Clapp

s94-4, 1

THE ON-LINE INSPECTION OF SEWN SEAMS PIs:

Timothy G. Clapp (Team Leader, NCSU) Kimberly J. Titus (NCSU)

L. Howard Olson (G. Tech) J. Lewis Dorri@ (G. Tech)

Code: S94-4 ABSTRACT: Apparel manufacturing is traditionally very labor intensive due to the extensive style and fabric variation of the products. Most of the sewing machine manufacturers and some of the larger apparel companies have developed semi-automated sewing stations to perform operations which are constant across a large style range. These normally require an operator to load the machine, which then automatically sews and stacks the components. Although such stations improve production efficiency, they remove the almost unconscious operator inspection of the operation. The result is that only major seam faults such as thread breaks are observed. Other faults, skipped stitches or non-included seams, for example, may not be detected until the garment is completed or perhaps laundered. At this point, the manufacturer’s cost is at a maximum. In order to reduce the number of defective garments it is necessary to develop complete seam monitoring systems that meet the apparel manufacturer’s requirements of flexibility, cost, and reliability. Several techniques capable of detecting faulty seams on-line have been investigated which include thread monitoring with piezoelectric sensors, optical monitoring, and the use of a beta-particle transmission gauge. Prototypes have been developed for testing and demonstration of these techniques. The details and results of these investigations are provided.

GOALS: SHORT TERM: Design, develop, and evaluate sensor technologies to measure and quantify the quality of seams. Produce proof-of-concept prototypes to demonstrate the suitability of those technologies recommended for real-time monitoring systems. LONG TERM: Produce a “black box” to be attached to a sewing machine which would provide infotmation that would allow for the real-time adjustment of the sewing machine settings in order to optimize performance.

TECHNICAL QUALITY AND ACCOMPLISHhIENTS: An interdisciplinary team of researchers from the College of Textiles at North Carolina State University and the School of Textile & Fiber Engineering at Georgia Institute of Technology have collectively addressed the following tasks: 1) industrial collaboration in order to compile a technology survey and determine design specifications, 2) the fundamental research of fabric and seams, 3) the investigation of technological concepts which could potentially be used in a seam monitor, 4) the construction of proof-of-concept prototypes to demonstrate those techniques of greatest feasibility, and 5) conduct in-plant real-time testing of constructed prototypes. The technical quality of the research is reflected by progress made through the third year of the project and by interest generated throughout the industry.

National Textile Center Annual Report: November 1996

173

s94-4,2

Paw-s: 1. J. Lewis Do&y, “New Developments for Seam Quality Monitoring in Sewing Applications,” IEEE Industry Applications Society, 1995 Textile Fiber and Film Industry Conference, May 3, 1995. Note: This paper was cited as “Best Paper” at this conference. 2. J. Lewis Dorrity and L. Howard Olson, ‘Thread Motion Ratio Used to Monitor Sewing Machines,” Textile

Process Control 2001 International Conference, Manchester, England, May 18, 1995. 3. R.N. Cox, K.J. Titus, and T.G. Clapp, “Development of an On-Line Monitoring System for Stitch Quality,” 1996 ISA Textile Division Symposium Proceedings, Vol. 1, pp. 39-48. 4. R.N. Cox, K.J. Titus, and T.G. Clapp, “An On-line Monitoring System to Recognize Stitch Defects,” to be presented to and published by the Textile Engineering Division Technical Session for ASME Congress, Nov. 19, 1996. 5. K.J. Titus, T.G. Clapp, Z. Zhu, and R.P. Gardner, “A Preliminary Investigation of a Beta Particle Transmission Gauge for Seam Quality Determination,” Textile Research Journal, in press for Dec. 1996.

