Table Of ContentsNext Page

Use of parallel processing in quality control

Dr Wayne Moore

School of Applied Sciences, CSU-Mitchell

Abstract.

This paper points out some of the applications of high speed computing in the food processing industry. It highlights the possible role of one aspect of high speed computing, that of parallel processing. A problem encountered in can sealing is presented and its possible solution by the use of transputers is described.

1. Introduction

The production of high quality food products is a very large industry in Australia and has the widest scope for scientific research and development. Research can encompass everything from biotechnology and statistics to engineering and computing. The latter section of research is essentially into the food packaging and processing section of the industry.

High speed computing to food production can have a great deal of applications. Some of the application areas are:

• production line monitoring;

• multiple sensor processing and AI applications;

• critical sections of processing that may need redundancy in its computing requirements.

The application techniques and hardware that are being developed in Australia and elsewhere are:

• vision systems (inspection for quality control and for grading of product)

• robotics (moving and handling of product)

• interfacing with Programmable Logic Controllers (PLCs)

As computers become more powerful the application areas become more diverse and manufacturing and process engineering techniques require extremely fast response times and calculating ability from computers.

2. Application Problems

The work at CSU-Mitchell has been motivated by a problem that arose in consultation with Uncle Ben's of Wodonga. This company is part of the MARS group of companies and processes dog and cat food for the domestic market and overseas markets. The problems that arise in pet food processing have their direct analogies with human food production methods.

The particular problem that was presented for solution was that of can sealing. Cans of pet food are sealed in at one point in the production line and are then passed on for further cooking before being packed into boxes for shipment. It is the point in the production line between the sealing and further cooking that requires an inspection point for faulty can seals. This entails the detection of 'product' (e.g. gravy, food) overflowing from the tins.

This problem needs high speed processing, data handling and data capture. The problem can be summarised in the following manner:

• cans are rotating at 200-300 revs/min when emerging from the sealer;

• cans moving on production line at 1200 cans/min (at Uncle Ben's);

• a scanning device needs to scan each can at least 2000 to 3000 times to be able to detect faulty seals.

The scan of a can results in a 'thin' line being available for use by suitable algorithms and a decision on whether to reject or accept it is made. In some industries, such as the brewing industry, it is possible to have bottles (or cans) moving at up to 2500 cans/min.

Other problems/areas that may lend themselves to the application of high speed computing are:

• detecting the 'm' on M&M sweets - quality control for the Japanese trade;

• neural-net applications in decision making;

• harvesting and general farming (robotics)

3. Computing Considerations

As already remarked, the solution to the 'can-sealing' problem lends itself to high speed computing. However, this can lead to some problems and a list of these is presented below.

High speed processing problems:

• usually need large and costly machines (e.g. CRAY or Connection machine);

• most computers not suitable for factory floor (not robust);

• not particularly adaptable to different types of problems (coherency);

• usually of a large size.

The 'can-sealing' problem requires very high sampling rates of the cans to make the image available to the computer for processing. The computer needs then to perform quite complicated image recognition algorithms to detect the faulty seals. Current high speed sequential processing is not available in an economic or practical form. One solution to this problem is to use a parallel processor.

Parallel processing is the division of a larger problem into smaller sections each of which can be 'solved' concurrently on a computer. Each of these sections is 'executed' on a number of computers. A practical and economic form of parallel processing is available from the INMOS company. This company manufactures a form of multiple instruction, multiple data (MIMD) parallel processor called a transputer.

The use of transputers is a possible way to the solution of the sampling problem in that transputers are economical and the problem lends itself to 'sectioning' and farming out of computational tasks.

The transputer has the following characteristics:

• transputers are fast (T800-10MIPS, H9000-150 MIPS)

• transputers are scaleable (i.e. can link many of them together easily)

• availability of 'good' language support (C, FORTRAN, OCCAM, ADA)

• a lot of worldwide development is taking place

As an example of the speed of transputers and its relative ease in applications that lend themselves to parallel processing, Table 1 presents the results of using 30 transputers to a thin-film problem using a simulated annealing technique.

The Table shows the speedup factor that can be achieved by increasing the number of transputers. This shows the 'scaleability' of such a parallel processing system which makes it an ideal solution to partitioning an image that has been captured from the production line and analysing it in a number of processors. This scaleability means that extra performance can be easily achieved by the simple addition of more processors. For further details refer to Glass and Morf(1) and Moore and Glass(2).

