan models than with live models. When analyzing seam and dart alignment with body landmarks is an important part of the fit analysis, judges may have difficulty with assessment. Neck There were no differences in scoring front neck, but back neck scores with 3D scan models were higher, and thus unreliable. Shoulder Judging shoulder fit is challenging because shoulder seam alignment and shoulder dart had significant problems in ability to score with 3D scan models. Sleeve/armscye The sleeve and armscye criteria, including sleeve grain, sleeve dart placement, and armscye L/T, were significantly unable to be scored with 3D scan models. Only sleeve L/T was reliable between live models and with 3D scan models. Bust There is significantly poor ability to score front bust darts from a scan and significantly different L/T scores by scan. Waist L/T at waist for dress slopers and pant slopers were evaluated without significant difference between model types. Hip Evaluation of hip and thigh criteria was not reliable. All hip and thigh L/T scores were lower with 3D scan models, except for pant sloper back upper hip. Hems Hems were able to be scored with 3D scan models as well as with live models, but the scores with 3D scan models were significantly lower. Overall, results indicate a need for awareness that fit scores may be lower when using 3D scan and that training is needed for evaluation of 3D scans. For criteria where fit can be evaluated, it may be possible to train judges to score 3D scan models as reliably as live models. Future advancements in 3D scanners, such as color, may improve the ability to see seams and darts, and thereby improve their ability to be evaluated. With current technology, there are significant differences between fit analyses with live or 3D scan models that must be considered before implementing them in research or industry settings.
Fit analysis using live and 3D scan models 99
IJCST 22,2/3
100
References Armstrong, H.J. (2006), Patternmaking for Fashion Design, 4th ed., Pearson Prentice-Hall, Upper Saddle River, NJ. Ashdown, S.P., Loker, S., Schoenfelder, K. and Lyman-Clarke, L. (2004), “Using 3D scans for fit analysis”, Journal of Textile and Apparel, Technology, and Management, Vol. 4 No. 1, pp. 1-12. Douty, H.I. (1968), “‘Visual somatometry’ in health-related research”, Journal of the Alabama Academy of Science, Vol. 39, pp. 21-34. Kohn, I.L. and Ashdown, S.P. (1998), “Using video capture and image analysis to quantify apparel fit”, Textile Research Journal, Vol. 68 No. 1, pp. 17-26. Nam, J., Branson, D.H., Cao, H., Jin, B., Peksoz, S., Farr, C. and Ashdown, S. (2005), “Fit analysis of liquid cooled vest prototypes using 3D body scanning technology”, Journal of Textile and Apparel, Technology and Management, Vol. 4 No. 3, pp. 1-13. Further reading Bye, E. and LaBat, K. (2005), “An analysis of apparel industry fit sessions”, Journal of Textile and Apparel, Technology and Management, Vol. 4 No. 3, pp. 1-5. Schofield, N.A., Ashdown, S.P., Hethorn, J., LaBat, K. and Salusso, C.J. (2006), “Improving pant fit for women 55 and older through an exploration of two pant shapes”, Clothing & Textiles Research Journal, Vol. 24 No. 2, pp. 147-60. Corresponding author Elizabeth Bye can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Automatic basic garment pattern generation using three-dimensional measurements Choong Hyo Kim, In Hwan Sul and Chang Kyu Park Department of Textile Engineering, Konkuk University, Seoul, South Korea, and
Automatic basic garment pattern generation 101 Received 28 May 2009 Accepted 4 October 2009
Sungmin Kim Faculty of Applied Chemical Engineering, Chonnam National University, Gwangju, South Korea Abstract Purpose – The purpose of this paper is to illustrate the generation of basic garment pattern using three-dimensional body measurement data. Design/methodology/approach – A pre-defined garment model is deformed using free-form deformation method and the model is flattened to generate flat patterns. Findings – The paper finds that individual basic garment patterns are automatically generated and verified to be well fit on human subjects. Research limitations/implications – The current approach is to focus on the generation of basic bodice patterns; however, other patterns can also be generated by this method by preparing more garment models. Practical implications – This method can reduce the time required to design basic patterns as well as enhance their fitness. Originality/value – The automatic generation of individually fitted garment pattern is one of the most important steps in future garment production process. Keywords Deformation, Garment industry, Textile technology, Modelling, Computer aided design Paper type Research paper
Introduction Recently, there have been many studies on the application of information technology to the production of customer-oriented fashion goods such as made-to-measure (MTM) garments. MTM garments are made based on three-dimensionally measured individual human body data through digital processes such as automatic pattern generation and virtual drape simulation. For automatic pattern generation, although a commercialized system has not yet been developed, there have been many studies of the development of garment patterns directly from three-dimensional (3D) body data. For example, optimum-fit patterns for each person were generated by drawing pattern outlines and darts directly on the surface of a parametrically deformable garment model (McCartney et al., 2000; Kim and Park, 2003). Most such 3D pattern generation methods were based on the data acquired by non-contact-type body scanners. However, it is difficult to flatten these garment models into patterns, so such methods are not suitable for practical application (Kang and Kim, 2000a, b; Kim and Park, 2003).
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 101-113 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018595
IJCST 22,2/3
102
In this study, a contact-type coordinate measurement system was used to reconstruct the body surface model with a topologically regular mesh structure. Parametric modeling was used to deform the body model into the target shape and size. For this, a free-form deformation (FFD) method was used instead of cross-section-based or morphing methods to deform the shape of the body model (Sederberg and Parry, 1986; Coquillart, 1998). The body model was projected into flat garment patterns using multiple darts, and the fit of the actual garments made from those patterns was verified on three subjects by self-sensory tests. Methodology 1. Preparation of body model 1.1 Contact-type coordinate measurement system. Generating a virtual 3D model using a non-contact measurement system includes the reconstruction of a surface model from so-called point cloud data. However, the resulting model tends to have a noisy surface and requires much post-processing to be converted into a smooth and topologically well-structured model, because it is composed of irregular triangular meshes. Therefore, non-contact-type measurement systems are not suitable for garment modeling, because smooth lines and regular mesh structure are very important when projecting the body model onto flat patterns (Gonzalez and Woods, 2003; Russ, 2007). In this study, we obtained a body model with a topologically regular surface mesh structure using a contact-type coordinate measurement system by measuring the absolute coordinate of each point on the lines drawn on a body model that are required for garment design. 1.2 Reconstruction of body model. Measured coordinate data must be converted into a parametrically deformable surface model to be used in garment pattern generation. The surface model consists of a number of triangles, and as the index of each point on a triangle has been determined previously, measured coordinate data can be converted into a 3D surface model with triangular meshes as shown in Figure 1. 2. Deformation of body model using a FFD method 2.1 Free-form deformation. FFD is a method to deform the shape of a 3D model that was developed by Sederberg and Parry (1986). This method has been widely used because the shape of a complex 3D model can be deformed easily using only a few control points, regardless of the modeling method used (Sederberg and Parry, 1986). 2.1.1 Formation of FFD lattice. An FFD lattice is a hexahedral set of control points used for the deformation of an enclosed 3D model. Control points in the lattice are usually located by B-spline interpolation as described below (Farin, 2002). An FFD lattice is defined in the Cartesian coordinate system. First, a hexahedral lattice is formed to enclose completely the 3D model being deformed. Then arbitrary numbers of control points are located along each side of the lattice at regular intervals. The distance between two adjacent control points can then be adjusted by non-uniform rational B-spline interpolation to map the proper knot vectors and control points. 2.1.2 Model deformation. Control points on the lattice are closely related to the points on the enclosed model, and the changes in their positions are directly reflected on the shape of the model. Continuity is guaranteed inside the lattice so that the deformation is smooth everywhere on the surface of the enclosed model. In addition, there are no restrictions on the movement of each lattice control point.
Automatic basic garment pattern generation 103
Figure 1. Reconstruction of 3D body model
2.2 Definition of FFD lattice on body model. 2.2.1 Formation of FFD lattice. One of the most frequently used body model deformation methods is the cross-section-based method, in which the body model is usually generated by connecting and skinning a series of cross-sectional shapes. Some representative cross-sections such as neck, bust, waist, and hip are selected, and their girths are adjusted to become those of the target body model. Girths of intermediate cross-sections between two representative cross-sections can be interpolated by a simple equation (Kang and Kim, 2000a, b). Although this method is very simple, the continuity of the model surface is not guaranteed, especially when the deformation becomes very large. In this study, the FFD method was used to deform the body model so that a continuous and differentiable surface model could be obtained by adjusting a few control points as shown in Figure 2. 2.2.2 Definition of FFD operators. Although the shape of an enclosed body model can be modified by manipulating the lattice control points, it is very difficult to adjust multiple points simultaneously to change the shape of the model along the three principal axes. Therefore, some operator functions must be defined to manipulate multiple control points. The body model is aligned as shown in Figure 2. For the definition of operators, control points with the same x coordinates are grouped as “sagittal” sections, and those with the same z coordinates are grouped as “frontal” sections, while control points with the same y coordinates are grouped as “layer” sections. Operator functions move or rotate multiple control points on specific sections according to the supplied parameter values, and the model deformation process can be managed in a simple and consistent way. The definitions of landmarks and measurement items required for the design of women’s bodice patterns are shown in Figure 3. As the landmarks were already drawn on the body model and measured by the coordinate measurement system, the geodesic distance between two landmarks can be calculated by considering the indices of
IJCST 22,2/3
104
y
x
Figure 2. Deformation of a 3D body model using the FFD method
z
3
1 2 4
h d e 5
6 a
b
c
f
7
8
Figure 3. Definition of body points and sizes for bodice formation
g
Body points for bodice deformation
Body sizes for bodice deformation
1. Cervicale 2. Anterior neck 3. Lateral neck 4. Acromion 5. Mesosternal 6. Nipple 7. Anterior waist 8. Posterior waist
a.Waist front length b. Waist backlength c. Bust point-bust point d. Neck base circumference e. Chest circumference f. Bust circumference g. Waist circumference h. Shoulder angle
two points on the body model. To change the size of the body model accurately, it is important to know the relationship between the parameters supplied to the operators and the actual changes in model geometry caused by the operators. Assuming that the changes in measurement items are independent of each other, the relationship between the input and output of an operator can be obtained by regression analysis. For example, a linear regression equation can be obtained by supplying a parameter value ranging from 250 to 50 mm to an operator and observing the resulting size change in the body model. Once the regression equation is established, the parameter value required to invoke a certain size change on the body model can be calculated by the inverse of that equation. According to the experiments made in this study, the actual changes made to the body model by each operator had good correspondence with the initially intended values, with R 2 values greater than 0.98. 3. Flat pattern generation 3.1 Definition of darts. Darts are necessary to form 3D garments from two-dimensional (2D) patterns and vice versa. Darts can have various widths, lengths, and shapes such as linear or curvilinear. There can be more than one dart in a pattern and usually the location of each dart is pre-determined according to the overall style of the garment. Potential darts and their positions for a bodice are shown in Figure 4. Darts on the body model can be defined as follows. First, a dart is marked on the actual body model using marking tape. Then the indices of the start and end points of the dart, and the end points of each segment intersecting with the dart line, are recorded. Finally, the shortest path for the dart can be obtained by finding and connecting all the “best” intersection points of each segment. The best intersection point of a segment can be located by finding the t-value that minimizes the total distance from the start point to the end point through the intersection point, where the t-value ranges from 0 to 1. Therefore, the shortest path for a dart can be determined by finding all the t-values, as shown in Figure 5. Once the dart path is determined, the triangular elements on the surface model must be restructured, as shown in Figure 5. If the incising line passes through one of the three points of a triangle, the triangle is divided into two triangles, while the triangle is divided into three triangles if the line intersects with two sides of the triangle as shown in Figure 5. 3.2 Projection of the body model. When projecting a 3D body model onto a 2D plane to form 2D garment patterns, some distortions are unavoidable, as shown in Figure 6. This distortion can be reduced by equalizing the lengths of each side on the triangular elements in 3D and 2D iteratively. A stopping criterion for this iterative process is defined in equation (1): 3D N X 3 2D X Lij 2 Lij C Strain ¼ 100 £ ð1Þ ð%Þ; L2D ij i¼1 j¼1 where N is the number of triangles, and Lij is the 2D and 3D length of the j-th side of the i-th triangle. The equalization process is stopped when the CStrain value becomes smaller than a pre-defined threshold value. Finally, the 3D body model is projected into flat patterns. 3.3 Generation of pattern outlines. As the projected patterns consist of triangular elements, it is necessary to generate a closed outline for subsequent pattern manipulation in garment production processes. An outline can be determined by finding every border
Automatic basic garment pattern generation 105
IJCST 22,2/3
b
a
3
d
2
4
106
1
5 c
6 7 11
8 9
Front
Figure 4. Definitions of potential darts for a bodice
1. Center front neck point 2. Neck line 3. Shoulder line 4. Shoulder point 5. Arm hole 6. Under arm
10
Back 7. Low under arm 8. Waist 9. Center front waist 10. Center front line 11. Bust point
a. Shoulder line b. Neck line c. Waist d. Arm hole
edge that has only one neighboring triangle and connecting the end points of all those border edges in the correct order, as shown in Figure 7(b). When the included angle at a point on that outline differs from 1808 by more than a pre-defined threshold value, the point can be regarded as a “corner” point. The outline can then be approximated using connected multiple smooth B-spline curves segmented by those corner points, as shown in Figure 7(c) (Farin, 2002). Cutting lines can be generated by duplicating and locating each segment at a certain offset called the ease from the original outline as shown in Figure 7(d). Experimental 1. Measurement of body model 1.1 Preparation of body model. In this study, a bodice model was chosen for automatic pattern generation, because it has the most complex shape among garment models
n = Last node n–1
Automatic basic garment pattern generation
t = 0~1
n–2
107 t = 0~1 0 = First node
(a)
(b) Notes: (a) Shortest path search algorithm for dart generation; (b) generated dart
Figure 5. Schematic diagram of dart generation
Figure 6. Schematic diagram of 2D projection
including neck, armhole, bust, and waist lines. For this reason, the design of an accurate bodice pattern is the most difficult process in garment design, and therefore bodice modeling is suitable for the verification of the performance of a newly developed method.
IJCST 22,2/3
108
Boundary edge
Common edge
Figure 7. Schematic diagram of sewing line generation
(a)
(b)
(c)
(d)
Notes: (a) Path search algorithm; (b) extraction of closed paths; (c) B-spline approximation; (d) cutting line generation
1.2 Measurement device. A contact-type MicroScribe G2 desktop digitizing system from Immersion Corporation was used to measure the surface geometry of a bodice model, as shown in Figure 8. Rhinoceros 3.0 modeling software was used to convert the measured data into an ASCII-format data file. 1.3 Definition of surface mesh structure. A regular mesh structure was drawn on the model using triangles and quadrilaterals using 1.0 mm wide marking tape. The lengths of sides ranged from 1 to 3 cm according to the complexity of the local surface shape. The coordinates of each mesh point were measured and each point was numbered for use in the formation of the triangular mesh structure. The index assigned to each point was also used in the measurements as well as in the definition of darts.
Automatic basic garment pattern generation 109
Figure 8. Multiple joint-type 3D coordinate measurement system
2. Pattern generation and trial test 2.1 Preparation of subjects. In this study, three human subjects were chosen to verify the pattern generation process using a self-sensory test with real garments. The bust girths of subjects ranged from 82 to 85 cm and waist girths from 67 to 70 cm. The basic measurement results of subjects’ upper bodices are shown in Table I. 2.2 Deformation of body model and generation of patterns. A five-layer FFD lattice was defined around the reconstructed bodice model. The layers were numbered from 0 to 4 from the bottom. Layer 1 is located at the under-bust level, 2 at the bust, and 3 at the front neck. Body deformation rules were established and calibrated by FFD operators using the actual measurement data. The body model was deformed to fit each subject and projected into flat patterns using multiple combinations of potential darts. In this study, the combination of front shoulder line, front waist, back waist, and back armhole dart seemed to be the best one. Projected patterns were plotted and cut by Mimaki CG-100AP apparel cutting plotter with cutting and sewing lines. 2.3 Trial test. Plotted patterns were sewn into real garments using white cotton fabric with low extensibility. A zipper was added on the back for wearability and a constant ease was added for each pattern for minimum comfort of subjects. A private fitting room with a full-body mirror was prepared so subjects could assess the fit of the garments.
Items Neck girth Waist girth Bust girth Upper bust girth Back length Front length Note: Unit: mm
1
Subjects 2
3
380 700 845 850 390 340
370 670 820 820 385 330
370 680 835 840 395 345
Table I. Upper body sizes of subjects
IJCST 22,2/3
2.4 Self-sensory test for fitting evaluation. A self-sensory questionnaire test was performed with 12 questions on fitting evaluation. Subjects stood relaxed with the garment on and answered each question using a satisfaction index ranging from 1 to 5. The results of the self-sensory test were analyzed by the SPSS statistical analysis software package using t-tests and one-way ANOVA tests.
110
Results and discussion The garment generated from the initially measured model is shown in Figure 9(a). A garment generated for a human subjected is shown in Figure 9(b).
(a)
Figure 9. Try-on test
(b) Notes: (a) Dummy model; (b) human subject
The results of the self-sensory test are shown in Table II. As shown in the results, Subject 2 felt most uncomfortable with the location of the neck circumference and the overall appearance (Q12) for appearance evaluation. In the comfort evaluation, Subject 2 felt uncomfortable with the neck circumference (Q5), shoulder width (Q6), front armhole width (Q7), front center length (Q9), and waist girth (Q11). It is thought that this discomfort was caused by the inappropriate location of darts after the large deformation of the original model, as Subject 2 had the smallest bust and waist sizes among the subjects. However, all the subjects felt approximately the same degree of comfort with the location of the other girths (Q2, Q3, and Q4) and the comfort on the back (Q8, Q10). Considering the overall results, Subject 2 differed from Subjects 1 and 3 ( p , 0.001). Subjects 1 and 3 felt comfortable with the garment, but the discomfort of Subject 2 indicates that a comprehensive experiment on the definition of the body model deformation rule seems to be necessary for the method to be applicable to various subjects. The results of the virtual try-on are shown in Figure 10. The development of a drape simulator is another research topic in our laboratory, and it could become an efficient tool for garment design if the pattern generation system developed in this study were to be integrated into the drape simulation system.
Automatic basic garment pattern generation 111
Conclusion The 3D pattern generation method is an innovative method by which optimum-fitting patterns can be obtained easily without trial-and-error-based traditional grading methods. We developed an automatic pattern generation system. A body surface model was reconstructed from 3D human body data measured by a contact-type coordinate measurement system. With this method, a surface model with a regular mesh structure could be generated, and its topological information could be used for efficient flat-pattern development. An FFD-based method was used to change the size and shape of the body model to match the bodies of human subjects. Finally, the body model was projected into flat garment patterns incorporating multiple darts. The fit of
1 Self-sensory question 1. Location of neck circumference 2. Location of upper bust circumference 3. Location of bust circumference 4. Location of waist circumference 5. Neck circumference 6. Shoulder length 7. Front body width at armpit level 8. Back body width at armpit level 9. Front center line length 10. Back center line length 11. Waist circumference 12. Overall appearance Average
M b
5.0 4.7 4.7 5.0 5.0b 4.3b 4.0b 5.0 5.0b 4.3 4.7b 4.3b 4.7b
SD 0.00 0.58 0.58 0.00 0.00 0.58 0.00 0.00 0.00 0.58 0.58 0.58 0.09
Subjects 2 M SD a
3.6 3.0 3.3 4.0 3.3a 2.3a 2.3a 4.3 3.3a 4.0 4.0a 2.7a 3.4a
0.58 1.00 0.58 1.00 0.58 0.58 0.58 0.58 0.58 0.00 0.00 0.58 0.05
3 M a
4.0 4.0 4.3 4.7 4.3b 3.7b 2.7a 4.7 4.7b 4.3 5.0b 4.7b 4.3b
SD
p-value
0.00 0.00 0.58 0.58 0.58 0.58 0.58 0.58 0.58 0.58 0.00 0.58 0.29
0.007 * * 0.058 0.068 0.252 0.014 * 0.014 * 0.011 * * 0.296 0.011 * 0.630 0.027 * 0.011 * 0.000 * * *
Notes: Significant at: *p , 0.05, * *p , 0.01, and * * *p , 0.001, respectively; superscripts were applied where a significant difference was observed after SNK test (a , b)
Table II. The results of the self-sensory test
IJCST 22,2/3
112
Figure 10. Virtual try-on test using 3D drape simulator
garments made from those patterns was verified on three human subjects by self-sensory tests that showed a good correspondence. In this method, changes made on the garment model such as dart insertion are instantaneously reflected on the resulting patterns, so designers can make customer-oriented MTM garments easily. Further research will be made on the practical application of this method on various garments through the construction of a database for model deformation rules. References Coquillart, S. (1998), “Extended free-form deformation: a sculpturing tool for 3D geometric modeling”, ACM SIGGRAPH Computer Graphics, Vol. 24 No. 4, pp. 187-96. Farin, G. (2002), Curves and Surfaces for CAGD: A Practical Guide, 5th ed., Morgan Kaufmann, San Francisco, CA.
