Copyright MCB UP Limited (MCB) 1998

M.R. Jackson: Department of Mechanical Engineering, Engineering Design Institute, Loughborough University, Leicestershire, UK

M.E. Preston: Department of Mechanical Engineering, Engineering Design Institute, Loughborough University, Leicestershire, UK

ACKNOWLEDGMENT: The authors would like to thank Guy Birkin & Co., Nottingham, UK and Dartex Machinery for their continued support during the research programme. This work was funded by the SERC/ACME Directorate, to whom thanks are acknowledged.

Introduction

One business segment which has not benefited from an integrated approach to the manufacturing process is the textile industry. This is true for the UK and world-wide. A case in point is in the lace fabric and lingerie manufacturing industry, where examples of high technology CADCAM knitting systems exist, but the rest of the manufacturing process such as cutting, materials handling and component assembly, has yet to catch up with this highly computerised front end design and knitting stage. Visual inspection by skilled operators is also applied at many stages of the process. Quality variability and bottlenecks at certain process stages are the result of this lack of integration, leading to high work in progress and increased product lead times.

Lace manufacturing is examined in this paper. Lace fabric is inherently flexible and, because of knitting tension variations along and across a typical lace breadth, the knitted pattern geometry bears only an approximate resemblance to the actual CAD pattern which is used as source data for the knitting machine control codes. Here is a tricky problem when cutting and inspection tasks have to be carried out. This is an expanding market (ITS, 1992) worth typically Pounds 1000 million per year in direct product sales.

Conventional lace manufacturing

The existing lace design and production process is a mixture of old and new technologies as illustrated in Figure 1. The front end design process starts with discussion with clients, where designs are initially explored by manual sketching, combined with previous design viewing. Once a design has been established in principle the designer transfers the individual thread runs to a highly specialised CAD package by using a manually operated point digitiser. This data is post processed to form knitting machine control codes for the thousands of electromagnetic actuators which manipulate the individual threads in the warp knitting process. The lace web thus produced (Figure 2) is typically 3 metres wide and 100 metres long.

The individual patterned lace breadths, known as ribbons (typically 15 mm to 250 mm wide), must eventually be separated from the lighter backing mesh waste. Before this separation process the lace is dyed for the required colour and then stentered by a heat setting process to provide adequate stiffness. The feel of the lace is adjusted using a paraffin based softener after stentering. The form of the lace is still in the 3 metre x 100 metre bulk at this stage. It is then separated by a high speed cutting process based on a band knife cutter. This process is used for straight slitting of the backing mesh normally, but can also scallop (i.e. cut along the edge) the less demanding patterns. The function of this part of the process is to separate the 3 metre wide web into more manageable widths of typically 1.5 metre.

Each reduced width lace web is then processed by a low cost scalloping machine (Figure 3) which follows the edge of the lace pattern using a mechanical edge detection technique. This accommodates the variations in edge position and geometry due to the previously mentioned pattern variations caused by knitting and stentering tensions. The scalloping process is slow and fairly inaccurate requiring constant operator attention and two passes through the cutting machine in order to produce one finished lace ribbon. Each 100 metre ribbon is then visually inspected in order to detect any faults which may have been caused by the cutting process, then rolled up onto spool holders and shipped out to the customer.

The volumetric flow rate of lace through the manufacturing stage can be represented by Figure 4. Here the bottleneck characteristic of the cutting room can be seen. The scalloping process also suffers from the inability to deal with closely interlocked lace patterns, known as centre-cutting. These must be cut by hand using scissors, resulting in corresponding throughput and cost disadvantages.

The mechanical scalloping process has so far been unable to address the industry goal of achieving dual edge cutting of a lace breadth in one pass through the machine. This would substantially reduce work in progress.

While "rigid" lace, which is in fact quite flexible, has been the mainstay of the lace manufacturing industry for many years, the newer "stretch" lace, which is very flexible and rapidly increasing in volume due to fashion demands, is very difficult to cut fast with the desired quality of finish. The industry in the UK is, therefore, facing a severe processing problem.