Presentations: 1. T.G. Clapp, K.J. Titus, G.R. Barrett, and Z. Zhu, “On-line Fabric and Seam Characterization Techniques,” presented to ISA, Raleigh, NC, Jan. 10, 1995. 2. T.G. Clapp, K.J. Titus, C.E. Farrington, research presentation for Sunny lshikawa of Union Special, Raleigh, NC, Feb. 13, 1995. 3. K.J. Titus and T.G. Clapp, “On-line Seam Monitoring Technology,” presented to American Bag Corp., a subsidiary of Milliken &Associates, Steams, KY, March 15,1995. 4. K.J. Titus, “Opportunities for Physicists in the Multi-billion Dollar U.S. Textile Industry,” Invited talk at the Joint Meeting of the American Physical Society and the American Association of Physics Teachers, Wash. DC. April 18, 1995. 5. J. Lewis Do&y, “New Developments for Seam Quality Monitoring in Sewing Applications,” IEEE Industry Applications Society, 1995 Textile Fiber and Fihn Industry Conference, May 3,1995. Note: This paper was cited as “Best Paper” at this conference. 6. J. Lewis Dorrity and L. Howard Olson, “Thread Motion Ratio Used to Monitor Sewing Machines,” Textile

Process Control 2001 International Conference, Manchester, England, May l&1995. 7. T.G. Clapp, K.J. Titus, R.N. Coy research presentation for Vice-President Al Gore, Nov. 13, 1995. 8. T.G. Clapp, K.J. Titus, R.N. Cox, and H.A. Foster, demonstration and research presentation to Diversified Systems Incorporated, Raleigh NC, April 19, 1996. 9. R.N. Coy K.J. Titus, and T.G. Clapp, “Development of an On-Line Monitoring System for Stitch Quality,” 1996 ISA Textile Division Symposium, Raleigh, NC, May 22-23,19%. 10. K.J. Titus, H.A. Foster, and B.R. Martin demonstration and research presentation to Mill&en & Associates, Raleigh, NC, August 7,1996.

Collaboration: 1. NCSU Dept. of Nuclear Engineering Levi Strauss & Co. Russell Corp. R&D in Alexander City, AL (TC)2 in Gary, NC Southern Tech Apparel Demonstration Site, Atlanta, GA Union Special, Huntley, IL On-Line Sensors, Waxhaw, NC

2. 3. 4. 5. 6. 7.

Visits: 1. Magnolia Finishing Plant of Milliken and Associates, Magnolia, SC 2. AMTEX Annual Report, Wash. D.C., March 2,1995

3. American Bag Corp. of Milliken and Associates, Stems, KY 4. Russell Corp., Alexander City, AL 5. Levi Technology Center, Richardson, TX

I74

National Textile Center Annual Report: November 1996

s94-4,3 6. Mill&en, LaGrange, GA 7. Eltex of Sweden, Greer, SC 8. Haggar, Dallas, TX 9. Royal Home Fashions, Durham, NC 10. Levi Strauss San Angelo Facility 11. Diversiiied Systems Incorporated, Greenville, SC

RESEARCH SUMMARIES: In the first year of the project, a technology survey was conducted by the team members at NCSU and Georgia Tech to determine the specifications for a seam monitoring device. The major stitch types identified for the study were the single and multi-needle chainstitches, lo&stitch, safety stitch, and overedge stitch. Seam types for both knit and woven goods included the single-fold knit hem, leg hem, and felled inseam and riser. Common seam defects for these seams included raw edges, needle cuts, non-inclusions, seam allowance and hem-width variations, puckering, mismatched seams, pleats, improper thread tensions, incomplete operations, and skipped stitches. A “black box” could be located between the folder and the presser foot, within a few inches behind the presser foot, or on the body of the sewing unit. Our research efforts were divided into two categories: stitch formation and formation of the seam. The first approach included techniques to monitor the motion of the sewing thread or optically “view” correctly formed stitches. Techniques to determine seam quality included the transmission of beta particles to determine the number of plies present within a seam. Highlights of these investigations are presented below. Stitch Quality Monitoring One of the main factors determining the quality of an apparel garment or other industrial sewn product is the quality of the stitching. An on-line system designed to monitor the formation of stitches could identify the presence of stitch defects immediately, thus eliminating defective products from proceeding downstream and providing important information regarding the performance of the sewing machine. As a collaborative effort, researchers at NCSU and Georgia Tech have investigated the movement of the sewing thread during machine operation in an effort to detect stitch defects. In the first approach, the time of thread motion was measured and compared to the time of a single sewing machine cycle to tie at a parameter termed the Thread Motion Ratio or TMR. This concept works if the true thread velocity is consistent from stitch to stitch. Although there is indeed variability in stitch formation, a great deal of experimental work has shown that averages of sets of stitches provide sufficient consistency to TMR that false indications of error occur much less than once per million cycles, or approximately once per eight hour work shift in an apparel plant. Prior work, including the initial proof-of-concept tests, had involved a 301 lockstitch [l] machine with a needle and bobbin thread. Other stitch formations recommended by the Technical Advisory Committee were the 406 two-needle chainstitch and the 504 overedge stitch, both formed with three sewing threads. The sensors used to monitor thread motion are low-cost, commercially-available piezoelectric transducers designed to detect yam break stops in weaving and knitting applications. They consist of a protruding ceramic eyelet through which the thread passes. As the thread moves National Textile Center Annual Report: November 1996

