Machine vision in the inspection of patterned textile webs

Considerable effort has been expended in research into machine visual defect detection in textile fabrics. Textile fabric inspection differs from many other industrial inspection tasks in that the product is usually in web form. It also requires very high resolution imaging to enable defects as small as a single missing yam to be detected. These factors make the use of line scan cameras virtually mandatory. A further problem in fabric inspection is that the material (and pattern) to be inspected can have a high degree of acceptable dimensional variation. This rules out simple template matching based inspection approaches. In order to enable commercially viable inspection throughputs (typically 1 m/s) to be obtained these combined factors suggest that `incremental´ approaches to processing the image are necessary. This paper discusses recent work in textile web inspection, illustrated by examples of the work on the high speed inspection of textile lace being carried out by the Mechatronics Research Group at the University of Birmingham