Gonzalez, R.C. and Woods, R.E. (2003), Digital Image Processing, 2nd ed., Prentice-Hall, New York, NY. Kang, T.J. and Kim, S. (2000a), “Development of three-dimensional apparel CAD system”, International Journal of Clothing Science & Technology, Vol. 12 No. 1, pp. 39-49. Kang, T.J. and Kim, S. (2000b), “Optimized garment pattern generation based on three-dimensional anthropometric measurement”, International Journal of Clothing Science & Technology, Vol. 12 No. 4, pp. 240-54. Kim, S. and Park, C.K. (2003), “Fast garment drape simulation using geometrically constrained particle system”, Fibers and Polymers, Vol. 4 No. 4, pp. 169-75. McCartney, J., Hinds, B.K., Seow, B.L. and Gong, D. (2000), “An energy based model for the flattening of woven fabrics”, Journal of Materials Processing and Technology, Vol. 107, pp. 312-18. Russ, J.C. (2007), The Image Processing Handbook, 5th ed., CRC Press, New York, NY. Sederberg, T.W. and Parry, S.R. (1986), “Free-form deformation of solid geometric models”, ACM SIGGRAPH Computer Graphics, Vol. 20 No. 4, pp. 151-60. Further reading Kim, S. and Kang, T.J. (2002), “Garment pattern generation from body scan data”, Computer-Aided Design, Vol. 35, pp. 611-8. Corresponding author Sungmin Kim can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
Automatic basic garment pattern generation 113
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
IJCST 22,2/3
Reconstruction of individualized dress forms using parameterized silhouettes
114
Xinrong Hu
Received 29 June 2009 Accepted 11 October 2009
Department of Computer Science, Wuhan University of Science and Engineering, Wuhan, China and School of Human Ecology, University of Texas at Austin, Austin, Texas, USA, and
Bugao Xu School of Human Ecology, University of Texas at Austin, Austin, Texas, USA Abstract Purpose – The purpose of this paper is to develop a fast parameterized modeling approach to generate individualized dress forms for realistic human bodies. Design/methodology/approach – An individualized dress form is created by deriving a new set of fitting functions from a number of key existing dressing parameters and pre-defined templates. The fitting functions only contain simple shapes of circular and/or elliptical arcs, which can be modified computationally based on a few personal dressing data. Findings – This paper reaffirms that individual body shape can be adequately described by a number of critical cross-section silhouettes, and a personalized dress form can be constructed based on key dressing parameters and templates. Originality/value – The fitting functions and relevant dressing data for specific cross-sectional silhouettes are determined, permitting a user to create personalized dress forms only by inputting a simple set of dressing parameters. Keywords Computer aided design, Fashion design, Modelling Paper type Research paper
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 114-126 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018603
1. Introduction Human body modeling has been one of the challenging tasks encountered by computer graphics researchers and fashion designers. Many parameterized modeling techniques have been applied for creating body models (Rose et al., 1998). In general, these techniques can be classified into two categories: one that models body surface from a sample of scanned data (Lewis et al., 2002; Blanz and Vetter, 1999; Kry et al., 2002; Sloan et al., 2001; Allen et al., 2002; DeCarlo et al., 1998; Dekker et al., 1999; Ju et al., 2000; Yu and Xu, 2008) and the second that models body surface from selected key feature parameters (Seo and Magnenat-Thalmann, 2003). Lewis et al. (2002) proposed “pose space deformation,” approaching the problem of geometric model deformation by using the scanned elastic surface of varying postures and blending them during the animation. Blanz and Vetter (1999) presented a “morphable face model” for manipulating an existing model according to changes in certain facial attributes. New faces were modeled by forming linear combinations of
the prototypes that were collected from 200 scanned face models. Manual assignment of attributes was used to define shape and texture vectors that, when added to or subtracted from a face, manipulated a specific attribute. Recently, Kry et al. (2002) proposed another extension modeling technique based on principal component analysis, allowing for optimal reduction of the data and thus expediting the modeling. Sloan et al. (2001) applied radial basis function for blending example facial models and the arm. Allen et al. (2002) presented another example-based method for creating realistic skeleton-driven deformation. DeCarlo et al. (1998) reduced the problem of generating simple face geometries into the problem of generating sets of anthropometrical measurements by using various modeling techniques. However, the modeling process needed minutes of calculation to produce a face model corresponding to the given measurement set. Dekker et al. (1999) used a series of anatomical assumptions in order to optimize, clean and segment data from a whole body range scanner to generate quad patch representations of human bodies and built applications for the clothing industry. Ju et al. (2000) introduced an approach to automatically segment the scan model to conform it to an animatable model. Yu and Xu (2008) presented an effective algorithm for reconstruction of the human body with the data from the two-view body scanner. Seo and Magnenat-Thalmann (2003) introduced a parameterized body modeling technique using key sizing parameters. They employed a two-phase algorithm to conduct global and local mapping, and used radial basis interpolation to form the body shape. Their proposed scheme took unorganized scanned data from realistic human body to generate appropriate shape and proportion of the body by forming deformation functions from the input parameters. A static body model is formed by altering the control vertices of a number of key template silhouettes with the user’s dressing parameters. The premise for this modeling concept is that the topology of the model is a known priori and shared by other resulting models. For dressing purposes, a strict fidelity model is not necessary for each user. An individualized dress form often suffices the need for fast visualization in virtual clothing. This simplification can tremendously reduce the needed measurements and make the modeling nearly real-time, which is critical for implementing online virtual try-on. 2. Parameterized silhouettes Since the head, hands, and feet of a person do not have direct effects on his/her dressing body appearance, most dress forms include only the torso and legs. The proposed modeling scheme for individualized dress forms will use regular dressing parameters, such as stature, shoulder width, bust girth, waist girth, hip girth, and the other correlative dressing sizes, to calculate the parameters necessary for constructing simplified silhouettes. To meet the requirements of the dressing, we can reconstruct a dress form by selecting some critical cross-sectional silhouettes as shown in Figure 1. The selection principle is that these silhouettes not only define the basic body shape needed for the dressing, but also can be reconstructed from the dressing parameters quickly. In order to achieve these objectives, we sampled some people to form the set of human body dressing parameters that can be used to model the template model shape. 2.1 Shoulder silhouette The critical parameters that determine the shoulder silhouette are shoulder breadth, depth and the height from heel to shoulder. It is also important to find the relationships
Reconstruction of individualized dress forms 115
IJCST 22,2/3
Shoulder Bust
116
Waist
Figure 1. Key cross-sectional silhouettes of a human body
Hip
Thigh
between these parameters and stature. The actual shoulder silhouettes for a male and a female are shown in Figure 2. These shoulder shapes may be represented with a simple shape shown in Figure 3. The simple shape consists of four connected geometrical arcs, two circular arcs and two elliptical arcs that are divided by the two blue lines. The key technique for re-constructing this shoulder shape is to solve for 20 selected vertices that determine two elliptical and two circular arcs. Seven vertices are extracted from each circular arc and three vertices from each elliptical arc. To solve for all these vertices, the vertices connecting a circle and an ellipse should be solved first, that is, to solve u1 as shown in Figure 3. The coordinates of a vertex on the ellipse arc can be represented: x ¼ C s*W s*cosðuÞ;
Figure 2. Actual shoulder shapes
y ¼ H s;
(a)
z ¼ C s*Ds*sinðuÞ þ Ds
(b)
Notes: (a) Male; (b) female
Ws
Ds
θ3
Figure 3. Fitted shoulder silhouette
θ2
θ1
ð1Þ
Here, u is inclined angle between the line that links the vertex on the ellipse arc and ellipse center and the horizontal line; Ws is shoulder breadth; Ds is shoulder depth; Hs is shoulder height from heel to shoulder; Cs is the weighted coefficient of the long axis and the short axis. The coordinates of vertex on circle arc can be represented: x < 0:5* ðW s 2 Ds*cosðuÞÞ;
y ¼ H s;
z < 0:5* Ds*sinðuÞ
ð2Þ
Reconstruction of individualized dress forms 117
Here, u is inclined angle between the line that links the vertex on the circle arc and circle center and the horizontal line. Then the shoulder shape silhouette is modeled by linking these vertices. 2.2 Bust silhouette Figure 4 shows two bust silhouettes. The critical parameters that determine the bust silhouette are bust girth and the height from heel to bust. For the dressing purpose, the bust silhouette can be modeled by the shape shown in Figure 4(c). Since the bust shapes of a male and a female are different, the depth of the simulating curve should be adjustable to reflect the difference. We propose to use four tangent ellipses to construct the desired simulating curve as shown in Figure 4(c). The four ellipses are labeled as A, B, C, and D, respectively. The medium bold curve in Figure 5 can be seen as the simulated curve of the bust silhouette, the shape of which can be manipulated as follows: . the depth of the front bust can be changed by adjusting the abscissa axes of ellipses A, B and D; . the smoothness of the back bust can be changed by adjusting the abscissa axes of ellipses A, B and C; . the bust width can be changed by adjusting the abscissa axes of ellipses A and B; and . the bust depth can be changed by adjusting the ordinate axes of ellipses A and B. It is important to locate the four tangent vertices generated by ellipses C or D and A or B. Let us assume that: . W and H represent the half abscissa axis and half ordinate axis of ellipses A or B, and they determine the bust width and depth;
(a)
(b)
Notes: (a) Male; (b) female; (c) simulated
(c)
Figure 4. Bust shapes
IJCST 22,2/3
C
CAL
AL
H
118
W
A
B
2R
D
DAL
Figure 5. Tangent ellipses and correlative parameters .
.
.
.
CW and CH represent the half abscissa axis and half ordinate axis of ellipse C, and they determine the width and depth of the back bust; DW and DH represent the half abscissa axis and half ordinate axis of ellipse D, and they determine the width and depth of the front bust; HYC is the distance between the original point and the center of ellipse C, and HYD is the distance between the original point and the center of ellipse D; and R is the half distance between the centers of ellipses A and B.
The formulas of ellipses C and B are: x2 ð y 2 HYCÞ2 ðx 2 RÞ2 y 2 þ ¼ 1:0; and þ 2 ¼ 1:0: CW2 CH2 W2 H From equations (3) and (4), we have:
›ðx 2 =CW2 þ ð y 2 HYCÞ2 =CH2 Þ ›ððx 2 RÞ2 =W 2 Þ þ y 2 =H 2 Þ ¼ : ›y ›y The coordinates of the tangent vertices between ellipses C and B are: 8 R > > ; if R ¼ W > > < 2:0 x¼ > ðR * CW2 2 W * CW * RÞ > > ; if R – W > : ðCW2 2 W 2 Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi H * W 2 2 ðx 2 RÞ2 y¼ W
ð3Þ
ð4Þ
ð5Þ
We can also calculate: rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 HYC ¼ y þ 1:0 2 * CH; CW2
ð6Þ
and two angles:
CAL ¼ arctan j
y R2x
j ;
DAL ¼ arctan j
y R2x
j
ð7Þ
Thus, when HB (the height from the heel to the middle bust) is known, the tangent vertices between two of ellipses A, B, C or D can be calculated with these parameters as follows. The tangent vertices of ellipses A and C: x ¼ W · cosðCALÞ 2 R;
y ¼ H B;
z ¼ H · sinðCALÞ:
Reconstruction of individualized dress forms 119
ð8Þ
The tangent vertices of ellipses B and C: x ¼ 2W · cosðCALÞ þ R;
y ¼ H B;
z ¼ H · sinðCALÞ:
ð9Þ
z ¼ H · sinðDALÞ:
ð10Þ
The tangent vertices of ellipses A and D: x ¼ W · cosðDALÞ 2 R;
y ¼ H B;
The tangent vertices of ellipses A and C: x ¼ 2W · cosðDALÞ þ R;
y ¼ H B;
z ¼ H · sinðDALÞ:
ð11Þ
2.3 Waist silhouette The critical parameters that determine the waist shape are the waist girth and height from heel to waist. The actual waist shapes are shown as in Figure 6, and can be modeled with two merged super-ellipses (Xiao, 2006). The general formula of a super-ellipse is: xn yn þ n¼1 n W H
ð12Þ
Here, W and H are the half abscissa axis and the half ordinate axis of the ellipse, respectively. If a vertex on the super-ellipse is represented with polar coordinates, that is: x ¼ r ðaÞ cosðaÞ;
y ¼ r ðaÞ sinðaÞ
Then, the super-ellipse can be written as: r ðaÞ cosðaÞ n r ðaÞ sinðaÞ n þ ¼ 1: W H
(a) Notes: (a) Male; (b) female
(b)
ð13Þ
ð14Þ
Figure 6. Actual waist shape
IJCST 22,2/3
120
Hence: rðaÞn ¼
1 : ðcosðaÞ=W Þn þ ðsinðaÞ=H Þn
ð15Þ
Figure 7 shown super-ellipses with different n-values. We choose the super-ellipse with n being three to represent the waist silhouette. These two super-ellipses used to simulate the waist silhouette can be represented as: ( x ¼ ðW þ c * sinðaÞÞ * cosðaÞ ; ð16Þ y ¼ ðH þ d * cosðaÞÞ * sinðaÞ W and H are the waist width and depth, and c and d are the coefficients that are used to adjust W and H. The waist girth can be computed with trapezoid element integral method. The vertices on the front super-ellipses are: x ¼ ðW 2 WR · sinðaÞÞ · cosðaÞ;
y ¼ H w;
z ¼ ðH þ FHR · cosðaÞÞ · sinðaÞ
ð17Þ
The vertices on the back super ellipses are: x ¼ ðW 2 WR† sinðaÞÞ · cosðaÞ;
y ¼ H w;
z ¼ ðH þ BHR · cosðaÞÞ · sinðaÞ ð18Þ
In equations (17) and (18), WR is the weighted coefficient of horizontal axis and Hw is the height from the heel to the waist. FHR is the weighted coefficient of vertical axis on the front super-ellipses and BHR is the weighted coefficient of vertical axis on the back super-ellipses. The bust shape silhouette can be modeled by linking these vertices. 2.4 Hip and thigh silhouette Figure 8 shown the silhouettes of two actual hip shapes. The critical parameters that determine the hip shape are the hip girth and height from heel to hip. 3 2 1
π 2
y H a
x (α) π
O
α
W x
Figure 7. Super-ellipses with different n
4 5 3π 2
The real hip shape can be modeled with the shape shown in Figure 9. Similar to waist silhouette model, the simulated hip silhouette is irregular, and it can also be modeled with two super-ellipses. The two super-ellipses are merged to construct the simulated hip silhouette curve. In fact, the hip girth size can be made up with a line and a curve as shown in Figure 9. The line size depends on the two topmost vertices. The coordinates of the two vertices can be determined from simple mathematical relationships. Let x denote the sine value of the included angle a1 formed by the topmost vertex and the center of the hip. Then: pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2W þ W 2 þ 8 · WR2 : ð19Þ sin a1 ¼ 4 · WR Similarly, y is related with the cosine value of the included angle a and: pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2H þ H 2 þ 8 · BHR2 : cos b1 ¼ 4 · BHR
Reconstruction of individualized dress forms 121
ð20Þ
The coordinates of a vertex of the simulated hip silhouette are as follows: x ¼ ðW þ WR · sinða1ÞÞ · cosða1Þ;
y ¼ ðH þ BHR · cosðb1ÞÞ · sinðb1Þ:
ð21Þ
In equation (21), W and H are the half horizontal and vertical axes of the super-ellipses, WR is the weighted coefficient of horizontal axis. In equations (22) and (23), FHR is the weighted coefficient of vertical axis on the front super-ellipses, and BHR is the weighted coefficient of vertical axis on the back super-ellipses.
(a)
Figure 8. Actual hip shape
(b)
Notes: (a) Male; (b) female β1
α1
α2 β2
Figure 9. Simulated hip silhouette
IJCST 22,2/3
The vertices on the front super-ellipses are: x ¼ ðW 2 WR · sinða1ÞÞ · cosða1Þ; y ¼ H H;
ð22Þ
z ¼ ðH þ FHR · cosðb1ÞÞ · sinðb1Þ
122
The vertices on the back super-ellipses are: x ¼ ðW 2 WR · sinða2ÞÞ · cosða2Þ; y ¼ H H;
ð23Þ
z ¼ ðH þ BHR · cosðb2ÞÞ · sinðb2Þ In equations (22) and (23), HH is the height from the heel to the hip. The bust shape silhouette can be modeled by linking these vertices and solving the hip girth. We can model the thigh root silhouettes (shown in Figure 10) with two super-ellipses in a similar way described above. 3. Human body shape modeling 3.1 Template model The template model of a human body is determined by the dressing parameters and is composed of triangular meshes, as shown in Figure 11. A coarse template model is made of 11 cross-section silhouettes and 16 vertices on each silhouette. In order to increase the detail of the surface fitting, each triangular patch can be subdivided into four triangles. Each newly generated vertex is mapped to the modeled surface and a displacement vector is recalculated. The refined template model improves the resolution with Loop (1987) subdivision. All cross sections are hierarchically linked so that the body shape can be preserved in any transformation (scaling, rotation, and translation). 3.2 Feature parameters and critical cross sections General template models can be generated using the standard male and female size parameters specified in the Chinese Body Size Criterion (General Administration of Quality Supervision , Inspection and Quarantine of China, 1989). Body configuration can be defined by the cross sections and the parameters that form the approximate silhouettes of these cross sections. By applying the user’s personal body parameters to the general template model, one can obtain a new model describing the personal body shape efficiently. The feature parameters to be used should be those familiar to the
Figure 10. The thigh root silhouette
(a) Notes: (a) Male; (b) female
(b)
Reconstruction of individualized dress forms 123
(a)
(b)
Notes: (a) Male; (b) female
user. In addition to gender and age range, we include the parameters, such as stature, shoulder width, shoulder depth, bust girth, waist girth, hip girth, thigh girth and the height at these landmarks, in the modeling. 3.3 Runtime results The Wolf dress forms of sizes eight and ten (Wolf Dress Forms, 2009) were scanned with our body scanner (Xu et al., 2002) and reconstructed with the methods presented above after their dressing parameters were supplied (Table I). A total of 13 regular dressing parameters were used to reconstruct these two standard mannequins. The experimental results comparing individualized body models with the actual dressing forms are shown in Figure 12. It can be easily seen that our system faithfully reproduces models that are consistent with the input parameters familiar to the user. 4. Summary In this paper, we presented a geometric solution for modeling dress forms based on key dressing parameters. The modeling utilized only a few cross-sectional silhouettes that are critical to defining individual body shape, and the pre-defined body templates. The silhouettes at important body landmarks were constructed by using simple shapes, such as circle, ellipse, and super-ellipse arcs, and the dressing parameters. The benefits of using this dress form modeling approach are that a realistic and individualized body model can be produced quickly based on straightforward measurements of the body and the pre-defined template model. In the near future, we plan to utilize more accurate dressing parameters obtained from the 3D body scanner, and to extend this framework to head, hands, and feet modeling with skin color and texture rendering.