17s

s94-4,4

through the eyelet, the sensor outputs a voltage consistent with the level of vibration detected. Piezoelectric sensors were chosen in part due to their high resonant frequencies which make it possible to detect sudden changes in the force applied on the transducer [2]. The latter fact is extremely important considering the high speeds involved in industrial sewing. As there is a direct cost associated with the sensors, effort went into reviewing feasibility of using a single sensor rather than multiple sensors in a multiple thread environment. Results indicate that a single sensor can work quite effectively although sensor location is important to overall effectiveness. The initial lockstitch work was done with needle thread sensing since the bobbin is inaccessible to an external sensor. However, with the 406 two-needle chainstitch, neither of the top threads offers sensing as well as the bottom looper thread. The bottom looper thread has the greatest consumption since it traverses the gap between the two needles on each stitch. The features which determine which thread is “best” are the amount of thread consumed, the tension balance among the threads, and the geometry of the stitch. With the 406 stitch, if either needle thread breaks or is missed in the formation process, then a large change in TMR occurs. This illustrates that more than simple break conditions can be detected. Any nonuniformity among the stitches is handled through averaging several individual stitches and using a large control limit before declaring a seam error condition. Establishing four standard deviations as a control limit reduces the number of false indications. For example, when monitoring the looper thread at 4300 RPM, the control limits were +5% from a mean TMR of 40%, while needle thread breaks were at least 15% below the mean or three times the control limit. The overedge stitch, on the other hand, offers two looper threads which consumes the majority of thread. Of the two, the top looper thread is much better in giving clear, certain results. Due to the balance of thread tensions and stitch geometry, the top looper thread is most affected by a loss of either the needle thread or the bottom looper thread. With an average TMR of 20% and control limits at &2%, the top looper thread exhibited a change of 15% with the loss of either remaining thread. Recently, a high-speed video system has been acquired to aid in the study of the effects of dynamic behavior of sewing thread and its possible effects on the TMR. The “double pulse” commonly observed in the thread motion is of prime interest. It is theorized that the pre-pulse may involve reverse motion caused by the yarn elasticity and the check spring action. High speed video may answer those questions surrounding this. The video system is capable of frame rates up to 500 frames/second which wiIl provide excellent observation of sewing dynamics at speeds up to 3000 rpm (50 rps). Frame-by-frame playback is available. Studies during the coming months will focus on top thread behavior at the top tension and check spring mechanism, gooseneck, and needle on the lo&stitch machine. Bobbin thread behavior and its effect on TMR will also be studied. Although a frame rate of 500 fps is high for video systems, it pales in comparison to the rates of film systems in the thousands. There are many good reasons for choosing video over Iilm provided the frame rates are adequate. The greatest advantages of using video systems are that the medium is inexpensive and reusable and that feedback is immediate. Fihn systems quickly consume large rolls of film which must be developed to be analyzed for correct lighting, angle, contrast, etc. This not only results in long delays, but results in higher costs and requires considerable photographic skill to efficiently collect data. Machine speeds of 4000 rpm and greater would likely require film systems. Through our research, we anticipate answering these questions initially at lower speeds.