The temptation to move the entire manufacturing operation to the Pacific Rim, where many operators could be applied to the cutting and inspection problem, is great, with the consequent loss of jobs and revenue in the UK. A solution which produces an acceptable solution for the UK economy as a whole is desirable.

Fundamental to the success of this strategy is a lace scalloping machine which can process bulk supplies of lace at high speed, typically 1 m/sec web speed and cut along both edges of a lace breadth simultaneously.

Enabling technologies

Machine vision is no longer the expensive and slow inspection research tool of a few years ago. The advent of increasingly fast image processing engines, typically Transputer, Digital Signal Processor (DSP) or digital hardware based, has meant that reasonably complex image processing algorithms can be executed cost effectively. Several examples of textile inspection work using machine vision can be found (Bradshaw, 1995; Sanby, 1995; Yunming, 1993).

While there is still work to be done in the inspection field, researchers are now focusing attention on investigating the possibility of using imaging information to determine what the process should do next. Some textile based research is evident (Ameziane, 1985; Kimoto, 1986; Russell, 1989) with a number of commercial machines now being offered for low speed adaptive cutting of fabrics (Anon, 1991; Hertzog, 1994; Khoury, 1991). The work is, of course, not restricted to the textile industry. Automatic fish processing is cited by Wu (1993) for example, where machine vision is used to guide a cutting knife for filleting cod roe.

Water jet cutting has been applied for many years across a range of industries but has not found favour in textiles. Cutting of floor coverings, tiles, metals and plastics are typical application areas. Water jet cutting and laser cutting are often compared (Anon, 1987; Mosavi, 1987).

Laser energy has been a favoured means of cutting, heat treating and welding of metals for some years now. A number of commercial machines now exist for cutting textiles (Roux, 1988). The favoured device is usually a CO[sub]2 gas laser which produces adequate edge quality and can be cost effective, but is limited to single ply cutting of mainly large workpieces on special machine frames, to provide a large work area, for example sail cloth. Laser beam workpiece velocities can be in excess of 2 metres/sec with laser power ranging up to 2 kW. Lower power, sealed tube laser technology is gaining acceptance, providing relatively low cost and low maintenance operation for applications such as plastics, paper and textile processing (Cummings, 1987).

The application of lasers has been mainly limited to date, to applications where pre-programmed cutting of shapes are carried out on fabrics where the location of the pattern needs only to be approximately located to the cut shape. This is typical of the majority of cutting situations tackled by laser, water jet of reciprocating knife where a combination of two-axis motion of workpiece and/or tool is effected by computer numerical control. These systems are in widespread use today.

There are other applications in textile cutting where an accurate cut must be made relative to the pattern on the fabric. Since the fabric pattern is often subject to a small variance which is unknown at a detail level, pre-programmed shape cutting cannot be applied. The idea of adapting the cutting path of the laser beam acting on information relating to reference features within the actual fabric pattern, determined by machine vision, becomes attractive to provide a solution for these types of application.

This approach, in principle, would expand the range of options open to lace fabric and lingerie garment designers since it would remove the restrictions currently imposed by the high costs associated with the cutting of shaped piece parts accurately to the local material pattern. These are impossible to cut automatically at present, and therefore must be cut by hand using scissors, with corresponding cost penalties. Lace "scalloping" or edge cutting is a case worthy of further examination to illustrate the technology approach and corresponding benefits.

As with all high technology proposals, the question, is a solution feasible, both from a technological and economic viewpoint, needs to be answered.

The following case study presents work undertaken at Loughborough University during the period 1991-1995 to establish if vision directed laser cutting of complex patterned, elastic materials might be realisable. The ideas presented here are exclusively for lace cutting, but the work not only demonstrates feasibility, it should inspire other researchers across a range of applications.

Case study

The following case study is presented from work undertaken during a three-year SERC/ACME directorate funded research programme and a further one year of industrial support.