Figure 11. Template models with different resolutions
37 38.5
8 10
Table I. Parameters of the dress forms (cm)
Shoulder width
13 13.5
Shoulder depth 135 137.5
Shoulder height 87 91
Bust girth 64 66.5
Waist girth 92 94
Hip girth 32.5 34.5
Thigh length 45 46.5
Calf length
124 126
Armpit height
118.5 121.5
Bust height
101.5 103.5
Waist height
82 85
Hip height
124
Size
75.5 76
Thigh height
IJCST 22,2/3
Reconstruction of individualized dress forms 125
(a)
(b)
(c)
(d)
Notes: (a) and (c) scanned; (b) and (d) simulated
References Allen, B., Curless, B. and Popovic, Z. (2002), “Articulated body deformation from range scan data”, Proceedings of the 2002 SIGGRAPH, Addison-Wesley, New York, NY, pp. 612-19. Blanz, B. and Vetter, T. (1999), “A morphable model for the synthesis of 3D faces”, Proceedings of the SIGGRAPH’99, Addison-Wesley, New York, NY, pp. 187-94. DeCarlo, D., Metaxas, D. and Stone, M. (1998), “An anthropometric face model using variational techniques”, Proceedings of the SIGGRAPH’98, Addison-Wesley, New York, NY, pp. 67-74. Dekker, L., Douros, I., Buxton, B.F. and Treleaven, P. (1999), “Building symbolic information for 3D human body modeling from range data”, Proceedings of the Second International Conference on 3-D Digital Imaging and Modeling, IEEE Computer Society, Washington, DC, pp. 388-97. General Administration of Quality Supervision, Inspection and Quarantine of China (1988), GB-10000-88 Human Dimensions of Chinese Adults, Chinese Standard Press, Beijing. Ju, X., Werghi, N. and Siebert, J.P. (2000), “Automatic segmentation of 3D human body scans”, Proceedings of the IASTED International Conference on Computer Graphics and Imaging 2000 (CGIM 2000), Las Vegas, NV, USA. Kry, P.G., James, D.L. and Pai, D.K. (2002), “EigenSkin: real time large deformation character skinning in graphics hardware”, ACM SIGGRAPH Symposium on Computer Animation, San Antonio, TX, pp. 153-9. Lewis, J.P., Cordner, M. and Fong, N. (2002), “Pose space deformations: a unified approach to shape interpolation and skeleton-driven deformation”, Proceedings of the 2000 SIGGRAPH, Addison-Wesley, New York, NY, pp. 165-72. Loop, C. (1987), “Smooth subdivision surface based on triangles”, Master’s thesis, Department of Mathematics, University of Utah, Salt Lake City, UT.
Figure 12. Dress forms of sizes eight and ten
IJCST 22,2/3
126
Rose, C., Cohen, M. and Bodenheimer, B. (1998), “Verbs and adverbs: motion interpolation using RBF”, IEEE Computer Graphics and Applications, Vol. 18 No. 5, pp. 32-40. Seo, H. and Magnenat-Thalmann, N. (2003), “An automatic modeling of human bodies from sizing parameters”, ACM SIGGRAPH Symposium on Interactive 3D Graphics, ACM, New York, NY, pp. 19-26. Sloan, P.P., Rose, C. and Cohen, M. (2001), “Shape by example”, ACM Symposium on Interactive 3D Graphics, Washington, DC, pp. 135-43. Wolf Dress Forms (2009), available at: www.wolfform.com, June. Xiao, L. (2006), “The Theory and Implement of Super-ellipse Curve Code System”, Economic Management Press, Beijing. Xu, B., Huang, Y. and Yu, W. (2002), “A 3D body scanning system for apparel mass customization”, Optical Engineering, Vol. 41 No. 7, pp. 1475-9. Yu, W. and Xu, B. (2008), “Surface reconstruction from two-view body scanner data”, Textile Research Journal, Vol. 5 No. 5, pp. 1-10. Corresponding author Xinrong Hu can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Style previewing in 3D using name-based sewing rules
Style previewing in 3D
In Hwan Sul College of Engineering, i-Fashion Technology Center, Konkuk University, Seoul, South Korea
127 Received 18 February 2009 Accepted 28 May 2009
Abstract Purpose – The purpose of this paper is to combine patterns from different garment sets and preview garment styles in 3D apparel design by giving sewing names to patterns and sewing edges. Design/methodology/approach – A new rule for 3D garment sewing is made. Unlike conventional vertex number-based method, patterns and their edges are given specific names. If two edges have a same edge name, they make a sewing line. Thus, patterns from different garments can be combined and draped with this method. Numbers of boundary mesh nodes were controlled using B-Spline to combine sewing edges of different lengths. Findings – It is found that by only assigning names to patterns and sewing edges, garment style can be previewed by substituting patterns. Originality/value – Styles and details of garments can be previewed in 3D by mixing patterns of different garment sets like in 2D technical flat sketching. Even patterns with different edge lengths can be combined by controlling the pattern meshes using B-Spline. Keywords Customization, Computer aided design, Garment industry, Fashion design Paper type Research paper
1. Introduction The development of computer hardware and software enabled 3D apparel computeraided design (CAD) system to visualize virtual garment from 2D pattern data effectively. The same technology is used also for animations and special effects in movies. The 3D virtual garment can be transferred via internet and such virtualization technology can promote business-to-business or business-to-customer industry. The most eminent benefit from using 3D apparel CAD is that we can see the garment design without making it in reality. But such an advantage is limited to a single garment set. The 3D CAD operation is similar to real garment design/sewing/stitching so only one style of garment (including pattern grading) is generated each time. The conventional sewing operation generates pairs of mesh nodes to sew. The list of sewing node pairs is used as sewing constraints in cloth simulation, thus combining patterns into one in 3D during drape simulation. But such sewing information is specific to each garment. Thus, patterns from different garment sets cannot be mixed. It is because the sewing information is based on the boundary mesh node ID’s. Meanwhile, when the garment designer draws technical flats in 2D, the designer can select any kind of collars, cuffs, and arm styles from previously stored database and generate a new design. Until now there was no way to implement this feature in 3D garment CAD system. Therefore, we propose new sewing rules based on edge names for 3D apparel CAD. The idea is to give standard names to every edge and record the sewing information with respect to the edge names, not to the mesh node ID’s. The advantage is that once
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 127-144 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018612
IJCST 22,2/3
the garments are given names, their detail patterns can be replaced with patterns from other garments. That is, straight collar on a man’s shirts can be substituted by wing collar or mandarin collar virtually without doing the 3D CAD work again. We defined this feature as 3D style previewing in that the user can preview various combinations of garment patterns just as in 2D technical flats.
128
2. Previous work CAD is widely used for many engineering products. Small mechanical parts, cars, or buildings are modeled and simulated via computer-based tools such as AutoCADw, Abaqusw and so on. Such computer-based simulation reduces the cost for refining final product from initial design. For apparels, the computer-assisted design is only limited to 2D. Pattern shapes (Eckert and Helmut, 1999) or technical flats (Ji et al., 2002) can be drawn using splines efficiently. For 3D garment design, Luo and Yuen (2005) devised a method to modify 3D garment simulation result from 2D pattern design. They used particle-based governing equation so that the modified patterns can their equilibrium shape. They used a female one-piece and slacks for example and showed modification result of neck shape. But their work focused on modification of a single piece of pattern and not on the replacement of patterns or conjunction of multiple patterns. They team also proposed a method to construct 3D garment mesh from 2D sketches (Wang et al., 2003). The sketch-based drawing method seems to be a candidate for next-generation design method in virtual environment, but currently it is hard to depict details of patterns in 3D. So our approach is not to draw 3D details directly, but to utilize conventional 2D patterns. The apparel industry already has database of garment patterns with size and style variations. Our name-based sewing rule can be standard nomenclature for combining patterns from different garment sets. Any combination of garment patterns, such as arm pattern from men’s shirt and bodice pattern from female’s shirt, can be sewn and simulated in 3D. We describe the methodology in the next chapter. 3. Naming rules for arbitrary pattern conjunction 3.1 Pattern names and grouping Garment parts are generally composed of multiple patterns. We sorted them out to pattern groups. For typical basic garments, the patterns can be categorized into several groups. Figure 4 shows the four kinds of basic female shirt patterns used in this investigation. We divided each pattern set into seven groups, such as collar (designated as “COLLAR”), bodice front (“FRONT”), bodice back (“BACK”), right arm (“ARM1”), left arm (“ARM2”), right cuff (“CUFF1”), and left cuff (“CUFF2”). There are no limits for specifying the number of groups, but the categorization into seven groups seemed appropriate in this case. The patterns of the same shaded rectangular regions in Figure 7 belong to the same group. The pattern groups become the basic units for style previewing. We used simple strings for the names of patterns and edges. The name strings were composed of two or three words, separated by underscore character. Each word means group name, ID, and type, respectively. For example, pattern name “FRONT_1” means the first pattern of front bodice group and edge name “EAST_1_INV” means an edge located in the eastern side with inverse type. The type string is a description for optional details for edges. Patterns of a same group have name with same
suffix, i.e. front bodice patterns of Figure 6(b) has name strings of “FRONT_1”, “FRONT_2”, . . . , “FRONT_7”, while the ID’s do not have to be continuous integers. Once all the patterns are given name, the next job is to give names for each edge. We classified the edges into two categories, such as inter- and intra-group edges. 3.2 Inter-group sewing edges Inter-group edges were defined as edge pair whose edges are from different groups. Figure 1 shows an example. It is necessary that inter-group edges should have same name throughout all the garment sets, because inter-group edges actually combine patterns from different sets. For example, right arm pattern and bodice pattern should have edges with the same name, “ArmholeFR” as shown in Figure 1. In this way, if any other right arm pattern has an edge name “ArmholeFR”, such as patterns “ARM_1” in Figure 6(b)-(d), it can replace the default one in Figure 1.
Style previewing in 3D
129
3.3 Intra-group sewing edge names Intra-group edges are defined as edges sewn to each other inside a same group. Intra-group names are not critical to style previewing and thus they can have any naming rules. But our test result showed that use of North-East-West-South terminology (abbreviated as “NEWS”) was helpful. As the DXF patterns are deployed on 2D plane, they have generally have four- to eight-neighbor intra-group patterns (Figure 2(a)). For example, patterns shown in Figure 2(b) are located in parallel horizontally. We called this as “West-East” connection and the left hand side edge is given name of “WEST_1” and the right hand side edge is given name of “EAST_1”. The same method applies to “North-South” (Figure 2(c)) or “northeast-southwest” connections. This approach has an advantage that the user does not have to worry about creating different names for edges. Additionally, dart edges had name “DARTA”, “DARTB” and so on. And patterns such as arm pattern that were sewn to themselves in a cylindrical shape had edge named “SELF”. Only edges those cannot be described with previous types had to be given extra names such as “ETC_1” or else,
Ar m ho le FR
ArmR_1
A rm ho le FR
Arm 1
Front_1
Front
Figure 1. Example of inter-group sewing connection between group “ARM1” and “FRONT”
IJCST 22,2/3
Eas
West
130
t
North
South
(b) East-west type sewing
South
North
South North
Figure 2. Example of intra-group sewing
(c) South-north type sewing
West
East
West
East
(a) Example of four-direction based intra-group sewing type
but there was no need to use extra intra-group names in the four template garments used in this study. 4. Implementation 4.1 User input The whole procedure is similar to conventional 3D apparel CAD except that the edge names are needed. If the pattern data in DXF format is ready, the user should arrange the patterns on the plane and input the pattern/edge names. The relative positions of patterns determine the East-West or North-South edge names so they should be fixed once they are aligned on the plane. The following procedures went on in an automatic way. A windows program in C þ þ language was written to implement them.
Style previewing in 3D
131
4.2 Boundary mesh node generation using B-Spline When replacing a pattern group with another in style previewing, the new patterns generally have different edge lengths. But we want edges with even different arc lengths can also be sewn. This can be easily done by setting the boundary mesh nodes equal between two edges even if their total arc length is different. This may result in seam puckering in 3D drape simulation, but the same phenomenon will occur in reality, too. Thus, all the input DXF patterns (Figure 3(a)) were converted to multiple numbers of B-Splines before mesh generation (Figure 3(b)). The actual number of splines was determined by number of extremal curvature points and existence of sewing edge.
(a) Original DXF point data
(b) Piecewise B-Spline interpolation
(c) Variation of number of boundary nodes and mesh density
Figure 3. Boundary mesh node control using B-Spline interpolation
IJCST 22,2/3
132
If DXF pattern line segments had high curvature change or sewing edge in it, they were dissected into multiple splines. After all the pattern edges were interpolated by B-Splines, the mesh density was controlled (Figure 3(c)) so that the sewing edge couples have the same number of boundary mesh nodes. 4.3 Mesh generation Triangular mesh was used by Shewchuk’s (2002) conforming Delaunay triangulation. The boundary mesh node ID’s were recorded for later sewing operation. 4.4 Group identification Based on the pattern names, patterns with names of a same first-word (e.g. “Front_”) were grouped. Figure 7 shows an example of automatic grouping result of the template sets. 4.5 Intra-group edge searching In each group, intra-group edges were found and recorded. 4.6 Inter-group edge searching Among multiple groups, edges of same name were searched. As it is a rule that there should be always one pair of inter-group edge name, the searching is not time consuming by ignoring already found intra-group edges. 4.7 Assigning sewing information After all the intra- and inter-group edge pairs are found, it is time to convert them to sewing information. Sewing information for particle method means the mesh node ID’s of two points to sew. Using the boundary node ID’s found in Section 4.2, the sewing information can be filled. 4.8 Drape simulation Now all the necessary information, such as mesh information and sewing information, is given. We used particle-based method for drape simulation (Shewchuk, 2002) using matrix symmetry for fast calculation (Baraff and Witkin, 1998). The material property of the patterns was set to that of cotton. It was critical to check the cloth self-collisions because there were frequent penetration around collars. K-DOP based hierarchical method was used (Oh et al., 2006) for cloth self-contact detection. The simulation speed was real time (30 frames/s) for 1,000 mesh nodes including collision detection time. 5. Results and discussion 5.1 Test for template garment sets Before trying style previewing, simple female shirts were used to check the validity of the algorithm. Four kinds of basic designs with different details were used as template garment sets from open database of Korean sewing technology institute (Figure 4). Figure 5 shows the technical flats. They had different styles of arm length, collar, and bodice. The four sets were designated as set A, B, C and D, respectively. Set A had long arms, straight collar of one-ply pattern and the bodice had three patterns. Set B had mandarin style two-ply collar, long arms and the bodice was composed of 13 patterns.
Style previewing in 3D
133
(a) Set A
(c) Set C
(a) Set A
(c) Set C
(b) Set B
(d) Set D
Figure 4. The template DXF pattern sets
(b) Set B
(d) Set D
Figure 5. Technical flats of the template sets
IJCST 22,2/3
134
Set C had short arm, two-ply straight collar and 11-pattern bodice. Set D had round collar, short arm, and four-pattern bodice with darts. Pattern data were imported in DXF format as shown in Figure 4. Buttons, button holes and some details were ignored for simplicity’s sake. After manually assigning name information for four garment sets, respectively, (Figure 6), the groups were found automatically (Figure 7). From the drape simulation result (Figure 8), it was verified that our naming-based sewing method works well for each template garment.
Self CuffL
L
FL Waist
Dart Dart
East West
Waist F
_1
R
FR
Arm
hol
eB
Ar m ho le FR
le
Dart Dart
ho
Front_2
Front_1
(c) Set C (d) Set D Notes: Bold characters: pattern name; inverted: intra-group sewing names; the rest: inter-group sewing names
Self
CuffL_1
ArmholeBL
ArmL_1 South North ArmL_2 Self
CuffR
Ar m ho le FR _2
R
m
West Front_4 East West Front_5 East West Front_6 WaistFL
Self
NeckBR NeckBL Self L Sho lderF ulde rFR Collar_2 Shou ArmR_1 Ar mh South ole FR Collar_1 North _1 Sho Self ArmR_2 ulde rFR rFL Self ulde Sho NeckFR NeckFL Self
CuffR_1
CuffL_1
CuffL CuffL
kF
Front_7
ec
Front_3
N
Ar
WaistFR Front_1 East West East Front_2
Back_2
Self
CuffL
CuffL_1
Back_5
East West Back_6 WaistFL
East West Back_4 WaistFL
Dart Dart
WaistFR
rm
A
BL
le
ho
rm
A
East
West
CuffR
CuffR
ArmholeBR
CuffR_1
_2
Back_1
Self CuffL
Cuff R
CuffR
CuffR_1
CuffL
CuffL CuffL_1
2 BL _ le
East
West
BR
Figure 6. Examples of sewing names for the four template garment sets
Self
ho
le
Front_3 WaistFL
ho Self
Back_3
WaistFR Back_1 East West Back_2
BL le A WaistFL
East
West
m
Front_1
WaistFR
Ar
Armh _1 oleBL oleBR Self _1 Sou th Armh Ar Sou th mh Nor th Self Nor th ole BR Front_4 Front_2 R L F kF L Shou lderF Nec Neck houlderF ArmR_1 R Collar_2 S ArmL_1 South R North Ar mh leF ho ole Collar_1 m FL Self Ar R Shou lderF Self Self lderF Shou L FL k c 2 _ Ne ArmholeFL_1 oleFR h rm A N. N. _2 .E. S.E W. N.E. . S.E W. N .W. leFL . S.W S mho . Ar
Self Self
Self
ArmL_1
(b) Set B
Dart Dart
R ckF
FR Ne
Ar m ho le FR WaistFR Armh oleBR Front_1 ArmholeFR_1 _1 East _2 A F le rmho o R West leBR Armh _2 Front_2 3 A _ rm R holeB East East oleF R_3 West Armh West Ne Front_3 ckF East NeckBR R West Front_4 East West FL NeckBL Front_5 Neck L_1 East East Armh holeB oleFL _1 Arm West Front_6 West Armh _2 East oleFL oleBL h rm _2 A West _3 Front_7 Armh oleBL oleFL Armh _3 WaistFL
WaistFL ArmholeBL
NeckBL
NeckBL
NeckFR NeckFL
NeckBR
le ho m
Ar
FL
mh
ole
ole
mh
(a) Set A
BL
ole
mh
Ar
FL
Self
L
Ar L
WaistFR
BR
ArmR_1
Front_2
Front_1
Collar_2 South North Collar_1
ole
le
Self
mh
ho
Self
Ar
rm
ckF
lderF
Self
Self
Self
A
FL le ho
rm
Ar
Self ArmL_1
Shou
Ne
R
lderF
Shou
Self
A
FR
CuffR
Shou Collar_1
ArmR_1
Self
L
lderB
R
BR
le
ho
lderB
ho
Shou
m
Self
NeckBR
ArmholeBR
Ar
Self
Self CuffR_1 Cuff R
Back_1
rm
WaistFR
5.2 Style previewing test As the four template garments has seven distinctive groups, there can be 47 ¼ 16,384 possible choices of mixing the pattern groups. Among them, we chose four mixing types for test purpose. 5.2.1 Type nos 1b-1d. Starting from the garment set A, the left and right arms/cuffs were replaced with those of set B (designated as type no. 1b), C (type no. 1c) and
Collar (B) Arm2 (B)
Arm1 (B)
Cuff2 (B)
Cuff1 (B)
Collar (A)
135
Arm2 (A)
Front (A)
Front (B)
(a) Set A
(b) Set B Back (D)
Front (C)
(c) Set C
Arm2 (C)
Cuff1 (D)
Collar (C) Arm1 (C)
Cuff2 (C)
Back (C)
Cuff2 (D)
Arm1 (A)
Cuff1 (C)
Style previewing in 3D
Back (B)
Cuff2 (A)
Cuff1 (A)
Back (A)
Collar (D) Arm2 (D)
Arm1 (D)
Front (D)
(d) Set D
D (type no. 1d). Figure 9 shows the concept of 3D style previewing for this example. Note that the user does not have to input edge names again, which were already input in the template sets in Figure 6. The user has only to select which groups to use for a new garment set. As shown in Figures 10-12, the new garment sets had different arm patterns. 5.2.2 Type nos 2c and 2d. In type no. 2, set B was used as a base and the collar patterns were replaced with those of set C (type no. 2c) and D (type no. 2d). Figures 13 and 14 show the result. It is evident that the new garment sets have different collar patterns, while maintaining the original bodice patterns. 5.2.3 Type no. 3. In this type, set B was used as a base and the right arm/cuff patterns were replaced by those from set C. Also the left arm and left cuff were replaced by set D. Figure 15 shows the result. There seemed some puckering around shoulders because the length of new collar was shorter than that of the original one. 5.2.4 Type no. 4. Type no. 4 garment set was composed of front/back bodice of set C, right arm/cuff of set A, collar of set B and left arm/cuff pattern of set D. Figure 16 is the result.