176

National Textile Center Annual Report: November 1996

s94-4,5

Stitch Formation Analysis In order to detect common single stitch defects such as shipped stitches and loose stitches, an understanding of stitch formation and the related sewing machine dynamics must be established. Current research efforts have focused on the 401 two-needle chainstitch machine used throughout the industry to construct felled seams in pants and jeans. As in most sewing units, both the needle and looper mechanism are connected to the main shaft of the sewing machine allowing a single stitch to be formed by a single rotation of this main shaft. The motion of the needle is coordinated with the looper mechanism such that as the needle penetrates and then emerges from the fabric, a loop is formed in the needle thread that can be caught and secured by the looper thread. The result is the formation of a chainstitch on the underside of the fabric. During the formation of the stitch, the analog output of a piezoelectric sensor indicating the needle thread motion is monitored. The data acquisition system utilizes signals from an encoder attached to the machine’s main shaft to identify the machine’s position within a stitch cycle; the encoder signal is thus used to sequence the A/D conversion of samples from the piezoelectric sensors. The result is the elimination of sampling effects related to speed by providing a consistent number of data points for each stitch, regardless of sewing speed. Figure 1 describes needle thread movement within a single stitch sewn at 2.5 stitches/second. The movement of the needle thread down through the fabric where it is secured by the looper thread is illustrated by a significant drop in the sensor output at the beginning of the stitch cycle. The constant voltage level observed in the middle of the stitch cycle represents a pause in the motion of the needle thread, while the looper mechanism is penetrating the needle loop and securing the stitch on the underside of the fabric. 4

Start of Stitch Needle Thread

0

100

200

300 Sample Number

400

500

Figure 1: Needle Thread Motion During a Single Stitch Since this motion describes a single stitch, the waveform over multiple, properly-formed stitches tends to be periodic. Consequently, deviations from the periodic waveform aid in identifying single stitch defects. Initially, it was thought that the motion of both the needle and looper threads were necessary to identify the occurrence of single stitch defects. However, analysis has shown that monitoring the needle thread is sticient in identifying the presence of one type of single stitch defect, the skipped stitch. This defect was identified by noticing the absence of the needle thread motion at the beginning of the stitch cycle. Because the needle thread is no longer linked properly with the previous stitch, tension levels decrease significantly to allow the needle, at the start of the next stitch cycle, to pull the needle thread down through the fabric without removing further thread from the supply bobbin. Therefore, no movement at this point in

National Textile Center Annual Report: November I996

177

S94-4,6

the stitch

cycle was registered by the piezoelectric sensors. However, due to the dynamics of high speed sewing, the thread will move during other portions of the stitch cycle whether or not the stitch is properly formed. Consequently, in order to detect skipped stitches, it is critical to precisely identify this portion of the stitch cycle. Figure 2 compares a properly-formed stitch with a skipped stitch sewn at 9.4 stitches/second (560 stitches/minute). The absence of needle thread movement at the proper position in the stitch cycle indicate the presence of a skipped stitch. The skipped stitch represented in Figure 2b was confirmed visually in the fabric sample. Similar results were obtained for sewing speeds of up to 3000 stitches/minute. Current research focuses on testing at higher speeds to confitm the viability of the piezoelectric sensors and this method. 3 2.5 & 2 G 1.5 >

1 0.5 0

4

Needle Movement

4io

I

0

I

960

Encoder Pulse (480 pulses per stitch)

Figure 2a: Two properly formed stitches

Stitch

0

460

960

Encoder Pulse (480 pulses per stitch) Figure 2b: Skipped Stitch over two stitch cycles

Optical

Stitch Monitoring

Another approach to stitch monitoring currently under investigation considers the reflection of a laser beam off a line of stitching. This principle is used in numerous industrial applications to measure surface structure, pattern conformity, and defect detection. The wavelength of light selected should exhibit low absorption (high reflection) characteristics for the incident material. Initially, spectrophotometric data was taken for ten dyed cotton twill fabrics ranging from white to black (both warp and weft threads), three varieties of woven Nylon 6,6 industrial fabric commonly used for airbags, and three types of dyed 50/50 polyester/cotton thread. Data was collected over

178

National Textile Center Annual Report: November 1996

s94-4,7

1100-2500 nm, in the near infrared (NIR) spectral region. This range was selected for study due to the ease of eliminating background noise caused by ambient light. As indicated in Figure 3, the reflection variation for the selected samples was minimal within the 2100-2300 nm range. However, considering the availability and affordability of light sources and detectors in the NIR region, a 1550 nm wavelength source was selected since the reflection characteristics for all the samples was above 35%.