It was envisaged from the outset of this work that integration of the key candidate solution technologies would be essential in order to achieve the desired goal.

Cutting path determination

The starting point for the project was to develop a means of determining the edge which must be cut along (i.e. the cutting path), in order to achieve separation of the lace patterned breadth from the waste backing mesh. The chosen solution must be fast enough and accurate to achieve 1 m/sec and 0.2 mm resolution along the executed cutting path. The solution must also be robust (i.e. must not lose tracking registration easily) due to the complexity of the lace pattern and delicate nature of the web material (mainly Nylon 66, sometimes with elastomeric threads). Studies of existing contact based edge detection systems showed that this approach not only limited the accuracy of feature following but, more fundamentally, restricted operating speeds to typically 0.2 m/sec or less due to elastic deformation of the fabric, caused by resistance in the edge following features, leading to excessive rucking up of the fabric, hence the need for constant operator interaction. A non-contact sensing system was envisaged so that interference of the edge determination system with the fabric would not be an issue.

Machine vision, using a line scan camera format seemed the most promising solution. With this approach only the pattern information immediately before the cutting zone is captured, which is in contrast to conventional area (or frame) based imaging where large amounts of data must be acquired and processed in order to analyse and determine the area of interest. This slows down the image processing function considerably.

This is of crucial importance since the lace is never stable geometrically and also, the pattern position changes with respect to time as the lace is transported through the cutting zone. Therefore, frame based imaging not only slows down the acquisition and analysis part of the imaging function, it also has the problem that most of the data in the frame is unreliable. The alternative approach of imaging and cutting a stationary portion of lace is simply not fast enough for industry requirements.

A 2048 pixel line scan camera can image an area of 200 mm wide across the lace web x 0.1 mm along the web. A personal computer (PC) based system would cost typically Pounds 8k excluding application software. Data capture speed of such a system is typically one line scan every 0.1 msec (i.e. 10 kHz); this would in principle allow 0.1 mm spatial resolution along the feed direction at 1 m/sec. This is therefore a fast and high resolution image acquisition system. The weakness in this approach is that the line scan camera has only the current line of information; the history of the local advancing pattern image must be maintained by computer software and/or hardware.

The solution to this local imaging problem on the proof of concept machine developed at Loughborough employs a specially developed One Dimensional Binary Cross-Correlation Algorithm. This operates on a subset of each new camera line scan bit pattern and correlates this for a statistical match against a digitised image of a pattern repeat for each different lace web (Figure 5). This approach is sufficiently fast to achieve matching within one image pixel (0.2 mm) at speeds of 1 m/sec, using a hybrid combination of hardware accelerator and digital signal processor (DSP) (Jackson, 1996; King, 1994).

Cutting along the determined path

Examination of a wide range of lace patterns indicates the difficulty of cutting along such an edge at high speeds. Elastic distortion of the fabric takes place during cutting with contact based machinery. This distortion is the key factor which influences cutting accuracy and is also the reason why a skilled operator is required to be in constant attention to prevent the fabric jamming in the edge following guides, cutting device or transport feed rollers. Contact cutting based systems are limited to typically 0.5 m/sec on easy to cut sections of rigid lace. When cutting more complex edge geometry and deeper scallops of the more difficult patterns, the actual lace feed velocity must be reduced to 0.1 m/sec often, especially on stretch lace. Non-contact cutting is therefore essential in order to avoid gross distortion of the fabric in order to achieve higher processing speeds.

Analysis of the edge profile of representative breadths of lace reveals that the cutting device must exceed 4 m/sec relative to the lace material in order to achieve a target web feed velocity of 1 m/sec on some patterns, while 3 m/sec relative velocity will be prevalent. Rapid changes in velocity are also needed. This rules out water jet based systems due to the relatively high inertia linear slides required to manipulate the jet or the workpiece. In contrast the laser is very attractive since the laser beam has no mass, therefore the only inertia is that of the rotary mirror based laser beam manipulation system. This is considerably lower than a linear slide. With this approach high velocities and high accelerations can be attained (Jackson, 1994).