Figure 7. Pattern DXF data and their automatic grouping result
IJCST 22,2/3
136 (a) Set A
(b) Set B
(c) Set C
(d) Set D
Figure 8. Drape simulation result of the four template garment sets
Figure 9. Conceptual view of 3D style previewing for mix type no. 1
It is notable that the user’s job is to only fill pattern names and edge names in Figure 6. The rest are done automatically. We implemented a computer algorithm to convert name-based sewing information into vertex ID’s for that. The drape simulation used time step of 33 ms for about 100 iteration loops. The drape simulation took about one minute in AMD Athlon 64 X2 3 Ghz PC with 2 GB RAM and Geforce 7900 GT graphic card. The sewing and meshing process took also less than a minute. But this could be skipped if all the mesh densities were set to an equal value. This means real-time pattern substitution can be possible. (The color figures and movies are available at: http://snowman0.com/StylePreview3D/).
Style previewing in 3D
Collar (A) Arm1 (B)
Arm2 (B)
Cuff2 (B)
Cuff1 (B)
Back (A)
137
Front (A)
(a) DXF pattern configuration
(b) Sewing result
(c) Simulation result
6. Conclusions A new name-based sewing rule for 3D garment patterns were proposed. Female shirts of four kinds of basic design were used for verification, but the same method can be applied to other kind of garments, such as slacks, blouse, one-piece dress, men’s suit and so on. The user’s work is only to prepare pattern data and assigning names
Figure 10. Result of mix type no. 1b
IJCST 22,2/3 Back (A)
Cuff2 (C)
Cuff1 (C)
138
Collar (A)
Arm1 (C)
Arm2 (C)
Front (A)
(a) DXF pattern configuration
(b) Sewing result
(c) Sewing and simulation result
Figure 11. Result of mix type no. 1c
Style previewing in 3D Back (A)
Cuff2 (D)
Cuff1 (D)
139
Collar (A) Arm1 (D)
Arm2 (D)
Front (A)
(a) DXF pattern configuration
(b) Sewing result
(c) Simulation result
Figure 12. Result of mix type no. 1d
IJCST 22,2/3
Cuff1 (B)
Collar (C) Arm1 (B)
Arm2 (B)
Front (B)
(a) DXF pattern configuration
(b) Sewing result
(c) Simulation result
Figure 13. Result of mix type no. 2c
Cuff2 (B)
Back (B)
140
Style previewing in 3D Back (B)
Arm1 (B)
Collar (D)
Arm2 (B)
Cuff2 (B)
Cuff1 (B)
141
Front (B)
(a) DXF pattern configuration
(b) Sewing result
(c) Simulation result
Figure 14. Result of mix type no. 2d
IJCST 22,2/3 Back (B)
Cuff2 (D)
Cuff1 (C)
142 Collar (B) Arm1 (C)
Arm2 (D)
Front (B)
(a) DXF pattern configuration
(b) Sewing result
(c) Simulation result
Figure 15. Result of mix type no. 3
Style previewing in 3D Back (C)
Collar (B) Arm1 (A)
Arm2 (D)
Cuff2 (D)
Cuff1 (A)
143
Front (C)
(a) DXF pattern configuration
(b) Sewing result
(c) Simulation result
Figure 16. Result of mix type no. 4
IJCST 22,2/3
144
to patterns and edges. We believe that our method can be used as a standard rule for 3D garment sewing description, because our method can be used to arbitrary garment data. References Baraff, D. and Witkin, A. (1998), “Large steps in cloth simulation”, Proceedings of the Annual Conference Series on Computer Graphics, Vol. 32, pp. 43-54. Eckert, C.M. and Helmut, E.B. (1999), “A garment design system using constrained Bezier curves”, International Journal of Clothing Science & Technology, Vol. 12 No. 2, pp. 134-43. Ji, Y.A., An, J.S., Lim, K.S. and Lee, D.H. (2002), “An introduction to a garment technical drawing system and its DB construction methodology”, International Journal of Clothing Science & Technology, Vol. 14 Nos 3/4, pp. 247-50. Luo, Z.G. and Yuen, M.M.F. (2005), “Reactive 2D/3D garment pattern design modification”, Computer-Aided Design, Vol. 37 No. 6, pp. 523-630. Oh, S., Ahn, J. and Wohn, K. (2006), “Low damped cloth simulation”, The Visual Computer: International Journal of Computer Graphics, Vol. 22, pp. 70-9. Shewchuk, J.R. (2002), “Delaunay refinement algorithms for triangular mesh generation”, Computational Geometry: Theory and Applications, Vol. 22 Nos 1/3, pp. 21-74. Wang, C.C.L., Wang, Y. and Yuen, M.M.F. (2003), “Feature based 3D garment design through 2D sketches”, Computer-Aided Design, Vol. 35 No. 7, pp. 659-72. Further reading Sul, I.H. (2010), “Fast cloth drape simulation using voxel based indexing and collision matrix”, International Journal of Clothing Science & Technology, Vol. 22 Nos 2/3 (in press). Corresponding author In Hwan Sul can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Fast cloth collision detection using collision matrix
Fast cloth collision detection
In Hwan Sul College of Engineering, i-Fashion Technology Center, Konkuk University, Seoul, South Korea
145 Received 25 July 2008 Accepted 12 June 2009
Abstract Purpose – The purpose of this paper is to develop a new and simple methodology for fabric collision detection and response. Design/methodology/approach – A 3D triangle-to-triangle collision problem was converted to simple 2D point-in-triangle problem using pre-computed 4 £ 4 transformation matrices. The object space was partitioned using voxels to find easily collision pair triangles. k-DOP was used to find inter-pattern collisions. Findings – Complex 3D collision detection problem is solved by simple matrix operations. Voxel-based space partitioning and k-DOP-based hierarchical methods are successfully applied to garment simulation. Originality/value – This paper shows that the collision matrix method can cover from triangleto-point to triangle-to-triangle collision with mathematical validity and can be simply implemented in garment simulation. Keywords Cloth, Simulation, Fabric production processes, Textile technology Paper type Research paper
1. Introduction Garment is 3D product which is sewn from 2D fabrics (patterns). To predict the final try-on shape of the garment, both material property of the clothes and bodice shape are needed. Material property of clothes determines the deformation of the garment while bodice shape constrains the final shape of the garment. For the modeling of fabric deformation, continuum approaches such as finite element method (FEM) or particle-based method were applied successfully. Kang and Yu (1995) adapted explicit FEM for fabric deformation. Etzmuss et al. (2003) also showed realistic virtual try-on simulation of a male suit. Although FEM is a reliable method for mechanical analysis, particle-based method was more widely used because calculation speed is an important factor for graphics and apparel design purpose, particle-based method seems to be more proper method for repetitive drape simulation such as garment designing. Modeling of fabric deformation has long been studied and it is not main concern of this paper. The other major factor that determines the garment product shape is collision reaction between garment and the body. Collision detection also becomes a rate determining step when the garment has multiple layers of fabrics. Collision detection among multiple bodies is an important theme in computer graphics area because no natural phenomena can be described without collision reactions. There have been many researches for accurate and fast collision detection. The simplest approach for detecting collision between triangular mesh elements is to check whether the two triangles overlap or not. GJK (van den Bergen, 1999),
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 145-160 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018621
IJCST 22,2/3
146
V-Clip (Mirtich, 1998) and I-Collide (Cohen et al., 1995) were the famous tools to compute the distance between objects. But comparing every triangle pair in each time step needs a massive calculation time and the burden becomes bigger when the number of mesh triangles increase. So wrapping the whole object into simple bounding objects such as boxes or spheres and collisions between bounding objects can be used for preliminary test to skip actual test. They are called as bounding box (Volino and Magnenat-Thalmann, 2000) or bounding sphere according to their shape. If the object mesh has complex shape, then the hierarchy of the bounding volumes can be used instead. Such hierarchy-based method showed notable time decrease. However, the actual triangle-to-triangle pair test is inevitable when the preliminary bounding volume test fails. This is the usual case in garment drape simulation where the cloth always contacts the body surface. Therefore, it is best to decrease the complexity of the triangle pair collision test to increase the overall drape simulation speed. Roughly, tens of iterations are needed for an expert apparel computer-aided design operator to initially align 2D pattern on proper 3D space even when the patterns are completely prepared. Therefore, fast drape simulation is the first necessary condition for fast garment pattern designing. This paper approximated the triangle pair collision test with simple matrix operations. The other time-consuming step in collision detection is to find possible collision pairs. Above-mentioned hierarchy method can also be used for finding possible collision pairs so we used k-DOP for hierarchical representation of meshes. Moreover, this paper also used voxelization of the space. The object 3D space was voxelized into multiple cells and each voxel recorded cloth or body mesh element ID’s at each frame. Voxel-based method was already applied for collision detection (Zhang and Yuen, 2000) but they voxelized the object itself so the voxels could not represent the original object smoothly due to memory limit. We voxelized the space into space roughly and used them indirectly as a storage to record the nearby mesh element ID’s. The detailed formulations are described in the following chapters and the results are displayed. 2. 3D voxelization 2.1 Voxel-based collision detection Not only actual triangular overlapping test, but also finding possible collision pair of triangles are rate determining step of drape simulation. Several approaches to find quickly possible pairs were done and the simplest approach is using voxel. The advantage of voxel is that the voxel index can be easily calculated from the coordinates. Figure 1 shows finding voxel ID from x-, y-, and z-coordinates. Once the voxel ID’s of reference triangle and target triangle are known, it can be known if they are possible collision pairs by checking whether they share common voxel ID. However, the disadvantage of voxel is aliasing and memory limit. As the voxel has box axis-aligned shape and the body elements have curved shape, there must be a voxelization error coming from aliasing. Smaller size of voxel can reduce the error, but using 1,0243 voxels need 109 space of memory. Without any further encoding procedure, simple voxelization cannot replace actual collision detection but it can be used only as an preprocessing method. Therefore, this paper used low density of voxels and the voxels are used only for registering element ID’s. As the cloth or body element deforms or moves, their position
Fast cloth collision detection
Voxel space
Voxel index j
Vi–1 j,k
Vi, j,k
→
Cloth node R = (x, y, z)
147 Vi–1, j–1,k
Vi,j–1,k
Voxel index i Voxel index k
index_i = int | (x–x_min) / VoxelSize | index_ j = int | (y–y_min) / VoxelSize | index_ k = int | (z–z_min) / VoxelSize |
Vi,j,k = index_i *n VoxelY *n VoxelZ +index_ j*nVoxelZ +index_k
is recorded in each voxels, respectively. Then, elements within same voxel ID become possible collision pairs. This method facilitates finding the possible pairs than finding nearly elements randomly. 2.2 3D voxelization of triangular prism Finding voxel ID of a vertex is easy as shown in the Figure 1. To find voxel ID’s of a triangular face, 2D rasterization algorithm was adopted. Bresenham (1965) algorithm is a method used for finding pixel area of a given polygon. This paper applied and expanded this method to 3D. As shown in Figure 2(a), edges of triangles are first voxelized. And then with respect to z-axis, overall voxels are separated into layers and each layer is applied Bresenham algorithm. Then, 3D voxel shape of the original triangle is acquired by accumulating the layers. But voxelizaing only triangle face can lead to errors so that nearby triangle pair can have different voxel ID’s. To avoid the error, this paper presents voxelization of triangular prism which is formed by upper and lower triangles. The upper and lower triangle is found by mesh enlarging and shrinking (Figure 2(b)). The coefficient of enlarging and shrinking was that of Taubin’s (1995) mesh smoothing algorithm so that the average volume is the same with the original mesh. Use of Taubin’s algorithm guarantees that each voxel only contains only one element ID, because Taubin’s method produces no volume change and no element entangling occurs. 2.3 Finding collision pair from voxel ID As each triangle element has multiple number of voxel ID’s, they should have a variable to keep voxel ID’s. To test collision of a triangle T, each voxels are opened and target triangle ID’s are acquired from the voxels. And then triangle-to-triangle test is
Figure 1. Finding voxel index from coordinates
IJCST 22,2/3
148 (a) 2D rasterization using Bresenham algorithm
Inflated layer
Original triangle
Figure 2. Voxelization scheme
Deflated layer (b) 3D voxelization of triangular prism
done with the target triangle. If collision has occurred, collision test of triangle T is finished. Otherwise another voxel or target triangle is compared until all the voxels are tested. This may seem redundant but actually voxels has only one element ID when the two objects are distant apart. Otherwise when the two objects are closer together, collision test end in the first voxel. Therefore, voxel-based collision pair finding is suitable for garment drape simulation. The simulation result shows the validity. 3. Transformation matrix for collision detection 3.1 Triangle-to-triangle collision detection Collision types between two triangles can be classified to three cases, which are node infiltration, edge infiltration, and face penetration (Figure 3). Node infiltration is the case in which a vertex lies under the reference triangle. Edge infiltration is similar to node infiltration but the intersection between edge and the reference triangle lies inside the reference triangle. Face penetration occurs when two triangles shear a common edge. To deal with three cases, both reference triangle and target triangle are rotated and translated so that the reference triangle lies on the xy-plane with one vertex clamped at the origin. Then the problem becomes a 2D point-in-triangle problem. Therefore, 3D triangle-to-triangle overlapping problem can be solved by two operators such as MapToXYPlane ( ) and PointInTriangle ( ). Assume ri, ui, fi are spherical coordinates of normal vector ni of reference triangle i. and pi0, pi1 and pi2 are the three vertices of triangle i. Then, MapToXYPlane() operation can be represented by a matrix operation Mi: p0 ¼ M i ð p 2 pi0 Þ þ pi0
→ →
p' ( p'x, p'y , p'z)
Z
p ( px, py , pz)
→
n'=(0,0,1) = (r' =1, 0,
Rotation and translation
→ n ( nx, ny, nz) = (r = 1, q, f)
π ) 2
Fast cloth collision detection
Y
149
→
p' ( p'x , p'y ,0)
Z Y
X
X
(a) Point infiltration →
p→( px, py, pz)
p' ( p'x, p'y,0) →
n' (0,0,1)
Rotation and translation
→
n
Y
Z →
→
q ( qx, qy, qz)
Y
q' ( q'x , q'y , q'z)
X
X (b) Edge penetration →
→
r'
→
p ( px, py , pz)
p'
r→( rx, ry , rz)
→
n' (0,0,1)
n→
Rotation and translation
Y
Z →
q ( qx, qy, qz)
Y
Figure 3. Three cases of triangular collisions
→
q'
X
X (c) Face penetration
2 6 M i ¼ M i ðu; fÞ ¼ 6 4 2 6 ¼6 4
cos fi
0 2sin fi
0
1
sin fi
0
32
cos ui
76 76 2sin ui 54 cos fi 0 0
cos fi cos ui
sin ui cos fi
2sin fi
2sin ui
cos ui
0
cos ui sin fi
0
sin fi
sin ui cos ui 0
0
3
7 07 5 1
3 7 7 5
And then conventional PointInTriangle( ) operation comes to check target vertex lies in the moved triangle.
IJCST 22,2/3
150
3.2 Precomputed collision matrix Transforming target triangle with respect to the reference triangle at every simulation frame is redundant when the target triangle belongs to bodice which does not move during drape simulation. In that case, transformation matrix Mi can be pre-calculated only at the first time. But still PointInTriangle() should be checked to know whether the target triangle lies inside the reference triangle or not. This can be omitted by adding additional transformation to Mi so that target triangle always have the three vertices at (0, 0), (1, 0) and (0, 1). By applying Mi, the first vertex always lies on the origin while the other two vertices do not. To match the other two points at desired positions, respectively, total five matrix operations were done. The 4 £ 4 matrix instead of 3 £ 3 of Mi was used to implement translation transformation. Figure 4 shows the graphical illustration of the procedure: . Step 1 (Translate): 3 2 1 0 0 2pi0x 7 6 6 0 1 0 2p 7 6 i0y 7 7 6 M Trans3D ¼ 6 7 6 0 0 1 2pi0z 7 7 6 5 4 0 0 0 1
.
.
M Trans3D translates the whole triangle by the displacement of pi0 from the origin. Step 2 (Rotation): 3 2 cos fi cos ui sin ui cos fi 2sin fi 0 7 6 6 2sin u cos ui 0 07 7 6 i 7 6 M Rot3D ¼ 6 7 6 cos ui sin fi 0 sin fi 0 7 7 6 5 4 0 0 0 1 M Rot3D rotates the triangle i so that the new normal vector becomes (0, 0, 1). Step 3 (Rotation2D): 2 3 sin ai 0 0 cos ai 6 7 6 2sin a cos a 0 0 7 6 7 i i 6 7 M Rot2D ¼ 6 7 6 0 7 0 1 0 6 7 4 5 0 0 0 1 where ai is the angle between x-axis and the edge p00i0 p00i1 . M Rot2D rotates the triangle so that edge p00i0 p00i1 lies on the þ x-axis.
pi0 n→( nx, ny, nz) =(r = 1, q, f) pi1
Z X
Fast cloth collision detection
Z M Trans3D
p'i0
Y
pi2
151
Y
X p'i1
(a) Matrix M Trans3D
p'i2
Z Z p'i0
Y M Rot3D n→( nx, ny, nz) = (r = 1, q, f)
X
p'i1
p'i2 (b) Matrix M Rot3D
→
n' (0,0,1) π = (r' = 1,0, 2 )
p''i0
Y
p''i1
X
Y
Y qi2
p''i2 M Rot2D p''i0
X
αi p''i1
qi0
qi1
X
(c) Matrix M Rot2D Y
Y qi2
q'i2
gi
qi0
MShear2D qi1
X
q'i0
q'i1
X
(d) Matrix MShear2D Y
Y 1
1 q'i2 MScale2D
q'i0
1 q'i1
X
X 0
(e) Matrix MScale2D
1
Figure 4. Procedure of triangle transformation
IJCST 22,2/3
.
Step 4 (Shear2D): 2 M Shear2D
152
.