Reflectance vs. Wavelength (NIR Range) for Various Samples

I

15X1 nm Wavelength

1550

i7m

it350

2000

-x-Green denim b Black du+m I -+-lndustrii fabric -x-Qokithread

2150

2300

2450

Wavelength [nm] Figure 3

The law of reflection states that a light beam incident on a surface at angle 8 will be reflected off the surface at the same angle e. This is particularly true for a laser light source for which the light is collimated thus will be incident at the same angle. Allowing for some absorption and scattering by the sample, the intensity of the laser beam will be reduced yet detectable by proper detection devices. With an appropriate choice of a suitably powerful laser and sensitive detector, the light reflected off of a fabric sample can be detected. Through focusing the beam down to a small size we can illuminate solely the stitch line, eliminating a large portion of background noise. As illustrated in Figure 4, as the stitch line moves relative to the beam, the curvature of the thread changes as a properly-formed stitch penetrates the fabric, changing the angle of incidence and causing the reflected beam to move away from the detector. As the thread comes back up through the surface of the fabric to form the next stitch, the reflected light comes back onto the detectors’ surface. The reflected light incident on the detector will create a signal, allowing the stitch to be “seen.” If a skipped stitch occurs, no change in the stitch’s curvature will occur, resulting in a continuous signal from the detector (Figure 4~). Stitch count, stitch length, and sewing speed must be known to successfully diagnose defects. Construction of the proof-of-concept prototype is currently underway.

National Textile Center Annual Report: November 1996

179

S94-4,8

Figure 4a

Figure 4b

Figure 4c

Beta-particle Transmission Gauge to Determine Seam Quality Another approach to detecting faulty seams is to monitor the formation of the seam. Numerous seam defects, such as raw edges, non-inclusions, seam allowance and hem width variations, and pleats, result from an improper number of fabric plies folded and caught by the stitching. By determimng the number of plies folded within the seam and secured by the sewing thread, the quality of the constructed seam may be assessed. Although optical ply detectors are commercially avaitable, their use is limited to fewer plies and lower weight fabrics. Other radiation systems such as beta-particle gauges, however, can provide a solution. Beta gauges are commonly used on-line in manufacturing processes to monitor the thickness of materials made in continuous sheets such as paper, plastics, and metal foils. Such sheet materials are of relatively uniform density such that their thicknesses may be computed from their density thickness. Density measurements such as those used to monitor tobacco packing density in cigarettes can also be made. Beta ray transmission designs have been used on-line in textile manufacturing to measure nonwoven fabric basis weights within a tolerance of 0.25% [3]. Unlike x-rays, microwave, infrared, and visible light radiation, beta rays can adequately separate the product from its environment and are not as composition-sensitive to additives and fillers.

180

National Textile Center Annual Report: November 1996

s94-q9

A beta particle is a high energy electron or positron which is emitted from. the nucleus of a radioisotope atom as a result of radioactive decay. This process for the strontium-90 radioisotope used in this study is described as

which yields a beta particle, b and a neutrino, v, when it decays to yttruim-90. In this particular case, yttrium-90 subsequently decays with the emission of another beta particle. Individual beta particles generally travel in a straight line, losing energy in small increments due to ionization interaction. For most of its range, the intensity of beta particles transmitted through an absorbing material experiences exponential attenuation which depends upon p the sample density and T the sample thickness. Provided B, the background counting rate, and R(O), the counting rate of beta particles when no material is present, can be determined from previous experimental data, the sample thickness can be calculated according to