Test work undertaken reveals that a significant laser power is required to cut through waste net fibres (0.15 mm diameter Nylon 66), typically 250 Watts at 6 m/sec (0.2 mm spot diameter at focus). This is due to the open nature of the backing mesh, with no opportunity to integrate the laser energy into the workpiece, such as when cutting paper.Nevertheless, laser cutting is attractive due to speed of response and edge sealing, to prevent fraying, as well as general cutting detail and accuracy. A further advantage is that the laser cutting tool does not dull during cutting and therefore does not attract costs associated with re-sharpening, as with knife or disc based cutting systems. A conceptual design for laser beam manipulation and an integrated vision system is shown in Figure 6.

Integrated fabric handling system

The bulk lace supply is typically three to four metres wide, 100-150 metres long when presented to the cutting room. The new solution must accommodate lace in this form, ideally. A number of design studies were undertaken to examine potential machine concepts, which included alternative proposals for lace handling systems. Following this work, a prototype test rig was designed and constructed to allow a number of different lace transport configurations to be studied. The results of this work showed that lace can be transported at speeds up to at least 2 m/sec, in a manageable way in a bulk form of 1.8 metres wide by 100 metres long.

The concept is shown in Figure 7, where the lace is joined as an offset endless belt so as to make a continuous strip presented to the vision and cutting zones set 10 mm apart along a 45 degree sloping table. This natural fall of the lace down the cutting table is crucial to smooth feeding. The right hand edge of the lace web is tracked laterally by two optical edge sensors situated prior to the drive roller and cutting table. These sensors and associated decision making logic action air jets to correct the position of the lace web relative to the centre line of the vision and cutting planes. The correct functioning of this tracking system is essential to maintain the lace breadth in the local field of operation on both the camera and cutting systems. Without the fine control (typically +/- 5 mm tracking variance) the vision algorithm and laser beam manipulation system could not be optimised for high speed yet accurate operation (Preston, 1994).

Results

The integrated proof of concept demonstrator (Figure 8) has been tested at 1 m/sec single path tracking and cutting on stretch lace. The tracking system is implemented by a hybrid hardware correlator and DSP system. Further tests have demonstrated dual path tracking and cutting using a tracking function based purely on two DSPs and an additional laser beam system for execution of the second cutting path (Table I). This approach has a maximum speed limited by the computational power of the DSP platform.

When a second hardware correlator is built these two features together (1 m/sec and dual path cutting) will form the basis of an advanced production machine. The increase in output over a conventional machine is potentially, truly formidable, and shows the positive advantage of this machine. Consider the data in Table II, which compares the performance of this new machine concept with that of the traditional lace scalloping machine.

Figure 9 shows a typical laser scalloped sample of lace. The cutting quality is self evident and has been judged by industry experts as being commercially acceptable.

Future view

The prospect for this type of advanced production machinery is good. But it is only when a number of technologies are successfully integrated at research level and realised at industrial level that the full impact will be felt. An increased number of systems integrating companies are necessary. These organisations must have close ties with research establishments; only then will the flow of ideas from the research sector to industry be swift and successful.

Conclusions

An advanced production machine proof of concept demonstrator for processing of patterned flexible materials such as lace has been realised. The system embodies machine vision with a fast, robust tracking algorithm capable of determining a pre-defined cutting path, even when the material is significantly distorted, in real-time, at 1 m/sec lace web feed velocity. A high speed, fast response CO[sub]2 laser beam manipulation system responds to the determined cutting path data derived from the material pattern by the vision system, again in real-time to an accuracy of0.2 mm typically, at lace feed velocities of1 m/sec. Dual path cutting has also been successfully realised. These results, which represent a substantial advancement of "smart" machine technology, provide a crucial example for other researchers and engineers to draw upon. An international patent has been filed by the authors and co-workers (Preston, 1993).