1 gi
0
1
0
0
1
0
0
6 60 6 ¼6 60 4 0
0
3
7 07 7 7 07 5 1
where gi is the shear angle of edge qi0 qi2 with respect to the y-axis. M Rot2D deforms the triangle so that point qi2 lies on the y-axis. Step 5 (Scale2D): 2 3 sx 0 0 0 6 7 6 0 sy 0 0 7 6 7 M Scale2D ¼ 6 7 6 0 0 1 07 4 5 0 0 0 1 where sx ¼ 1=q0i1 ; sy ¼ 1=q0i2 . M Scale2D shrinks or deflates the triangle so that it matches (0, 0) (1, 0) and (0, 1). So the final matrix would be combination of the above transformations: P i ¼ M Scale2D · M Shear2D · M Rot2D · M Rot3D · M Trans3D
It is apparent that matrix Pi transforms any original triangle i to fixed triangle T0 ¼ {(0, 0), (1, 0), (0, 1)}. Then, the point-in-triangle test can be simply done by checking any point p lies inside T0. This is much easier than 3D triangle-to-triangle test where the triangle has arbitrary coordinates. Although Pi contains shearing and scaling deformation and they are not affine transform, they do not affect collision geometry because all the space is deformed linearly. The actual collision was classified as three types. The first is point penetration type, where a point lies inside the triangular column whose cross-section is the same with the reference triangle Ti. The second is edge penetration type, where an edge e crosses the reference triangle, i.e. e > T i – 0 and e > T i , T i . The last is face penetration type, where a target triangle T1 meets with reference triangle, i.e. T 1 > T i – 0 and T 1 > T i , T i . It is easy to check point infiltration type by seeing if the point lies insides the area formed by y ¼ 0, x ¼ 0 and y ¼ 2 x þ 1 curves. Point infiltration type alone can be used for simple collision test instead of all the three tests, but in that case convex mesh elements have a dead angle. Therefore, three types of tests should be used simultaneously. 3.3 Approximation for transformation Pi When the collision checking is only for the body-to-cloth and the body does not have motion, preparation of Pi can be done only one time. But when the body has motion or the checking is cloth-to-cloth, the coefficients, ai, gi, sx and sy cannot be known until the vertices are pi1 and pi2 transformed by M Rot3D · M Trans3D . It is not desirable to calculate
Pi at every frame. If we assume that the shear angle of edge pi0 pi1 and pi0 pi2 does not change during simulation and the initial values can be used as constants. Then, the approximated matrix Q i can be simply acquired as: Q i ¼ M Scale2D · M Shear2D · M Rot2D · M Rot3D · M Trans3D 2 3 Q i12 Q i13 Q i14 Q i11 6 7 6 7 Q i22 sy sin a sin f Q i24 7 6 Q i21 6 7 ¼6 7 6 cos u sin f sin u sin f cos f Q i34 7 6 7 4 5 0 0 0 1 where: Q i11 Q i12 Q i13 Q i14 Q i21 Q i22 Q i24 Q i34
¼ sx cos u cos fðcos a 2 g sin aÞ 2 sx ðg cos a þ sin aÞsin u ¼ sx ðcos aðg cos u þ cos f sin uÞ þ sin aðcos u 2 g cos f sin uÞÞ ¼ sx ð2cos a þ g sin aÞsin f ¼ 2sx ðcos að pi0x cos uðg þ cos fÞ þ pi0x ð2g þ cos fÞsin u 2 pi0z sin fÞ2 sin að pi0x cos uð21 þ g cos fÞ þ pi0x ð1 þ g cos fÞsin u 2 pi0z g sin fÞÞ ¼ 2sy ðcos u cos f sin a þ cos a sin uÞ ¼ sy ðcos u cos a 2 cos f sin a sin uÞ ¼ sy ð2pi0x cos aðcos u 2 sin uÞ þ sin að pi0x cos u cos f þ pi0x cos f sin u2 pi0z sin fÞÞ ¼ 2pi0x ðcos u þ sin uÞsin f 2 pi0z cos f
Qi’s can be prepared preliminarily and can be updated by inserting normal vector information such as ui’s and fi’s. 4. Finding inter- and intra-pattern collision pairs 4.1 Finding collision pairs using k-DOP tree Using space dividing scheme including voxel-based method is a good choice for culling possible collision pairs of non-deformable objects such as body mesh data. Since body data are pre-made and can be modified to a minimum number of mesh elements so that the size and number of voxels can be minimized. It can be also adapted to the case of inter- or intra-pattern collisions, but the voxel size should be more fine and the voxel information should be updated in every frame because the patterns deforms and moves (Teschner et al., 2004). So we used a better collision culling scheme using k-DOP tree (Klosowski et al., 1996). k-DOP tree is an extended method of axis-aligned bounding box (AABB) to a user defined number of coordinate axes. It should also be updated in each and every frame but it can be easily updated just the way as the AABB’s are updated. We constructed one k-DOP tree for each garments pattern meshes and the bottommost leaf nodes contained the actual k-DOP information of mesh triangle elements. If the garment patterns are far away, the root nodes of k-DOP trees do not collapse and the collision detection can be skipped. If the patterns collides at all, the child nodes among the k-DOP trees are compared and the colliding triangle element pairs can be known instantly.
Fast cloth collision detection 153
IJCST 22,2/3
154
Figure 5. Examples of body mesh voxelization
5. Results and discussion 5.1 Body mesh voxelization Figure 5 shows the example of body mesh voxelization. Both the arbitrary object (Figure 5(a)-(d)) and the human body data (Figure 5(e)-(f)) can be transferred to voxel space with the desired voxel sizes. Of course, there is trade-off between voxel size and efficiency. The smaller voxel size becomes, the voxels contains less elements and the possible collision pair of triangle elements reduces which results into faster collision detection speed. The bigger the voxel becomes, the smaller memory is needed for keeping the voxel information. Even though the respective voxel contains very small amount of data (ID’s of triangle elements which resides in the voxel), the total number of voxels can exceed one million if we voxelized the 1 m3 space with 1 cm interval. But if we
(a) Original Stanford bunny mesh data
(b) Voxels of 2 cm intervals
(c) 0.8 cm
(d) 0.4 cm
(e) Female avatar mesh
(f ) 4 cm voxelization
can reduce the number of mesh elements of the body data, we can use bigger size of voxels. In this paper, 10 cm voxel was used for the body mesh shown in Figure 5(e), which has 7,034 triangular elements (excluding the hair, shoes, and garments mesh in Figure 5(f)). 5.2 Inter-pattern collision detection using k-DOP tree To resolve the collision between garment patterns, each pattern was given with k-DOP tree structure and the collision among k-DOP nodes were used as a preliminary test to find possible collision pairs. Figure 6 shows a real time free fall simulation of 24 pieces
Fast cloth collision detection 155
(a) Bird-eye view of initial state
(b) Top view of the final stacking
(c) Lateral view
Figure 6. Real time stacking simulation of 24 colored papers on a marble plate
IJCST 22,2/3
of 4 £ 4 cm papers (or patterns). Each paper was given a different color to easily identify the penetration. The figure shows no inter-pattern collisions occur even though the simulation is real-time. Figure 7 shows the several k-DOP nodes in different levels. A 18-DOP (which has degree of freedom of nine axes) was used but simply their
156
(a) Simulation result
(b) Root nodes of k-DOP trees
(c) 2nd level nodes
Figure 7. k-DOP’s of different hierarchy levels
(d) 6th level nodes Note: k-DOP's are roughly shown as AABB's for the sake of convenience
AABB’s were drawn for simplicity’s sake because the actual 18-DOP is difficult to draw in real time (Klosowski et al., 1996). Once the possible collision pairs are known by k-DOP tree test, the actual collision detection between two triangles are done by using collision matrix Mi (where i is the index of the lower lying triangle element). With this strategy, the self-collision (collision inside each pattern) can be also searched within the tree. But including self-collision takes results into slower simulation speed and self-collision occurs in specific cases such as dress, shirt collar or skirt. But in those cases, possible self-collisions can be avoided by assigning multiple number of k-DOP trees to each patterns. In this investigation, only simple patterns which definitely have no self-collision were used. 5.3 Garment try-on collision test To verify the exactness of the collision matrix-based test, virtual garment try-on simulation was done. The garment pattern data was imported from DXF file of real garment patterns and the textures were from graphical artworks. Semi-implicit particle-based method of Baraff and Witkin (1998) was used for cloth simulation with time step of 20 ms. Simple shirt, pants, and vest patterns were used and they were triangulated so that the edges have approximately have 1 cm length. The patterns were initially aligned around the body mesh (Figure 8(a)) and three types of tests were done. Figure 8(b) shows the shirts above pants case. Figure 8(c) is the reverse case. In both cases, upper patterns are covering the lower patterns successfully. If the inter-pattern collision occurs, the outer patterns were lifted to collision-free position with respect to the lower pattern normal vector. Also, the inner patterns were given pressing force from the outer patterns. Figure 8(d) shows the vest is pressing and preventing the shirt pattern from falling-off. Figure 9 shows the collision status with red spheres (cloth-to-body) and blue cones (cloth-to-cloth). 5.4 Calculation time Table I summarizes the simulation time of colored papers and try-on trials. Cloth meshes of about thousand vertices can be simulated in real time (15 frames/sec in this case of stacking colored paper). Complex cloth and body meshes increases the total simulation time, but they increase linearly. 6. Conclusions Various techniques exist for collision detection algorithm but most of them are focused on objects moving independently. In the case of garment simulation, the body and cloth is under constant collision state and effective collision culling algorithm is important. We used voxel-based space dividing scheme for fast body-to-cloth collision detection and k-DOP-based hierarchical method for cloth pattern-to-pattern collision detection. The actual collisions between two triangles were checked by multiplying simple 4 £ 4 collision matrices to vertex coordinates. Using the collision matrix, the 3D collision test problem is simplified to 2D point in triangle test. The garment trial simulation tells that the collision matrix-based method can be almost real time for small number of vertices. Even for the large number of mesh vertices, the results showed reliability.
Fast cloth collision detection 157
IJCST 22,2/3
158
Figure 8. Garment collision detection using k-DOP tree and collision matrix
(a) Initial position of patterns
(b) Shirt above pants
(c) Pants above shirt
(d) Vest above shirt and pants
Fast cloth collision detection 159
Figure 9. Collision status
Notes: Red sphere: cloth-to-body penetration; blue cone: cloth pattern-to-pattern contact
Cloth mesh data Body mesh data Drape simulation time (ms/frame) No. of No. of No. of No. of No. of Semi-implicit Collision patterns vertices elements vertices elements integration detection Total Color paper stacking Pants and shirt Vest, shirt, and pants
24
1,440
2,040
242
478
33.8
30.2
64.0
10
5,587
9,202
7,034
13,850
188.0
125.0
313.0
18
7,065
11,533
7,034
13,850
236.0
147.0
383.0
Note: Including rendering
References Baraff, D. and Witkin, A. (1998), “A large steps in cloth simulation”, Proceedings of the Annual Conference Series on Computer Graphics, pp. 43-54. Bresenham, J.E. (1965), “Algorithm for computer control of a digital plotter”, IBM Systems Journal, Vol. 4 No. 1, pp. 25-30. Cohen, J.D., Lin, M.C., Manocha, D. and Ponamgi, M.K. (1995), “I-COLLIDE: an interactive and exact collision detection system for large-scale environments”, Proceedings of the ACM Interactive 3D Graphics Conference, pp. 189-96. Etzmuss, O., Keckeisen, M. and Strasser, W. (2003), “A fast finite element solution for cloth modeling”, Proceedings of the 11th Pacific Conference on Computer Graphics and Applications, pp. 244-51.
Table I. Speed of simulation and collision detection
IJCST 22,2/3
160
Kang, T.J. and Yu, W.R. (1995), “Drape simulation of woven fabric by using the finite-element method”, Journal of Textile Institute, Vol. 84 No. 4, pp. 247-57. Klosowski, J.T., Held, M., Mitchell, J.S.B. and Sowizral, H. (1996), “Efficient collision detection using bounding volume hierarchies of k-DOPs”, IEEE Transactions on Visualization and Computer Graphics, Vol. 4 No. 1, pp. 21-36. Mirtich, B. (1998), “V-clip: fast and robust polyhedral collision detection”, ACM Transactions on Graphics, Vol. 17 No. 3, pp. 177-208. Taubin, G. (1995), “Curve and surface smoothing without shrinkage”, 5th International Conference on Computer Vision, pp. 852-7. Teschner, M., Kimmerle, S., Heidelberger, B., Zachmann, G., Raghupathi, L., Fuhrmann, A., Cani, M.-P., Faure, F., Magnenat-Thalmann, N., Strasser, W. and Volino, P. (2004), “Collision detection for deformable objects”, Proceedings of the Eurographics, pp. 119-35. van den Bergen, G. (1999), “A fast and robust GJK implementation for collision detection of convex objects”, Journal of Graphics Tools, Vol. 4 No. 2, pp. 7-26. Volino, P. and Magnenat-Thalmann, N. (2000), Virtual Clothing: Theory and Practice, Springer, Berlin. Zhang, D. and Yuen, M.M.F. (2000), “Collision detection for clothed human animation”, Proceedings of the 8th Pacific Conference on Computer Graphics and Applications, pp. 328-37. Corresponding author In Hwan Sul can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Study of heat transfer through layers of textiles using finite element method Yuchai Sun College of Textiles, Dong Hua University, Shanghai, China and Hebei University of Science and Technology, Shijiazhuang, China
Xiaogang Chen
Study of heat transfer
161 Received 7 January 2009 Accepted 12 September 2009
School of Materials, The University of Manchester, Manchester, UK
Zhonghao Cheng Hebei University of Science and Technology, Shijiazhuang, China, and
Xunwei Feng College of Textiles, Dong Hua University, Shanghai, China Abstract Purpose – The purpose of this paper is to present the results of a study on heat transfer through a textile assembly consisting of fabric and air layers based on a theoretical model capable of dealing with conductive, convective and radioactive heat transfer. Design/methodology/approach – Quantificational results were given out by the aid of finite element (FE) analysis software MSC MARC Mentat. Findings – Significant findings through this paper include the change in heat flux against time and the transit temperature distribution at the cross-section of the fabric assembly. The size of the air gaps has a significant influence on the heat transfer. The balance heat flux drops by 40 per cent when the air gap increases from 2 to 10 mm. The influence of the air gap tends to become smaller as the air gap is further increased. The number of fabric layers in the textile assembly has a noted influence, more so when the ambient temperature is lower. Comparisons between the theoretical and tested results show a good agreement. Originality/value – This paper has established a new method for clothing comfort study by making use of a general purpose FE method software package. Keywords Finite element analysis, Heat transfer, Flux, Textile making-up processes Paper type Research paper
1. Introduction Heat transfer through a textile assembly or a fabric system is a complex process, involving conduction, radiation and convection. The combined heat transfer across the fabric system, consisting of fabric and air layers, is not simply the sum of what each mechanism would do in the absence of the others. The three heat-transfer mechanisms work together to determine the characteristics of the overall heat-transfer process. At different conditions, these heat-transfer mechanisms are of different importance. It has long been realised that heat transfer through textile assemblies involves multiple mechanisms. Peirce and Rees (1946) pointed out in 1946 that at the outer surface of the clothing exposed to the air, heat is lost by means of both convection and radiation.
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 161-173 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018630
IJCST 22,2/3
162
However, their work stopped short in establishing a model to describe the combined influence from the two. Farnworth (1983) presented a theoretical treatment of heat transfer through a bed of fibres considering conduction and radiation. It was illustrated that the theoretical results fitted well with the experimental data from testing some fibrous insulating materials. He reported that no detectable convective heat transfer took place inside the fibre bed. He assumed that the fibre bed is placed between two plates and thus the boundary conditions were not included in his work. In a more recent study, Mohammade et al. (2003a, b) presented a theoretical equation of the combined thermal conductive, convective, and radioactive heat flow through heterogeneous multi-layer fibrous materials. They used the effective thermal conductivities of the samples obtained by an experimental method, along with Fricke’s formula (Fricke, 1924; Fricke et al., 1990) for thermal conduction, to calculate the radiation thermal conductivity at high temperatures. Again, they did not take into account the boundary conditions at the surface of textile material. For clothing textiles, heat transfer is a complicated transient process. Generated form the body, heat transfers through the air gap between skin and fabric, then through the fabric system, to the outer surface of the fabric system. At the outer-surface (OS), heat is lost into the environment by heat exchanging with the surrounding air. During this process, conduction, convection and radiation are all involved, may be to different extent, in determining the total heat loss. The establishment of a theoretical model to describe this process is important for the understanding of heat-transfer process and therefore important for the design of fabric of different requirement of heat-transfer properties. This present study aims to describe the whole process of heat transfer from the very beginning when fabric contacts with skin to the dynamic heat transfer balance, and attempts to set up a theoretical model which combines heat conduction through the textile assembly with convection and radiation as the boundary conditions at the outer surface of the assembly. Finite element (FE) simulation is used to solve the heat-transfer problem as defined earlier, where changes in the structural parameter of the fabric assembly and in ambient temperature during the heat-transfer process are taken into consideration. From the methodology point of view, this research also aims to identify and evaluate a new technique through the use of MARC Mentat, a commercial FE package, to characterize the heat-transfer features through the textile assembly, including details of changes in heat flux against time, the transit temperature distribution at the cross-section of the fabric assembly as well as how ambient temperature, size of air gaps, number of fabric layers and air layers between fabrics affect thermal properties of the system by using FE method (FEM). 2. Theories of heat transfer To understand the thermal properties of the textile system, it is necessary to assess the contributions of the various heat-transfer mechanisms that may be operative. These mechanisms are conduction, convection and thermal radiation for dry heat transfer. 2.1 Conduction Fibres and air intermingle together in any textile yarns and fabrics. In another word, the fabrics are neither homogeneous nor isotropic. However, with the preposition that the average heat-transfer properties of fabrics are to be measured and calculated through the
theoretical and practical work, it is reasonable to assume that a fabric is a homogeneous and isotropic material in heat transfer. In addition, since thickness dimension of a fabric is substantially smaller than the fabric width and length dimensions in normal clothing situations, it is also feasible to consider the heat transfer through a fabric is a one-dimensional problem. Under such assumptions, the transient heat-transfer process through the insulating material is described as (Yang and Tao, 1999):
›T l ›2 T ¼ · 2 ›t cr › x
163 ð1Þ
where: T ¼ temperature (8K); t
¼ time (s);
l
¼ conductivity (W m2 1 K2 1);
r
¼ mass density (kg m2 1);
c
¼ specific heat (W S kg2 1 K2 1); and
x
¼ direction of heat transfer.
2.2 Convection As one of the basic heat-transfer mechanisms, convection involves the transport of energy by means of the motion of the heat-transfer medium, in this case the air surrounding the human body. When cold air moves past a warm body, it sweeps away warm air adjacent to the body and replaces it with cold air. It has been found that there is no convection inside clothing insulation even with a very low density (Peirce and Rees, 1946). For this reason, this paper considers convective heat transfer only at the outer surface of the textile assembly. In the FE analysis, the convective heat transfer will be set as a boundary condition. The heat flux due to convection can be expressed as follows (Incropera and DeWitt, 2002): q ¼ hðT G 2 T 1 Þ
ð2Þ
where: q
¼ heat flux (W m2 2);
h
¼ film coefficient (W m2 2 K2 1);
TG ¼ out surface temperature of the fabric (8K); and T1 ¼ temperature of the ambient atmosphere (8K). 2.3 Radiation The heat loss carried out by radiation from a clad human body to the environment is a situation where the clad human body as the heat source is enveloped by the environment. In this case, the heat flux by radiation at the outer surface of the textile assembly is governed by the following equation (Incropera and DeWitt, 2002): 4 Þ q ¼ s · 1ðT G4 2 T 1
Study of heat transfer
ð3Þ
IJCST 22,2/3
where:
s ¼ Stefan-Boltzmann constant, which is 5.6703 £ 102 8 W m2 2 K2 4; and 1 ¼ emissivity of the surface.
164
2.4 Combined convective and radiative boundary condition There is a temperature difference between the outer surface of the textile assembly that clad the human body and the environment, and this temperature difference causes heat transfer between the textile assembly and the environment in the form of convection and radiation. In this research, convection and radiation together are taken as a boundary condition at the outer surface of the textile assembly, which couples together with conduction to determine the total heat-transfer process through the textile assembly. According to Fourier’s law and energy conservation, the relationship between the temperature gradient in the thickness direction of the textile assembly and combined convective and radiative heat transfer can be expressed by equation (4) (Farnworth, 1983; Yang and Tao, 1999; Incropera and DeWitt, 2002):
›T 4 ¼ hðT G 2 T 1 Þ þ s · 1ðT G4 2 T 1 Þ ð4Þ ›n where n indicates the distance in the thickness direction. Equations (1) and (4) are the theoretical background for dry heat transfer through textile assemblies. 2l
3. Mesh generation and boundary conditions 3.1 Mesh generation Mesh definition is the process of converting a physical problem into discrete geometric entities for the purpose of analysis (Chen, 2002; Rao, 1982). What is most important about heat-transfer process is the overall temperature distribution across the textile assembly and the heat flow through it. Under the assumption that the solid fibre and the air are mixed evenly together in a fabric, what is important for mesh generation is the precision of the calculation and the normality of the elements shape. In this research, quad elements are used during the FE simulation. 3.2 Boundary conditions In the case of heat transfer through a fabric assembly, convection and radiation contribute jointly to the heat loss at the outer side of the textile assembly and they are therefore used in setting up the boundary conditions for FE analysis. The film coefficient is an important parameter heat transfer by convection. According to Rapp, the film coefficient for nature convection is (McIntyre, 1980): h ¼ 4ðW m22 K21 Þ The emissivity of a surface describes how effective it is at radiating energy compared with a black body. For textile fabrics, it is reasonable to assume the emissivity 1 ¼ 0.9 (McIntyre, 1980). 4. Theoretical results by FEM In the simplest terms, the discipline of heat transfer is concerned with only two factors, i.e. temperature, and the flow of heat. Temperature represents the amount of thermal
energy available, whereas heat flow represents the movement of thermal energy from one location to another. When textile fabrics contact with the skin of human body, heat transfers through the textile assembly because of the temperature difference between skin and the outside environment. After sometime, the transfer process reaches a dynamic balance when the temperature difference and the heat flow through the textile system become constant. For such a situation, this research uses the balance heat flux to describe the heat-transfer property for the fabric assembly. At the same time, the transient maximum heat loss from the skin when the fabric first touches the human body is also studied towards the understanding of transient cool feeling when clothes is just put on. During the research, different ambient temperature, different layer arrangement and different air intervals between the fabric layers are all taken consideration.