Table 1. Speedup (performance) values for various combinations of integration points and transputers

No. of
Transputers

Speed-up Factor

(21 pts)

(52 pts)

(94 pts)

(200 pts)

(1000 pts)

           

1

1.0

1.0

1.0

1.0

1.0

2

1.85

1.96

-

-

-

3

2.8

-

-

-

-

4

3.2

3.8

3.8

3.9

3.98

5

3.6

-

-

-

-

6

4.2

5.0

5.5

5.7

5.95

7

5.3

-

-

-

-

8

5.2

6.4

7.1

7.6

7.92

9

5.2

-

-

-

-

10

4.9

7.1

8.3

9.4

9.86

12

6.0

8.0

9.9

11.69

 

16

5.4

8.9

12.0

13.4

15.42

20

5.2

9.9

13.1

16.5

19.19

24

6.3

9.6

13.8

17.4

22.54

30

5.5

10.6

13.6

20.2

27.09

Table 2 illustrates the relative performance of well known sequential computers compared to the use of transputers. The last line shows the relative speedup of transputers compared to a DEC 6310 computer, a quite powerful machine.

This work was performed at Landis and Gyr, at Zug in Switzerland. It is hoped that Australian companies can be as far sighted as such research and development oriented companies as Landis and Gyr.

Table 2. Comparison of a 5-layer test calculation for thin-film optical layer design for a number of different machines.

Computer

Number of processors

Time (sec.) for integration pts

   

21

52

94

200

1000

             

VAX-11/780

1

171

413

736

1560

7915

VAX 3600 Series

           

workstation

1

74

179

320

675

3348

VAX-6310

1

51

123

218

467

2264

INMOS T800

1

140

341

613

1300

6484

T800 Array

30

25

32

45

64

240

Speedup Factor
30-T800/VAX6310

 

2

4

5

7

10

4. Research Efforts at CSU-Mitchell

A team at CSU-Mitchell is working on the application of parallel processing to the 'can-sealing' problem and other more generic types of problems.

The group is the Centre for Parallel Processing and Industrial Control(3) and has one publication on image processing to its credit but hopes to involve a number of research students to image processing problems. The group is also in the early stages of setting up a prototype system of conveyor belt and associated cameras and image capture equipment for research students and staff. Some members of the team are also engaged in developing parallel processing software for Fast Fourier Transform (FFT) analysis for applications in image processing, turbulent flow and other numerical methods. Work is also being done on interfacing transputers with PLCs (this will combine image processing on the conveyor belt prototype system with robotics.

The Centre is also fortunate in having the support of industry.

Some of the technical objectives that are envisaged for the solution of the can sealing problem are:

• development of guidelines on the use of parallel processing to scanning procedures;

• make the findings easily accessible to other researchers in the industry;

• development of some software tools for use in the development, specification and execution of parallel programs.

Some obstacles to be overcome in the application of parallel processing to food processing problems are:

• industrial timidity in using new ideas (e.g. parallel processing). The MARS group seem to be less prone to this problem than others!);

• lack of any 'centre of expertise' in parallel processing with large systems (problem of fragmentation);

• lack of any suitable training programs for individual application.

5. Summary

Food processing stands to benefit from the application of parallel processing to the packaging end of the food production line. Future developments end of the food line might involve an 'earlier' application of high speed computing to the production end of the food line. A team is working on some of the problems of quality control in the food processing industry at Charles Sturt University (Centre for Parallel Processing and Industrial Control) using parallel processor technology.

References

1. Glass, A.S. and Morf, R. (1990). "Optmizing the Performance of Spectrally Selective Photodiodes by Simulated Annealing Techniques". Sensors and Actuators, Vol.21, pp.564-569.

2. Moore, W.E. and Glass, A.S. (1990). 'Parallelization of Application Programs'. The Transputer in Australia. Proceedings of the 3rd Australian Transputer and Occam User Group Conference, Sydney, June 1990.

3. Sheridan, P., Hintz, T. and Moore, W.E.T. (1991). 'Spiral Architecture in Machine Vision'. Proceedings of the 4th Australian Transputer and Occam User Group Conference, Canberra, September 1991.

Previous PageTop Of PageNext Page