References

1. Ameziane, M., Bonnet, P. and Postaire, J.G. (1985, "Vision applied to a cutting process in the textile industry", 7th International Conference on Automated Inspection and Production Control, Birmingham, UK, March, IFS (Publ.) Ltd, Kempston, UK, pp. 185-93.

2. Anon (1987, "Water-jet and lasers take on tough materials", Plastics Technology.

3. Anon (1991, "Intelligent laser cutter", Via Combattenti, 11th International Textile Machinery Exhibition, Hanover.

4. Bradshaw, M. (1995, "The application of machine vision to the automatic inspection of knitted fabrics", in Acar, M. (Guest ed.), Special Issue - Mechatronics in Textile Industries, Mechatronics, Vol. 5 Nos 2/3, Pergamon Press, New York, NY, pp. 233-43.

5. Cummings, M.B. (1987, "Low-cost laser systems can replace mechanical equipment in packaging applications", Packaging Technology, March/April.

6. Hertzog (1994, Machinery Catalogue.

7. International Trade Statistics Yearbook (1992), Vol. II, Trade by Commodity, Department of Economic and Social Development Statistical Office, United Nations, New York, NY, 1993.

8. Jackson, M.R., Preston, M.E. and Tao, L.G. (1996, "Real-time cutting path determination using machine vision-based incremental pattern tracking", Real-Time Imaging, Vol. 2 No. 4, pp. 249-64, Academic Press, New York, NY.

9. Jackson, M.R., Preston, M.E., Yang, S. and Tao, L.G. (1994, "Laser cutting parameters for the high speed cutting of net fabrics", Proceedings of Europto (SPIE/EOS), The European Symposium on Optics for Productivity in Manufacturing, Frankfurt/Main, FRG.

10. Khoury, J. (1991, "PC-based vision in laser cutting of upholstery fabric", Photonics Spectra.

11. Kimoto, I. and Yamafuji, K. (1986, "Automation of sealant painting and lace cutting using pattern tracking techniques", The International Journal of Advanced Manufacturing Technology, Vol. 1 No. 4, pp. 101-7.

12. King, T.G. and Tao, L.G. (1994, "An incremental real-time pattern tracking algorithm for line-scan camera application", Mechatronics, Vol. 4 No. 5, pp. 503-16, Pergamon Press, New York, NY.

13. Mosavi, R.K. (1987, "Comparing laser and water jet cutting", Lasers & Optronics.

14. Preston, M.E., Jackson, M.R., Yang, S., Tao, L.G. and King, T.G. (1994, "The design of high speed lace cutting system using a vision controlled CO2 laser", International Conference on Machine Automation - Mechatronics Spells Profitability, Proceedings of the ICMA '94..

15. Preston, M.E., Jackson, M.R. and King, T.G. (1993, "Automatic operations on materials", International Patent Application No. PCT/GB93/01663..

16. Roux, R. (1988, "The laser in the French clothing industry", Laser Focus World, Annual Review of Laser Processing..

17. Russell, R.A. and Wong, P. (1989, "An application of computer vision to lace cutting", Robotics and Autonomous Systems, Vol. 5, pp. 91-6.

18. Sanby, C., Norton-Wayne, L. and Harwood, R. (1995, "The automatic inspection of lace by machine vision", in Acar, M. (Guest ed.), Special Issue - Mechatronics in Textile Industries, Mechatronics, Vol. 5 Nos 2/3, Pergamon Press, New York, NY, pp. 215-31.

19. Wu, Q.M. and deSilva, C.W. (1993, "Automatic adjustment of the cutting position of a vision-based fish processing machine", IEEE Pac. Rim, Conf. Commun. Comput. Signal Process., pp. 702-5.

20. Yunming, L. and Dong, L. (1993, "Rule-based fabric inspection using machine vision", Journal of China Textile University (Eng. Ed.), Vol. 10 No. 3.