Looking Beyond Irregularities-
Enhancing Machine Vision Technology to Inspect Carpets

PROSPECT has been developed to inspect carpets as they come off the production line. The machine vision system is able to identify defects in carpets more effectively and more cheaply than ever before. Using PROSPECT, fewer faults go undetected and less waste is produced, generating both environmental and economic benefits. 

PROSPECT was developed by Kjaergaard Industri Automatik A/S, experts in machine vision and quality control, in collaboration with the High Performance Computing and Networking (HPCN) experts EPCC and the carpet manufacturers Foamtex. The development project was funded by the European Commission, and runs from 1 August 1998 to 31 March 1999. It is presently in its test phase. 

Kjaergaard gained experience with PROSPECT by first using the technology to inspect colour print on cans using line-scan cameras. However, the challenge was to enhance the technology to cope with the many irregularities that are present in weaving products.  

The key to the PROSPECT system is the use of powerful but inexpensive computing technology to analyse the image of the carpet. The system must be able to process the images quickly in order to keep up with the speed of the carpet as it comes off the production line. The system tackles this challenge using state-of-the-art Digital Signal Processor (DSP) technology and vision systems.  

The ability to use artificial vision and image processing techniques to detect such minute faults across an entire product has only recently become available. The computing power required is significant and until now has been very expensive. Not anymore: EPCC is running the ENTICE project as part of a network of technology transfer nodes (TTN) established by the European Commission to help companies, and in particular small and medium sized enterprises, to exploit High Performance Computing and Networking (HPCN) technology.  

Left: The system can also detect different colour errors.   Right: A thread bursting will show in the picture and result in an error message from the system.

In carpet manufacture, faults can occur with colour, pattern, or pile; or faults can be introduced by the carpet backing process. A prototype of the PROSPECT system has been implemented and tested in the Foamtex factory. The most expensive faults from the manufacturerÕs point of view were chosen from a range of over 20 possible faults in carpet production. The system was set to detect the four faults most frequently reported as customer complaints. These faults include tiny areas - from 2 x 6 millimetres - where some of the carpet pile is missing or where there has been an unsuccessful repair of missing pile. Such faults are difficult to see with the human eye but their detection is vital for customer satisfaction.

Work carried out by EPCC has shown that HPCN technology can be used, for example, to expand the system to accommodate wider carpets, faster production, and to detect different types of faults. This ability to ÒscaleÓ the system to cope with the workload makes it versatile and easy to upgrade. Detection rates of over 95% have been achieved when using this type of technology to detect faults in other manufacturing processes, and the partners hope that the carpet industry will soon be able to see the same benefit.

Contact Details: 
Kjaergaard
- el & industri automatik a/s
Vestergade 1-9 
DK-8723 L¿sning 
Info@kia.dk 
www.kia.dk