Study of heat transfer
165
4.1 Parameters of the textile assembly Conduction heat transfer through the textile assembly is affected by the material properties of the fabric and by the construction of the textile assembly which contains fabric layers and air intervals. In this study, cotton fabrics are used as the fabric layer in the textile assembly. The material parameters for air and the cotton fabric are shown in Table I (Yao and Zhou, 1990; Rohsenow and Hartnett, 1973; Raznjevic, 1976). 4.2 Description of the fabric and air system Figure 1 shows schematic illustrations of models used for the textile assembly, where two types of models are involves. The first type is to simulate the tight-fit of a piece of clothes on the body where the fabric layer touches the skin. The other type of models assumes that there is a air gap between the inner-fabric layer and the skin. In all cases, between any two layers of fabrics in the textile assembly, there is a layer of air. 4.3 Theoretical results 4.3.1 Influence of size of air gap. The size of the air gap between human body and fabric is affected by the profile of the body, the body movement and the fitness of the clothes. It is assumed in this analysis that the air gap between the skin and the fabric is the constant for the location considered, though in practice this gap can never be of the same size. Different sizes of the air gap between skin and the inner fabric are simulated to see its effect on the heat-transfer properties. Figure 2 shows the influence of the air Mass density (kg m2 3)
Specific heat (W kg2 1 K2 1)
Conductivity (W m2 1 K2 1)
1.17 234
1,027 1,217
0.026 0.071
Air Cotton fabric
Table I. Parameters of air and the fabrics
A layer of fabric A layer of air Skin
Tight-fit
1 layer system
2 layers
3 layers
Figure 1. Illustration of the textile assembly models
IJCST 22,2/3 Outer-surface temperature y/°C
166
34 32 30 28 26 24 22 0
2
4 6 Air gap x/mm (a) OS temperature
8
10
0
2
4 6 Air gap x/mm (b) Balance heat flux
8
10
0
2
Balance heat flux y/W.m–2
120 100 80 60 40
Maximum heat flux from skin y/W.m–2
20
Figure 2. Influence of the size of air gap between skin and inner fabric
1,600 1,400 1,200 1,000 800 600 400 200 0 4 6 8 Air gap x/mm (c) Maximum heat flux from skin
10
gap on the OS temperature, balance heat flux, and the maximum heat flux from skin. The ambient temperature is 208C and the skin temperature is 33.58C. It is clear from Figure 2 that the air gap between skin and the inner fabric have significant influence on heat transfer. Balance heat flux is the energy flow form the body through the textile assembly to the environment when heat-transfer process achieves its balance situation. Balance heat flux has been used to describe the warmth-keeping property of clothes. When the size of air gap increases, balance heat flux is decreased most rapidly when air gap increase form 0 to 1 mm and the balance heat flux reduced 26 per cent per mm. As the air gap increases from 1 to 2 mm, the balance heat flux is reduced 20 per cent per mm. Further increase in the air gap shows a gradual slowing down of the balance heat flux. When air gap increases from 9 to 10 mm, the reduction in the balance heat flux becomes a mere 7.8 per cent per mm. A similar pattern is shown for the OS temperature against the change in the size of air gap. The maximum heat flux from the skin relates to the transient cool feeling when putting on a piece of clothes. The maximum heat flux is reduced by 80 per cent when air gap is increased form 0 to 1 mm. When the air gap changes from 3 to 10 mm, the reduction rate of the maximum heat flux changes from 10 to 2 per cent per mm. All these suggest that an air gap between the skin and the inner fabric helps the warmth-keeping as well as reducing the transient cool feeling. So, it is necessary to create an air gap between the skin and the underclothes. Increasing the roughness of the fabric is one way for air gap creation. 4.3.2 Influence of fabric layers. Layered clothes are considered in this analysis. The assumption adopted here is that the size of air gap between the inner-fabric layer and the skin and that between any adjacent layers are both 1 mm, though in reality, air gaps are never neatly defined as this. The four models of textile assembly shown in Figure 1 are used. The models are subject to different environmental or ambient temperatures, ranging from 10 to 308C. In Figure 3, the “Fabric” curves refer to the tight-fit model where a single-layer of fabric touches the skin tightly and there is no air gap in between. The results reveals, as shown in Figure 3(a), that in all cases the OS temperature drops as the number of layers of the textile assembly increases, indicating that addition of more layers of fabrics with air gaps in the assembly is an effective measure to reduce heat transfer. It is more effective when the environment temperature is low. It also shows that the environment temperature affects the OS temperature positively proportionally. This is an indication that the increase in environment temperature reduces the temperature difference and hence reduces the heat transfer from the body through the textile assembly to the OS of the textile assembly. This is supported by the curves shown in Figure 3(b) where the balance heat flux decreases as the environment temperature goes up. It is of interest to see that when the environment temperature increases to a certain value all curves in Figure 3 will eventually meet at a single point, respectively. It can be speculated from Figure 3(a) that this temperature is actually the skin temperature of the human body, which is 33.58C. Balance heat flux through the textile assembly and the maximum heat flux from the skin at this point both become zero as can be seen in Figure 3(b) and (c). These would suggest that at such an environment temperature, the composition of the textile assembly will not cause any difference in heat transfer. At such a situation, the heat source, i.e. the human body, and the textile assembly have become one new heat source. This of course,
Study of heat transfer
167
Outer-surface temperature y/°C
IJCST 22,2/3
168
33 32 31 30 29 28 27 26 25 24 23 22 21
Fabric 1Layer 2Layers 3Layers
10 12 14 16 18 20 22 24 26 28 30 32 Environment temperature x/°C
Maximum heat flux from skin y/W.m–2
Balance heat flux y/W.m–2
(a) OS temperature
Figure 3. Influence of fabrics layer at different environment temperature
200 Fabric 180 1Layer 2Layers 160 3Layers 140 120 100 80 60 40 20 0 10 12 14 16 18 20 22 24 26 28 30 32 Environment temperature x/°C (b) Balance heat flux
2,500 2,000
Fabric 1Layer 2Layers 3Layers
1,500 1,000 500 0 10 12 14 16 18 20 22 24 26 28 30 32 Environment temperature x/°C (c) Maximum heat flux from skin
is a situation that is not easily achieved in practice as this balance can be interfered by many factors such as the body movement and the constantly changing ambient condition. It is also seen in Figure 3(c) that the maximum heat flux from skin reduces more rapidly for the tight-fit model than for the three-layered models whose curves overlap in Figure 3(c). For all models, the increase in ambient temperature reduces the transient cool feeling of the textile assembly and the number of fabric layers in the textile assembly becomes insignificant. 4.3.3 Temperature and heat flux distribution with the one-layer model. It is of interest to see how temperature and heat flux propagate from the heat source through the textile assembly. In this study, the one-layer fabric model is used with the air gap between the skin and the fabric being 10 mm. The ambient temperature is set to be 208C. The temperature distribution across the 10 mm air gap and at the inner-surface and OS of the fabric against time is shown in Figure 4. It is evident that at any given time the temperature gradient exists, with higher temperature closer to the skin and vice versa. The second phenomenon is that the temperature of air closer to the skin increases much more quickly than that further form the skin. For instance, the air temperature 2 mm away from skin gets closer to its balance in a matter of seconds, whereas the air temperature 8 mm away from the skin takes about 50 s to reach its balance. Figure 5 shows the heat flux distribution against time. On the skin when the air gap between the skin and fabric is 0 mm, the heat flux peaks within the first few seconds creating a sharp cool sensation after the clothes are just put on. The heat flux from the skin becomes smaller when the fabric is further away from the skin. The dynamic process for heat flux dies down eventually to reach the heat flux balance. It is not a surprise to see that the OS of the fabric responded most slowly to reach the heat flux balance.
Study of heat transfer
169
5. Experimental analysis To test the theoretical predictions, the OS temperature of the fabric was measured under different ambient temperature, variable fabric layers, and different sizes of air 32 2 mm 4 mm 6 mm 8 mm I.S O.S
31 30
Temperature y/°C
29 28 27 26 25 24 23 22 21 20 0
50
100 Time x/S
150
200
250
Figure 4. Temperature distribution
Heat flux y/W.m–2
IJCST 22,2/3
170
Figure 5. Heat flux distribution
75 70 65 60 55 50 45 40 35 30 25 20 15 10 5 0
0 mm 2 mm 4 mm 6 mm 8 mm I.S O.S
0
10
20
30 Time x/S
40
50
60
gap between the inner layer of fabric and the skin. The testing apparatus mainly consists of two parts, a heating plate providing a fixed temperature of 33.58C to stimulate the skin, and a data acquisition system with a temperature sensor to test and record the OS temperatures of the textile assembly. All tests were carried out in an artificial climatic chamber where the relative humidity was 65 per cent and the temperatures were made changeable from 12 to 308C. Figure 6 shows the composition of the testing apparatus, where 1 is the temperature sensor, 2 is the outer-fabric layer, and 3 is the textile assembly holder. Table II enlists the comparison of OS temperatures under three different situations between theoretical and experimental approaches. When examining the influence of the ambient temperature, a single layer fabric was used where the air gap between the fabric and the skin was 1 mm. In the case of fabric layers, while the ambient temperature was set to be 208C, the air gap between adjacent fabric layers and between the inner-fabric layer and the skin was chosen to be 1 mm. When considering the
1
RS232 RS232
89C52
D/A
Heating control
PC Heating plate
Figure 6. Illustration of the testing apparatus
Temperature measurement
A/D
PID
2
3
influence of the air gap, the single layer fabric model was used again with ambient temperature setting to 208C. Both experimental and theoretical data indicate that the ambient temperature affect the heat transfer from the skin to the OS of the clothing fabric. At a lower ambient temperature, the OS temperature is lower than it is subjected to a higher ambient temperature, leading to a larger temperature gradient which facilitates higher rate of heat transfer. In terms of the number of layers of fabrics involved in the textile assembly, addition of fabric, and air gap, layers provides better insulation against heat loss. Within a reasonable range, increase in air gap size leads to reduced OS temperature, indicating less heat loss. Devising a fabric assembly with controllable air gap sizes would enable the same piece of garment be used under different climatic environment. It is evident from Table II that the theoretical and experimental data agree well in all cases. This gives confidence in using the general purpose software MARC Mentat, for analysis on heat transfer for clothing comfort applications. This is of significance as it reduces the necessity of developing specialised software tool for clothing comfort. OS temperature was also measured against time. In this experiment, the single layer fabric model with 10-mm air gap between the fabric and the skin was used, and the ambient temperature was set to 208C. Figure 7 shows the change of OS temperature
Approach
Ambient temperature (one-layer fabric) 158C 258C
Theory Experiment
27.1 26.9
Fabric layers (ambient temperature 208C) 1 2 3
30.1 29.7
29.4 28.9
27.2 27.0
Study of heat transfer
171
Air gap (ambient temperature 208C) 5 mm 10 mm
25.8 26.0
24.7 24.9
22.9 23.1
Table II. OS temperature (8C)
23.5
Outer-surface temperature y/°C
23.0 22.5 22.0
Experimental result Theoretical result
21.5 21.0 20.5 20.0 0
50
100
150
200 Time x/S
250
300
350
400
Figure 7. OS temperature changes against time
IJCST 22,2/3
172
against time, obtained both experimentally and theoretically. It can be seen that a similar trend was demonstrated between the theoretical and experimental results, with the theoretical curve showing sharper temperature increase within the first 50 s. The reason for this is likely that the temperature sensor takes time to respond to the temperature changes. In both cases, the OS temperature stabilises between 22.8 and 23.08C. The experimental curve also demonstrated a fluctuation because the heating plate was set to work intermittently in simulating the skin temperature. 6. Conclusions Heat transfer through a layered textile assembly is studied using the FE approach. After analysing the nature of heat transfer from a human body through a textile assembly, the study presented a theoretical model that takes all three forms of heat transfer into consideration. While the heat conduction was regarded as the main mechanism for heat transfer, convection and radiation were involved as boundary conditions in solving the problem. On the other hand, four models of textile assemblies were created in the study, including the tight-fit, and three-layered models which allow an air gap between adjacent fabric layers and between the skin and the inner-layer fabric. Based on these models, simulations were carried out using a general purpose commercial FE package, MARC Mentat, and the relationships between structural parameter of the textile assembly and the thermal properties have been established. The establishment of theoretical models and the use of FE tool have enabled quantitative description of heat transfer with accuracy as verified by the experimental results. The size of the air gaps has a significant influence on the heat transfer. The balance heat flux drops by 40 per cent when the air gap increases from 2 to 10 mm. The influence of the air gap tends to become smaller as the air gap is further increased. The number of fabric layers in the textile assembly has a noted influence, more so when the ambient temperature is lower. Another aspect of work that has been reported is the heat flux from the skin when the skin touches the fabric under different situations, leading to further understanding of transient cooling sensation. This paper also reported on the setting up of a simple apparatus for testing surface temperature of fabrics, which is capable in dealing textile assemblies with different sizes of air gaps. It has shown that the theoretical results agree significantly well with the experimental results. This has demonstrated a new route for thermal analysis relating to clothing comfort. References Chen, H.H. (2002), MARC Finite Element Course, China Machine Press, Beijing. Farnworth, B. (1983), “Mechanisms of heat transfer through clothing insulation”, Textile Research Journal, Vol. 53, pp. 717-25. Fricke, H. (1924), “A mathematical treatment of the electric conductivity of disperse systems: the electric conductivity of a suspension of homogeneous spheroids”, Phys. Rev., Vol. 24. Fricke, J., Buttner, D., Caps, R., Gross, J. and Nilsson, O. (1990), “Solid conductivity of loaded fibrous insulation, insulation materials, testing, and applications”, in McElroy, D.L. and Kimpflen, J.F. (Eds), ASTM STP 1030, American Society for Testing and Materials, Philadelphia, PA, pp. 66-78. Incropera, F.P. and DeWitt, D.P. (2002), Fundamentals of Heat Transfer, 5th ed., Wiley, Somerset, NJ.
McIntyre, D.A. (1980), Indoor Climate, Applied Science, London. Mohammade, M. and Banks-Lee, P. (2003a), “Determining effective thermal conductivity of multilayered nonwoven fabric”, Textile Research Journal, Vol. 73, pp. 802-8. Mohammade, M. and Banks-Lee, P. (2003b), “Determining radiative heat transfer through heterogeneous multilayer nonwoven materials”, Textile Research Journal, Vol. 73, pp. 896-900. Peirce, F.T. and Rees, W.H. (1946), “The transmission of heat through textile fabrics, Part II”, Journal of the Textile Institute, Vol. 37, pp. T181-T204. Rao, S.S. (1982), The Finite Element Method in Engineering, Pergamon Press, Oxford. Raznjevic, K. (1976), Handbook of Thermodynamic Tables and Charts, Hemisphere, Washington, DC. Rohsenow, W.M. and Hartnett, J.P. (1973), Handbook of Heat Transfer, McGRAQ-HALL Book, New York, NY. Yang, S.M. and Tao, W.Q. (1999), Heat Transfer, Higher Education Press, Beijing. Yao, M. and Zhou, J.F. (1990), Textile Materials, China Textile Press, Beijing. Corresponding author Xiaogang Chen can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
Study of heat transfer
173
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
IJCST 22,2/3
174 Received 10 November 2008 Accepted 24 April 2009
Comparative analysis of hand properties and compositions of trace elements in linen fabrics produced in different regions Takako Inoue, Kengo Ishihara and Kyoden Yasumoto School of Life Studies, Sugiyama Jogakuen University, Nagoya, Japan, and
Masako Niwa Nara Women’s University, Nara, Japan Abstract Purpose – The purpose of this paper is to examine ladies’ linen fabrics produced in different regions – Japan, Italy, and Poland – to ascertain differences in mechanical, thermal, and air permeability properties. Design/methodology/approach – The paper investigates mechanical properties, air permeability, and thermal conductivity. The silhouettes of Polish, Italian, and Japanese linen fabrics are different. The thermal conductivities of the Polish linen fabrics are high. The levels of 72 elements were analyzed and remarkable differences were observed in the levels of 16 elements, including Li, Al, Si, Ti, Cr, Ni, Rb, and Y, Ag, among Polish, Italian linen fabrics, and linen fabrics made in Japan. Another ten elements were detected at some level in either the samples of Polish linen fabrics or linen fabrics made in Japan. Findings – There are differences among the Polish, Italian, and linen fabrics made in Japan, but the differences are not remarkable. Research limitations/implications – This paper is a wide world regional study of linen characterisation. Practical implications – Another ten elements are detected at some level in either the samples of Polish linen fabrics or linen fabrics made in Japan. There are differences among the Polish, Italian, and linen fabrics made in Japan, but the differences are not remarkable. Originality/value – The paper presents useful Measurement instrumentation, analysis and characterisation of linen fabrics from different regions of the world. Keywords Textiles, Mechanical properties of materials, Permeability, Thermal conductivity Paper type Research paper
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 174-186 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018649
Introduction We had clarified that the Polish linen fabrics and linen fabrics made in Japan (typical Hari type ladies’ garment fabrics produced in different districts) differed in terms of mechanical properties and thermal conductivity (Kawabata et al., 2000). Linen is an annual grass of the flax family, produced primarily in comparatively cold areas including northern France, Belgium, Russia, East European countries, and China (Japan Linen, Ramie & Jute Spinners’ Association, 2010). Early in the Showa era, The authors would like to thank Akira Ishikawa of Unix, Inc. for providing the linen fabrics made in Japan.