The variable ~1 can be determined from @L/P, the absorption coefficient of the material (area per unit mass) which is determined empirically for a given arrangement of source, sample, and detector. Generally, it is found that for a given beta source N/p depends primarily on the “arial” density or area per unit mass of the sample and depends on sample composition, particularly the presence of heavy elements, to a much lesser degree. The variety of fabrics chosen for this study are intended to represent those commonly used throughout the industry as determined from our first year survey. Significant attention is given to high volume textiles, namely twill denim and jersey knit. Prior to investigating sewn seams, the transmission of beta particles through single and multiple plies of fabric was studied. Statistical fl~tuations, calculated as the square root of the counting rate, are within 2% of the average. Clearly, the number of layers can be determined up through 4 layers or more for all fabrics tested, including 16 oz. denim. The observed counting rates differ only slightly from those predicted from the exponential relationship; this deviation is attributed to the variation in fabric density as observed across the 1.0 mm collimated beam which is more pronounced for ribbed knits and corduroy. Two types of seams have thus far been considered: the single-fold edge ftish (EFc-2) commonIy used for knits and the felled seam constructed from twill denim. Defective single-fold hems are less likely to be “overlooked” than felled seams. A felled seam, used throughout the jeans industry to construct inseams, risers, and seat seams, is a folded seam secured by two or three rows of stitching. Because no raw edges are visible on either the inside or outside of the garment, defects are not obvious even to a skilled operator. Since the quality of felled seams cannot be assessed visually without removing stitching and consequently destroying the sample, seam quality was cor&med from a radiographic study. X-ray exposures obtained from the sample seams clearly indicate the number of plies within the stitching and the location of the panel edges relative to the stitching lines. This information was used to confirm that the beta gauge data correctly indicates the number of plies within the seam, allowing for a proper assessment of the seam quality.

National Textile Center Annual Report: November 1996

I81

s94-4, 10

The introduction of a radiation source such as a radioisotope into a non-automated manufacturing environment may encounter a great deal of concern regarding operator safety. Several precautions should be taken to greatly reduce this concern such as chasing a source with an activity level as low as possible while yet providing a satisfactory counting rate for the desired application. Providing a greater distance between the source and operator also serves as protection, for the intensity of the radiation falls off as the inverse square of the distance from the source. Particularly in industrial environments, shielding the source serves as a solution to operator safety. For the ?Sr beta source used in this research, l/4” Plexiglas adequately shields the source from the user. Minimal bremsstrahhmg will be produced since the amount of this radiation is proportional to the square of the atomic number and Plexiglas is composed of low atomic number (2) elements. As with radiation gauges currently incorporated into manufacturing environments, beta-particle transmission gauges could be safely used at semi-automated apparel assembly stations without posing a threat to the operator. References: [l] U.S. Federal Standard 751a, “Stitches, Seams, and Stitchings,” General Services Administration, Washington, DC, January, 1965. [2] Jones, R. J., and Munden, D. L., 1980, “A Study of the Mechanics and Geometry of the 2-Thread Chainstitch, Part 2, The Development of an Apparatus for the Measurement of Dynamic Thread Tension,” Clothing Research Journal, No. 3, pp. 115-137. [3] Boeckerman, P.A., “Meeting the Special Requirements for On-line Basis Weight Measurement of Lightweight Nonwoven Fabrics,” Tappi Journal, Dec. 1992,166-172.

RESOURCE MANAGEMENT The management structure of the project is based on the establishment and pursuit of common goals and objectives by a team of interdisciplinary researchers while maintaining a sense of individual ownership and responsibility to meet those goals. The management team is led by Dr. T.G. Clapp, who coordinates reporting, communications, and resource allocation for the project. Meetings are held as necessary to discuss ideas and disseminate results of specific tasks. A list of contributors to the project in addition to the PIs reflects the diversity of backgrounds and interests of the individuals. Other Faculty: Dr. Robin Gardner, Dr. Kuruvilla Verghese (NCSU, Nuclear Eng.) Visiting Scholar: Dr. Joachin Gayler (Wuppertal, Germany) Graduate Students: Robert Cox (NCSU, TE/EE), Zhaofeng Zhu (NCSU, Nuclear Eng.), Howard Foster (NCSU, Physics/Textile Material Science), Rob Schoenborn (G.Tech), Melissa Wetherington (G.Tech), Susan McWaters (G.Tech), Elizabeth McFarland (G.Tech, TE), Wendy Bishop (G. Tech, EE), Todd Kleine (G. Tech) Undergraduate Students: Adam Davis (TE), Neal Baldwin (TE)

182

National Textile Center Annual Report: November 1996