Universitat Autonoma de Barcelona, Centre de Visió per Computador (CVC)
The method is aimed at detecting yarns defects prior to their entry in the textile manufacturing machine to avoid yarn breaks, or low textile final quality. It relies on yarns image acquisition and processing to identify the defects of all yarns entering the machine at once. A system setup and a software package have been developed to implement the method at production scale with an industrial partner.
The method consists in: > capturing a set of yarns using a high definition lineal camera, > splitting the image into sections to attenuate possible negative effects of non uniform lighting conditions, > and treating the images obtained using restrictive and less-restrictive criteria > The two images are then subtracted to each other, > and the defects on the yarns are detected by comparing the pixels of the differential image with predetermined defect patterns.
There are no influence of inhomogeneous or background lighting variations.
A combination of horizontal and vertical scanning can be implemented to enhance the detection reliability.
Yarn numbers, yarn breaking and defects such as deviated fibers, yarn separations or junctions can be effectively detected.
Innovative aspects and applications
> All yarns checked at once > High reliability and flexibility > Full traceability with yarn reports > Real-time detection errors
> Successfully tested on production lines
State of development
> Currently, two system setups are used industrially since several years: - 275+ threads - 90 m/min warp speed rate
> Software package developed including predetermined defect patterns library > Classification of error among three types (low, medium and high) > On line report per thread. Summary report at the end of the production
> Improvement of the error severity classification > Detection improvement via the development of machine learning algorithm > Improvement of the inspection time and number of threads via algorithms speed up