Russia was the country with the most flax-producing land, followed by Latvia, and then Poland ( Mori, 1949). Linen was imported to Japan from Belgium, China and others as fiber, from Italy, Poland and others as single yarn, and from China, Italy and others as woven fabric (Japan Linen, Ramie & Jute Spinners’ Association, 2010). We investigated the mechanical properties, hand evaluation, thermal conductivity, air permeability and the contents of trace elements of 100 percent linen fabrics. The samples used were 19 fabrics provided by the Polish National Research Laboratory, as well as an additional 16 fabrics made in Japan, and three fabrics made in Italy but collected in Japan. There is a paper reporting the results of a study, which lasted almost 30 years, of effects of climatic and soil conditions in Poland on five varieties of fiber flax (Stanislaw et al., 1997). In this paper, the effects of environmental conditions on the growth, development and yield of the following varieties of fiber flax were estimated: Fortuna, Minerwa, Svapo, Waza, and Nike. The environmental parameters were the soil composition, the type and pH of the soil, the climatic conditions, the time of agronomic operations, and the agronomy level. In this paper, we researched the differences in the properties of 100 percent ladies’ linen fabrics which were produced in the different districts. The levels of trace elements in these fabrics were analyzed to investigate whether significant correlations exist among trace element levels and the mechanical properties of 100 percent linen fabrics which are used to make suits, jackets, trousers, skirts, blouses, one-piece dresses, etc. Experiments Samples In total, 19 sample fabrics made in Poland from fiber grown in Poland were provided by the Polish National Fiber Research Laboratory. The sample fabrics made in Japan were eight fabrics provided by company A, four fabrics by company B, and four fabrics by another company. The growing districts for the linen fabrics from company A were Ireland and North France; the yarn was processed in Italy and woven, dyed, and finished in Japan. The growing districts for the linen fabrics from company B are not known, but the fabrics were woven, dyed, and finished in Japan in the same way as those from company A. The three sample fabrics made in Italy were woven, dyed and finished in Italy. Fabric analysis The mechanical properties of all of the linen sample fabrics were measured with the KES-FB-AUTO system (Kawabata et al., 1990). The tensile properties and compression properties were measured under standard conditions and high-sensitivity conditions applied to thin fabrics (Matsudaira et al., 1984). The measurement items and conditions are shown in the Appendix. The air resistance, RA(kPa s/m) of the sample fabrics was measured with a KES-F8 air permeability tester and the thermal conductivities, l(W/mK), of the sample fabrics were measured with a KES-F7 thermo labo. Trace element levels in fabrics The sample fabrics were weighed into a quartz digestion reaction container. After nitric acid (Wako Pure Chemical Industries, Ltd, Osaka, Japan) was added for trace element measurement, wet digestion was performed using microwave digestion apparatus (Multiwave and Perkin Elmer). The sample fabrics were weighed into a quartz digestion reaction container. After nitric acid (Wako Pure Chemical Industries, Ltd, Osaka, Japan)
Trace elements in linen fabrics
175
IJCST 22,2/3
176
was added for trace element measurement, wet digestion was performed using microwave digestion apparatus (Multiwave and Perkin Elmer). The clear solution was transferred to 50 ml polypropylene volumetric tubes, resulting in a 1:10 dilution; 1,000 g of a 0.1 mg/g yttrium (Y) standard solution in 5 percent HNO3 was added as an internal standard to create a concentration of 10 ng/g. Trace element concentrations were determined by an ICP-DRC-MS. The mass spectrometer settings and plasma conditions were optimized with a solution of 10 mg indium/l and the instrument operating conditions were as follows: ratio-frequency power, 1,100 W; plasma gas flow, 15 l/min; auxiliary gas flow, 1.2 l/min; nebulizer gas flow, 0.91 l/min. Data collection was performed from mass numbers 6-15, 19-39, 42-210 and 230-240. The signal intensities in each mass number of the element were compared to the signal intensities of 10 mg/ml of Mg, Rh and Pb, and the concentrations of each element were calculated using the Total Quant Method mode (ELAN Ver. 2.4, PerkinElmer Sciex Instruments). Results and discussion Mechanical properties The global chart (Inoue and Niwa, 2003) shown in Figure 1 was prepared to identify the region of the mechanical properties of ladies’ garment fabrics for different intended uses. The scale of the horizontal axis is normalized by the mean value and the standard deviation of the mechanical property of the 280 global ladies’ garment fabrics (Inoue and Niwa, 2003). The mean values and standard deviations are shown for the Polish linen fabrics and linen fabrics made in Japan. The mechanical properties of Italian linen fabrics were plotted individually. The mechanical properties of Polish linen fabrics showed high bending properties B, 2HB, low EM, LT, WT, and high tensile resilience RT. The shear stiffness G, shear hysteresis 2HG and 2HG5 of the Polish linen fabrics were low, and those of the Italian linen fabrics were the lowest. Italian linen fabrics have a low shearing stiffness. As for the surface properties, Polish linen fabrics have high SMD, but MIU was low and the surface was smooth. Of the samples used in this study, the thickness and weight of the Polish linen fabrics were higher than those of the linen fabrics made in Japan. Silhouette and hand evaluation Figure 2 shows a discrimination graph for the optimum silhouette design of ladies’ linen fabrics. The fabrics were discriminated using the mechanical properties of the three optimum silhouette designs (Niwa et al., 1998) – Drape-type has a beautiful drape silhouette, Hari-type has an anti-Drape silhouette which forms a suitable space between the body and clothing, and tailored-type has a silhouette which covers the body with a beautiful, dimensional shape. Niwa and others obtained a silhouette discrimination equation (Niwa et al., 1998) which discriminates optimum silhouette design using tensile properties, bending properties, shearing properties and weight per unit area of fabric. We used this equation. Ladies’ linen fabrics are located broadly within the Hari and tailored silhouettes. The Japanese company B linen fabrics falling at the the center of Hari-type to the outside of Hari-type show strong Hari-type characteristics. The other four fabrics made in Japan have Hari-type silhouettes only. The primary hand values KOSHI, SHARI, HARI, and FUKURAMI of ladies’ linen fabrics were calculated from the mechanical properties of linen fabrics using the KN301S equation (Kawabata, 1980), and then the mean values were plotted on a criteria chart for
Tensile
–3σ
–2σ
EM1
0.2
0.3
0.4 0.5
EM2
0.2
0.3
0.5
EM
0.2
LT
0.3
0.4
0.4
WT
0.7
0.005
2HB1
0.005 0.005
0.01
0.005
0.005
20
0.9
0.2
1.0
0.01
1.1 3
0.5 1
0.5 0.1
0.05
4
1
0.5
0.1
2 2
3
1
0.5
0.1
0.05
177
100 0.3
0.1
0.05
50 20
2
0.05
0.01
30
10
5
0.1
0.01
0.001
B
20
80
0.05
0.001
2HB2
4
60
0.01
B2
10
5
0.5
40
B1
10
5
0.8
0.1
RT
Bending
0.6
4 3
2
1
4
Trace elements in linen fabrics
(N = 280) 3σ
2σ
3 3
2
1
0.5
σ
2
1
0.5
0.05
(X-M)/s 0
–σ
2
1
0.5
1
2HB Shear
0.001
G
0.05
2HG
0.01
0.1
0.05
0.1 0.05 0.05
0.1
1 4
1
0.5
0.1
0.5
1
0.5
0.01
2HG5 Compr.
0.005
5 5
0.5
10 10
50
LC 0.4
0.6
0.8
1.0
WC 0.01
0.005
0.5
0.1
0.5
60
80
1
RC 20
Surface
MIU
40
0.05
MMD
0.1
0.2
0.25
0.01
0.005
SMD
0.15
0.5
0.3 0.05
1
5
0.35 0.1
10
T 0.05
0.1
0.5
5
W 2
3
4
5
10
20
30
40
50
Notes: This chart is normalized by the ladies garments fabrics population; suffix 1: warp direction; suffix 2: weft direction; : Japanese linen fabrics including A and B companies fabrics (n = 16); Polish linen fabrics (n = 19), mean value and : ± standard deviation
100
:
ideal men’s suiting (mid-Summer) (Kawabata et al., 2002) in Condition 1 of Figure 3. The KOSHI, SHARI and HARI of Polish linen fabrics and Japanese company B linen fabrics are characteristically high. The mechanical parameters and three basic components of tailorability which are concerned with total appearance value (TAV) as men’s suiting are plotted in Condition 2 of Figure 3. The TAVs of Polish linen fabrics, Japanese company B linen fabrics and Italian linen fabrics were close to the highest value
Figure 1. The mechanical properties of linen fabrics
IJCST 22,2/3
4 Tailored
178
2
Z2
s
0
s
s
–2
2s Drape
2s
Hari
–4
Figure 2. Discrimination graph for the optimum silhouette design of linen fabrics
2s
–4
–2
0 Z1
2
4
Notes: : Polish linen fabrics (n = 19); : Japanese company A linen fabrics (n = 8); ∆: Japanese company B linen fabrics (n = 4); : Japanese linen fabrics excluding A and B companies fabrics (n = 4); : Italian linen fabrics (n = 3)
of 5, so suitable silhouette formability and a high-quality finish suit can be predicted. The Z1Z2 discrimination parameters of the linen fabrics were also calculated using the mechanical properties of the three categories (Kawabata and Niwa, 1991) – cotton-, silk-, and wool-like. The results are shown in Figure 4. The three zones show each of the three categories. Compared with all of the linen fabrics, these are located between “Wool Like” and “Cotton like”. The fabrics other than Polish linen fabrics are located on the cotton-like side, and Polish linen fabrics are located near the wool-like side. Thermal conductivities and air permeability Thermal conductivities and air permeability are shown in Figure 5. The thermal conductivities of Polish linen fabrics are high, so these fabrics are found to be suitable as mid-Summer clothing. The relationship between air resistance and porosity is shown in Figure 6. The relationship between thermal conductivity and porosity is shown in Figure 7. The air resistance decreases as porosity increases, and a tendency for low thermal conductivity was also recognized. However, a tendency related to differences in production regions was not seen. Compositions of trace elements and production districts The trace element levels in Polish linen fabrics and other fabric samples acquired in Japan are shown in Table I. Of the 16 elements for which significant trace levels were
–3σ
Condition 1: Hand –σ σ 0
–2σ
Hand value Stiffness (Koshi) Anti-drape (Hari) Crispness (Shari) Fullness (Fukurami) 1 THV
1
2
1 0
3
4
2
3
4
1
2
3
2
3
1
Trace elements in linen fabrics
5
6
6
7
5 4
5
4
6
5
2
3σ
7
8
8 7
9
9
8
6
3
2σ
9
10
7
8
4
179
10
9
5
Condition 2: Suit appearance Mechanical parameters –3σ –2σ EL2 BS2 SS BP SP (B/W)1/3 (G/W)1/3
2 0.02
3
4
σ
0
5
0.1
0.2
3
4
1.0
5
0.3
2.0
3.0
4.0
0.4 0.5
10
1.5 4.0
40
5.0
1.0 20
2.0
3.0
30
0.3 0.4 0.5
2.0 0.2
3σ
20
0.1
1.0
0.05
2σ
10
0.03 0.04 0.05
0.4 0.5
2
–σ
30
40 50
2.5 5.0
6.0
7.0
Three basic components of tailorability –3σ Formability Elasitic potential Drape TAV
–2σ
0 0 0
–σ
0
2σ
2
4
2
4
2 0
σ
2
4
3σ 6 6 6
4
Notes: This chart is normalized by the men’s suiting population; the perfect property zone is shown by the shaded zone; : Polish linen fabrics (n = 19); . : Japanese . Japanese company B linen fabrics (n = 4); company A linen fabrics (n = 8); ∆: . : Japanese linen fabrics excluding A and B companies fabrics (n = 4); . : Italian linen fabrics (n = 3), mean values are shown
Figure 3. Hand value and suit appearance of linen fabrics plotted on a criteria chart for ideal suiting (mid-Summer)
IJCST 22,2/3
3
2
180
Silk Wool 2s
1
s
Z2
s
2s
0
–1
s
s
–2 s Cotton
–3
2s –3
–2
–1
0
1
2
3
Z1
Figure 4. Wool silk cotton graph for linen fabrics
Notes: : Polish linen fabrics (n = 19); : Japanese company A linen fabrics (n = 8); ∆: Japanese company B linen fabrics (n = 4); : Japanese linen fabrics excluding A and B companies fabrics (n = 4); : Italian linen fabrics (n = 3)
0.11
0.10
l (W/mK)
0.09
0.08
0.07
0.06
0.05
0.04 5 6 7 8 9 0.01
Figure 5. Thermal conductivity l and air resistance AR
2
3
4 5 6 7 89 0.1 AR (kPa·sec/m)
2
3
Notes: : Polish linen fabrics (n = 19); : Japanese company A linen fabrics (n = 8); ∆: Japanese company B linen fabrics (n = 4); : Japanese linen fabrics excluding A and B companies fabrics (n = 4); : Italian linen fabrics (n = 3)
Trace elements in linen fabrics
3 2
0.1 AR (kPa·sec/m)
7 6 5 4 3
181
2
0.01 7 6 5 75
80
85 Porosity* (%)
90
Notes: : Polish linen fabrics (n = 19); : Japanese company A linen fabrics (n = 8); ∆: Japanese company B linen fabrics (n = 4); : Japanese linen fabrics excluding A and B companies fabrics (n = 4); : Italian linen fabrics (n = 3); *: thickness at 0.5 g/cm2
Figure 6. Air resistance AR and porosity
0.11 0.10
l (W/mK)
0.09 0.08 0.07 0.06 0.05 0.04 75
80
85 Porosity* (%)
90
95
Notes: : Polish linen fabrics (n = 19); : Japanese company A linen fabrics (n = 8); ∆: Japanese company B linen fabrics (n = 4); : Japanese linen fabrics excluding A and B companies fabrics (n = 4); : Italian linen fabrics (n = 3); *: thickness at 0.5 g/cm2
seen, three of the elements, Li, Cr, Cu, were present at significantly higher levels in the Italian fabrics than in the fabrics produced in the other two districts. The levels of 11 elements (Al, Ti, Ni, As, Rb, Y, Mo, Ag, Cd, Ba, and La) were significantly higher in the Polish linen fabrics. The levels of only two elements (Si and Cl) were
Figure 7. Thermal conductivity l and porosity
IJCST 22,2/3 Elements
182
Table I. Comparison of element quantities
Li Be B Mg Al Si P S Ci K Ca Sc Ti V Cr Mn Fe Co Ni Cu Zn Ga Ge As Se Br Rb Sr Y Zr Nb Mo Ru Rh Pd Ag Cd In Sn Sb Te I Cs Ba La Ce Pr
Poland (n ¼ 19) Mean(ppb) SD 1.153 0.018 9.525 488.079 128.531 37.929 0 1,437.114 54.223 38.410 180.519 2.287 8.747 0.027 0.582 3.562 13.718 0.281 0.848 1.423 5.311 0.080 0.034 0.023 0.087 0.726 0.095 3.214 0.019 0.089 0.005 0.010 0.003 0.007 0.040 0.038 0.076 0.001 0.294 0.003 Tr 0.001 0.001 0.269 0.002 0.004 0.001
2.198a 0.024 5.513 327.969 61.298b 160.918a 0 6,097.157 155.764a 30.026 448.763 3.975 5.161b 0.019 0.474a 5.152 12.470 0.436 0.627b 1.330a 3.731 0.170 0.055 0.028b 0.142 0.336 0.079b 2.809 0.014b 0.088 0.004 0.008b 0.005 0.012 0.052 0.029b 0.106b 0.002 0.455 0.002 0.000 0.001 0.201b 0.001b 0.003 0.000
Origin Japanese market (n ¼ 16) Mean(ppb) SD 0.306 0.004 9.903 334.243 52.202 283.148 0 8,142.128 232.737 24.392 486.535 4.369 2.377 0.022 0.328 0.710 12.500 0.041 0.240 1.392 3.829 0.007 0.001 0.003 0.067 1.660 0.023 1.414 0.007 0.063 0.003 0.004 0 Tr 0.014 0.011 0.004 0.001 0.723 0.002 0 0.001 0 0.094 0.001 0.002 Tr
0.226a 0.005 13.313 232.221 39.882a 268.307b 0 10,174.247 220.991b 16.398 581.508 5.207 1.864a 0.016 0.384a 0.622 7.115 0.080 0.185a 3.545a 6.487 0.007 0.001 0.002a 0.106 4.688 0.018a 1.642 0.008a 0.076 0.003 0.003a 0 0.017 0.014a 0.044a 0.001 2.102 0.002 0 0.001 0 0.185a 0.001a 0.002
Italy (n ¼ 3) Mean(ppb) SD 5.868 0.006 8.831 326.155 45.865 91.334 0 10,874.434 257.739 39.578 904.343 7.541 3.330 0.019 4.526 0.502 12.503 0.557 0.249 6.945 1.329 0.006 0.002 0.009 0.134 0.366 0.024 1.693 0.006 0.148 0.003 0.004 0 0.001 0.019 0.045 0.005 0 0.275 0.003 0 Tr 0 0.160 0.001 0.003 0
Anova
7.273b 0.004 2.413 324.104 8.348a 83.505ab 0 6,833.870 146.859 ab 4.419 1,203.097 5.333 0.926ab 0.004 4.068b 0.423 2.760 0.682 0.087 ab 8.069b 0.277 0.001 0.000 0.006 ab 0.095 0.106 0.004 ab 1.680 0.001 ab 0.030 0.000 0.002 ab 0 0.001 0.002 0.030 ab 0.003 ab 0 0.0327 0.001 0 0 0.086 0.000 0.000 0
ab ab
**
** **
*
** **
** *
*
** **
*
** *
* *
(continued)
Elements Nd Sm Eu Gd Tb Dy Ho Er Tm Yb Lu Hf Ta W Re Os Ir Pt Au Hg Ti Pb Bi Th U
Poland (n ¼ 19) Mean(ppb) SD 0.001 0 Tr 0.003 Tr 0.002 0 Tr 0 0.001 0 Tr 0 0.001 0 0 Tr Tr 0.002 0.001 Tr 1.079 0.287 0.016 0.045
0.001 0 0.004 0.003 0 0 0.001 0 0 0.001 0 0 0.002 0.001 1.296 0.238 0.012 0.037
Origin Japanese market (n ¼ 16) Mean(ppb) SD 0.001 Tr 0 Tr 0 Tr 0 0 0 0 0 Tr 0 Tr 0 0 0 Tr Tr 0.001 0 0.850 0.033 0.009 0.041
0.001 0 0 0 0 0 0 0 0 0 0 0 0.001 0 1.617 0.068 0.006 0.071
Italy (n ¼ 3) Mean(ppb) SD 0.001 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.001 0.001 0 0.309 0.056 0.008 0.084
Trace elements in linen fabrics Anova
0.000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.000 0.000 0 0.058 0.007 0.002 0.090
Notes: Means and SD values are significantly different *p , 0.05 and * *p , 0.01; Tr, a very small quantity of the element was detected; values without common superscripts are significantly different
significantly higher in the linen fabrics made in Japan. Several elements (Ru, Te, Cs, Eu, Tb, Er, Yb, Ir, and Tl) were detected in Polish linen fabrics but not in the linen fabrics made in Japan, and only one element (Sm) was detected in the linen fabrics made in Japan but not in the Polish linen fabrics. Looking at the eight elements (Mg, Al, Si, S, Cl, K, Ca, and Fe) for which the levels exceeded 10 ppb, the levels of four of these elements (Mg, Al, K, and Fe) were higher in Polish linen fabrics than in the linen fabrics made in Japan, and the levels of the other elements (Si, S, Cl, and Ca) were higher in the linen fabrics made in Japan than in Polish linen fabrics. These results indicate that the Polish linen fabrics contained a larger variety of elements at extremely low concentrations compared to the linen fabrics made in Japan (Figure 8). Conclusions . The mechanical properties of Polish, Italian, and Japanese linen fabrics are different. The silhouettes formed by these fabrics are also different. . The primary hand values calculated from the mechanical properties of linen fabrics, KOSHI, SHARI, and HARI, are high, and the mean value of formability,
183
Table I.
IJCST 22,2/3
0.11 0.10 0.09 l (W/mK)
184
0.08 0.07 0.06 0.05 0.04 0
5
10
15 Sulfur (ppb)
20
25
30 ×103
Notes: : Polish linen fabrics (n = 19); : Japanese company A linen fabrics (n = 8); ∆: Japanese company B linen fabrics (n = 4); : Japanese linen fabrics excluding A and B companies fabrics (n = 4); : Italian linen fabrics (n = 3)
Figure 8. Thermal conductivity l and the quantity of sulfur
.
.
elastic potential and drape of suit appearance show suitable values. Polish, Italian, and Japanese company B linen fabrics can be predicted as yielding suitable silhouette formability and a high-quality finish suit. The thermal conductivities of the Polish linen fabrics are high at approximately 12 percent, so these fabrics are found to be suitable as mid-Summer clothing. The elements found in the fabrics, as well as the amounts of trace elements, differed depending on the production region.
References Inoue, T. and Niwa, M. (2003), “Objective evaluation of the quality of ladies’ garment fabrics ( paper translated from the Japanese ed.)”, J. Text. Eng., Vol. 49 No. 2, pp. 33-45. Japan Linen, Ramie & Jute Spinners’ Association (2010), National Attached Table Classified by Imported Statistical Items, Japan Linen, Ramie & Jute Spinners’ Association, Tokyo, pp. 1-15, available at: www.asabo.com Kawabata, S. (1980), “The standardization and analysis of hand evaluation”, The Hand Evaluation and Standardization Committee, 2nd ed., The Textile Machinery Society, Osaka. Kawabata, S. and Niwa, M. (1991), “Recent progress in the objective measurement of fabric hand”, International Conference of Textile Science ‘91, Technical University of Liberec, Czechoslovakia, Vol. 1, pp. 12-19. Kawabata, S., Niwa, M. and Yamashita, Y. (2002), “Recent developments in the evaluation technology of fiber and textiles: toward the engineered design of textile performance”, J. App. Polym. Sci., Vol. 83, pp. 687-702.
Kawabata, S., Niwa, M., Ito, K. and Nitta, M. (1990), “Application of objective measurement to clothing manufacture”, Int. J. Cloth. Sci. Tech., Vol. 2 Nos 3/4, pp. 18-33. Kawabata, S., Niwa, M., Koztowski, R., Manys, S., Nakano, K. and Inoue, T. (2000), “Fabric hand property of Polish linen fabrics for ladies’ outer wear”, Int. J. Cloth. Sci. Tech., Vol. 12 No. 3, pp. 193-204. Matsudaira, M., Kawabata, S. and Niwa, M. (1984), “Measurements of mechanical properties of thin dress fabrics for hand evaluation”, J. Text. Mach. Soc. Japan (predecessor Journal of J. Text. Eng.), Vol. 37, pp. T49-T57. Mori, S. (1949), Diamond Industrial Complete Book (10) of Linen Production, Diamond, Tokyo. Niwa, M., Nakanishi, M., Ayada, M. and Kawabata, S. (1998), “Optimum silhouette design for ladies’ garments based on the mechanical properties of a fabric”, Textile Res. J., Vol. 68, pp. 578-88. Stanislaw, R., Jadwiga, B. and Krzyztof, H. (1997), “Die ertragsfa¨higkeit der faserflachssorten in verschiedenen umweltbedingungen”, Max-eyth-gesellschaft im VDI (VDI/MEG), Heft 22, pp. 39-40. (The Appendix follows overleaf.) Corresponding author Takako Inoue can be contacted at:
[email protected] To purchase reprints of this article please e-mail:
[email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints
Trace elements in linen fabrics
185
Thickness and weight
Compression
Shearing
Surface
Bending
Shear stiffness Hysteresis at f ¼ 0.58 Hysteresis at f ¼ 58 Linearity Compressional energy Resilience Thickness at 0.5 gf/cm2 Weight per unit area
2HG 2HG5 LC WC RC T
W
MMD SMD
G
Mean deviation of MIU Geometrical roughness
LT WT RT B 2HB MIU
EM
Tensile
Tensile strain at max. load Linearity Tensile energy Resilience Bending rigidity Hysteresis Coefficient of friction
Symbols Characteristic value
Table AI. Characteristic values of basic mechanical properties and measuring conditions for KESF measurements Weight of specimen per unit area
mg/cm2
Upper limit pressure: 50 gf/cm2 Thickness at 0.5 gf/cm2 pressure
Contactor for geometrical roughness: a steel piano-wire, with 0.5-mm dia. and 5-mm length. Contact force: 10 gf Shear deformation under constant tension of 10 gf/cm Max. shear angle, f ¼ ^ 88
Pure bending Max. curvature, K ¼ ^ 2.5 cm2 1 Contactor for friction measurement: ten parallel steel-piano-wires with 0.5 mm dia. and 5-mm length simulating finger skin geometry,. Contact force: 50-gf
:10 gf/cm
Upper limit tensile force (max. load): 500 gf/cm :50 gf/cm
Strip biaxial deformation
Standard (Kawabata, 1980)
gf/cm degree gf/cm gf/cm – gf cm/cm2 % mm
– mm
– gf cm/cm2 % gf cm2/cm gf cm/cm –
%
Unit
High sensitivity (Matsudaira et al., 1984)
186
Mechanical properties
Measuring conditions
IJCST 22,2/3 Appendix
The current issue and full text archive of this journal is available at www.emeraldinsight.com/0955-6222.htm
Analysis of tactile perceptions of textile materials using artificial intelligence techniques Part 1: forward engineering B. Karthikeyan School of Engineering and Textiles, Philadelphia University, Philadelphia, Pennsylvania, USA and Department of Electrical and Computer Sciences, College of Engineering, Temple University, Philadelphia, Pennsylvania, USA, and
Forward engineering
187 Received 2 September 2008 Revised 24 May 2009 Accepted 24 May 2009
Les M. Sztandera School of Business Administration, Philadelphia University, Philadelphia, Pennsylvania, USA Abstract Purpose – The first of a two-part series, this paper aims to discuss the design and development of an artificial intelligence-based hybrid model to understand human perception of the tactile properties of textile materials and create an objective system to express those tactile perceptions in terms of measurable mechanical properties. Design/methodology/approach – A forward engineering system using the Model Free Algorithm approach of the Artificial Intelligence Technique to predict the tactile comfort score is presented. Findings – Human perception of tactile sensation is based on the weighted stimulus perceived by the human neural system. Originality/value – Contribution to intelligent textile and garment manufacture. Keywords Artificial intelligence, Modelling, Mechanical properties of materials, Textiles, Programming and algorithm theory Paper type Research paper
Introduction Textile materials form a unique branch in the field of material science. They exhibit an entirely different set of mechanical properties than most other conventional engineering materials. Textile materials find their application not only in the clothing sector but also in several other areas including geo-synthetics, medical transplants, etc. Properties such The research reported was supported in whole by Laboratory for Engineered Human Protection through grant W911QY-04-0001 from Department of Defense/US Army Natick Soldier Systems Center. The authors would like to acknowledge Contracting Officer Technical Representative, Carole Winterhalter, Warfighter Science, Technology and Applied Research Directorate, US Army Natick Soldier Research, Development and Engineering Center, Massachusetts 01760. The authors also thank Dr Howard Schutz, University of California-Davis, California, and Visiting Scientist at Natick, Massachusetts, and Dr Armand Cardello, Senior Research Scientist, Natick, Massachusetts.
International Journal of Clothing Science and Technology Vol. 22 No. 2/3, 2010 pp. 187-201 q Emerald Group Publishing Limited 0955-6222 DOI 10.1108/09556221011018658
IJCST 22,2/3
188
as inhomogeneity, discontinuity, anisotropicity and easy deformability make them a special case of materials suitable for such distinguished engineering applications (Amirbayat, 1991). They are non-linear and undergo large deformations even under low stress and buckling loads. Hence, the tactile perceptions of a textile material by humans form a different platform of study. A pioneering work of Stylios (1998) laid ground in terms of defining principles for aesthetic measurements of textile materials. It was further advanced when the relationship between the drape attributes and fabric bending, shear and weight was modeled using artificial neural networks (ANNs). It was found that using the natural logarithm of the material property divided first by the weight of the fabric produced the most predictive model (Stylios et al., 2002). In this paper, we analyze and map mechanical properties into perceived tactile comfort. Detailed analyses of both handfeel and mechanical properties and their relationships with tactile comfort, as well as which material properties influence tactile perceptions and why, can be found in Sztandera (2008a, b). Several kinds of testing equipment are available to quantify the mechanical properties of textile materials; the most widely used is the Kawabata Evaluation System (KES-F) for fabrics. The mechanical properties characterized by KES-F systems are listed in Table I (Kawabata et al., 1982). These parameters are analogous to the psycho-physiological evaluation system of human experts who subjectively evaluate the handle property of textile materials (Hu, 2004). In order to evaluate the hand value of a fabric material, the KES-F properties are used to determine the primary hand value, which is used subsequently to derive the total hand value (THV). The THV finds its application in garment manufacturing wherein a two-dimensional fabric material is transformed into a three-dimensional clothing system. In the process, mechanical properties greatly influence the handling and conformation ease of the fabric. Though THV characterizes the fabric in terms of its hand value, it does not depict significantly human perception of tactile comfort properties of the fabric.
Group
Property
Description
Tensile
EMT LT WT RT B 2HB G 2HG 2HG5 LC WC RC MIU MMD SMD W T
Elongation (%) Linearity of load-extension curve (ND) Tensile energy (gf.cm/cm2) Tensile resilience (%) Bending rigidity (gf/cm degree) Hysteresis of bending moment (gf.cm/cm) Shear rigidity (gf/cm. degree) Hysteresis of shear force at 0.5 degrees of shear angle (gf/cm) Hysteresis of shear force at five degrees of shear angle (gf/cm) Linearity of compression-thickness curve (ND) Compressional energy (gf.cm/cm2) Compressional resilience (%) Coefficient of friction (ND) Mean deviation of MIU (ND) Geometrical roughness (mm) Fabric weight per unit area (mg/cm2) Fabric thickness (cm)
Bending Shear Compression Table I. Mechanical property of textile materials evaluated by KES-F module
Surface characteristics Fabric construction
Models based on energy equations (De Jong and Postle, 1977), finite element analysis (Ghosh et al., 1990a, b), stochastic formulations (Behery, 2005), and ANN (Behera and Karthikeyan, 2006) exist to identify the interrelationship between the structure of textile materials and their functional properties. Linear models to predict the tactile comfort of textile materials in terms of both subjective and objective measurements are also found (Wong et al., 2003). Researchers have determined that human tactile perception of textile material is non-linear (Wong et al., 2004) thereby limiting the application of existing models. Moreover, these models are domain specific and their extent of extrapolation is limited. Thus, a system to portray non-linear tactile perception of textile materials in the light of comfort and in terms of their mechanical properties becomes essential. In the current research, a hybrid system based on artificial intelligence techniques was developed to identify the underlying relationship between the fabric mechanical properties and the human perception of tactile comfort. The major tools of AI are ANN and adaptive neuro fuzzy inference system (ANFIS) engines. These tools are capable of identifying/mapping the dynamics of the non-linear relationship that exists in the problem under study.
Forward engineering
189
Materials and methodology A diversified set of fabrics including woven, knitted and non-wovens materials was selected for the evaluation. Laminated fabrics with water, fire, and chemically retardant finishes are included. The range of their mechanical properties as measured using the KES-F module is given in Table II. While these 17 properties form the input of the ANN/ANFIS engine, human perception score of tactile comfort is used as the output. The human perception score is measured using the comfort affective labeled magnitude (CALM) scale shown in Figure 1. The scale developed by the Natick Soldier Systems Center ranges from 2 100 to 100 where a score of 2100 represents the greatest imaginable discomfort and a 100,
Property EMT LT WT RT B 2HB G 2HG 2HG5 LC WC RC MIU MMD SMD T W
Minimum value
Maximum value
0.98 0.48 2.40 22.46 0.01 0.01 0.45 2 0.32 1.70 0.23 0.03 38.35 0.18 0.01 2.18 0.03 4.10
35.87 1.31 61.90 65.30 5.37 6.98 25.54 25.54 39.23 0.58 1.66 94.27 0.38 0.18 27.35 0.23 64.88
Table II. Range of the mechanical properties as measured using KES-F module
IJCST 22,2/3
100
Greatest imaginable comfort
80 Extremely comfortable 60
190 40
Very comfortable
Moderately comfortable
20 0 –20 – 40 – 60 –80
Figure 1. CALM scale
–100
Slightly comfortable Neither comfortable nor uncomfortable Slightly uncomfortable
Moderately uncomfortable Very uncomfortable
Extremely uncomfortable Greatest imaginable discomfort
greatest imaginable comfort. The other labels are distributed in a progressive ratio scale (Cardello et al., 2003). In total, 33 fabric specimens were selected for the evaluation of mechanical properties and the same set of fabrics were evaluated by 50 human subjects for their perceived tactile comfort score. The fabric samples were sequenced in random order for the subjects to evaluate. The mechanical properties were tested with five replicas making a data set of 165 £ 17. The domain specificity is reduced to a greater extent by incorporating fabrics that find application in various domains, namely, battle dress uniforms, ballistic protection, vests, linings, etc. Comfort scale We repeat here the process of the developing the scale, after (Cardello et al., 2003), as it completely defines the way our output variable, tactile fabric comfort, was formed. According to the scale developers (Cardello et al., 2003) in order to develop a sensitive, reliable, and valid labeled magnitude scale of comfort, 35 volunteers, none of whom were members of the descriptive hand panel, were recruited from a random list. Word adjectives that could be used to modify the terms “comfortable” and “uncomfortable” to reflect intensity differences were compiled from previous scaling literature and from standard English language resources. The adjectives “greatest imaginable” and “greatest possible” were included to define scale values commensurate
with a common fixed end-point of positive and negative affective experience, as used in previously developed labeled magnitude scales (Cardello et al., 2003). These adjectives were used to create 41 word phrases, which in combination with two non-polar terms (“neutral” and “neither comfortable nor uncomfortable”), resulted in a total of 43 phrases to be used in scale development. The 43 phrases were printed on separate pages and assembled in random order into testing booklets. Before testing, subjects were provided with written instructions on the procedure to be used in scaling the semantic meaning of the phrases. Oral instructions with an example were also provided. Subjects sequentially rated each of the phrases to index the magnitude of comfort or discomfort connoted by the phrase, using a modulus-free magnitude estimation procedure. In this procedure, subjects assign an arbitrary number to indicate the magnitude of comfort or discomfort reflected by the first phrase (positive numbers used for comfort, negative numbers for discomfort). Subjects then make all subsequent judgments relative to the first, so that if the second phrase denotes twice as much comfort as the first, a number twice as large is assigned; if it denotes one third as much comfort, a number one-third as large as the first is assigned, etc. All ratings were made in spaces provided in the testing booklet. A subset of phrases was chosen to construct a labeled magnitude scale of comfort (Cardello et al., 2003). The criteria for selecting terms were low variability in perceived semantic meaning, parallelism in the terms used to describe comfort and discomfort, and selection of an equal number of comfortable and uncomfortable phrases (a decision based on evidence from the preference scaling literature showing that balanced scales are better for differentiating products). Examination of the standard errors of the geometric means for each of the phrases (Cardello et al., 2003) led to the elimination of several phrases (e.g. “mediocre comfort,” “barely comfortable,” and “a little comfortable”) due to their variable semantic meaning to the subjects. Other phrases were eliminated because of a lack of suitable parallelism in terminology for the purpose of establishing bipolarity (e.g. “superior comfort” and “oppressively uncomfortable”). Applying the remaining criterion to the phrases resulted in the selection of 11 phrases for use in the scale: five associated with comfort, five associated with discomfort, and one neutral term (“neither comfortable nor uncomfortable”) to define the zero point. The geometric mean magnitude estimates of the positive and negative phrases were transformed to range from 0 to þ 100 (positive phrases) and 0 to 2 100 (negative phrases). The phrases were then placed along a 100 mm vertical analogue line scale in accordance with their transformed values. The resulting labeled affective magnitude scale of comfort is shown in Figure 1. The CALM scale shown in Figure 1 has several advantages over other comfort scales commonly used in the literature (Cardello et al., 2003). With this scale, the level of comfort or discomfort experienced by an individual can be readily indexed by simply placing a mark somewhere on the line. This stands in contrast to the difficulty often encountered by subjects using magnitude estimation procedures. However, by having positioned the phrases of comfort/discomfort along the analogue line scale at points representing the magnitude of their semantic meaning as determined by a magnitude estimation procedure, it becomes possible to treat the measured distances along the scale as ratio level data. This stands in contrast to category scales of comfort, which provide only ordinal data. The ratio nature of the CALM scale enables statements to be made about whether a particular sample is 20 percent, 40 percent, three times, etc. as
Forward engineering
191
IJCST 22,2/3
192
Table III. Comparison of a biological and ANN
comfortable (or uncomfortable) as another sample. In addition, it does not require that the data be normalized, as is the case with magnitude estimates. Last, by using the “greatest imaginable” comfort (or discomfort) as end-points on the scale, the scale enables better discrimination between samples/conditions that are either very high or very low in comfort/discomfort and establishes a common ruler by which comfort/discomfort ratings of different subjects can be compared. Artificial neural network An ANN functions as a model based on connectionism. It represents the architecture of a human brain in the sense of interconnected neurons. A neural network essentially consists of several interconnected neurons that can be labeled as “input”, “hidden,” and “output” depending on the layer in which the neuron is present. The analogy between a biological neuron and an artificial neuron is given in Table III. Neural networks are capable of learning the relationship between input and output parameters by adjusting the weights and biases of the interconnected link. A neural network with three layers (one input layer, one hidden layer, and one output layer) can approximate any mathematical function with a satisfactory precision level. There exist neural networks with more hidden layers and the architecture strictly depends upon complexity of the relationship between the inputs and the outputs along with severity of noise in the data (Negnevitsky, 2002). The neural network needs to be trained with data sets whereby the system is provided with the inputs and the outputs; the weights are adjusted in the interconnections to fit with the targeted output. Once the neural network has learned the mapping function, it can predict the output when any unknown instance of the input is fed. This is analogous to how the human brain learns and reacts to unknown situations. Several architectures exist that can be used to model a neural network engine. Feed-forward back-propagation is one of the widely used learning algorithms to train a multi-layer perceptron neural network. It falls under the category of a supervised learning algorithm wherein, at the training phase, the network propagates the input pattern in a sequence through the layers of the network to generate an output. The error in the output, compared with the targeted output, is propagated backwards from the output layer to the input layer. The weights are modified to minimize the error by gradient descent method at each epoch (Baughman and Liu, 1995). Schematic representation of a feed forward neural network with error back propagation is given in Figure 2(a). The 17 mechanical properties form the input vector x(n) and the averaged Tactile perception score forms the output vector Y(n). The total activity at any internal neuron is given by equation (1): Natural neural network
ANN
Soma Dendrite Axon Synapse Electrical potential Threshold potential
Neuron or node Input Output Interconnections Weighted sum Threshold value
Forward engineering
Input signals EMT LT
193
Tactile comfort score
WT
W
Input layer
Hidden layer
Output layer
Error signals (a)
f (x) =
2a (1 – e–bx )
– a;
a = 1, b = 2
1
0
–1
–3
–1
0 (b)
1
3
Notes: (a) Schematic of feed forward back propagation neural network; (b) transfer function to map the inputs to the hidden layer
g kj ¼ f
n X i¼1
Figure 2.
! w kij x ki 2 t kj
ð1Þ
IJCST 22,2/3
194
Where: w kij
is the weight of the ith input to the jth neuron of the kth layer.
x ki
is the ith input.
t kj
is the threshold at the jth node of the kth layer.
f( )
is the activation function.
g kj
is the output of the jth neuron of the kth layer.
The objective of a transfer function is to convert the weighted sum of the inputs to a range that can be [0, 1] or [2 1, 1], and in the present case the latter is followed. The hyperbolic tangent function used as an activation function between the hidden layer and the output layer induces an accelerated learning process (Negnevitsky, 2002). This function, shown in equation (2) and schematically in Figure 2(b), has an additional advantage in the calculation of error gradient. The error gradient is determined just with the partial derivative of the transfer function multiplied by the error in the neuron’s output: f ðxÞ ¼ fðxÞtanh ¼
2a 2a 1 þ e2bx
ð2Þ
where, a and b are constants. Thus, for the neuron j in the output layer, the error gradient is calculated as:
dj ¼ fðxj Þ½1 2 fðxj ÞDj where:
dj
is the error gradient of the function f(x).
xj
is the output of jth neuron in the output layer.
Dj
is the difference in the actual output and the calculated output of the jth neuron in the output layer.
Hence, after each epoch, the weights are updated according to equation (3): Dw kij ¼ a:g kj :d kj
ð3Þ
where: Dw kij is the incremental weight of the ith input of the jth neuron of the kth layer.
a
is the learning rate and is usually kept at 0.1.
The details of the architecture and the set parameters for training and testing phases are as shown in Table IV. Thus, the neural network-based model is analogous to the biological system of perception. The mechanical properties are analogous to the stimuli and the conceived perception is reflected in the tactile comfort score. Analogous to the biological neural network, the ANN identifies the relationship between the input and the output. A schematic comparison of the function of a typical biological and ANN system is shown in Figure 3.
Forward engineering
Parameter
Value
Learning algorithm Number of input neurons Number of hidden layers Number of hidden neurons Number of output neurons Number of Epochs Mean square of error (termination criteria) Activation function Total data set Training data set Validation data set Training data set
Feed forward back propagation 17 1 25 1 1,000 0.001 Hyperbolic tangent 165 (33 fabrics £ five samples) 116 (