The failure to provide textiles within defect tolerance limits can lead to whole batch recalls, resulting in costly customer claims and downstream production delays. Poor defect management is also a major source of industrial waste. The potential addressable market for patterned textile vision systems is estimated to be well over €5.8 billion when considering on loom, on knitting machine, in process and final finished fabric inspection.
A ground-breaking and unique new fabric inspection technique for accurately detecting the most subtle of defects on patterned fabrics during high-speed production has been developed by Shelton Vision Ltd., Leicester/UK.
The patent-pending system has been integrated into the company’s WebSpector platform – a powerful plain fabric inspection system – and validated through factory trials on a purpose-built full scale in-house demonstration system with sophisticated fabric transport capabilities. A first system has already been ordered by a manufacturer of both plain and patterned fabrics, including camouflage, in Colombia.
This follows the successful conclusion of a 21-month Innovate UK project in which techniques for the resolution of complex pattern deformations were developed by machine vision and computer scientists in the company, backed up by the machine vision and robotics department at Loughborough University, Loughborough/UK.
Traditional methods for defect detection rely on human inspection, with detection rates under 65%, while the Shelton WebSpector machine vision system offers a sophisticated platform for automated defect detection of over 97%, but until now has been restricted to plain textiles.
»What our new system basically does is essentially make the ‘good’ pattern invisible to the detection software. We have developed template matching techniques for the resolution of complex pattern deformations in order for the system to pick up defects in the pattern as well as underlying defects.«
While pattern matching and neural network approaches have previously been tried for patterned textiles, they have failed to provide a practical solution due to the extreme complexity associated with pattern matching on deformable substrates like textiles, as well as the time required to train a neural network for each pattern type.
The full system consists of a camera and lighting system for optimum image capture at high speed and associated image processing hardware; self-training software utilizing statistical analysis to automate the system configuration for new textile products; an advanced suite of defect detection algorithms for the detection of all textile defect types; an AI-driven defect classification system which learns and automates defect naming in real time, as well as a real time defect grading capability based on client decision rules; and a system for recording and retrieving complete roll map images for subsequent review and quality control.
The generation of textile roll maps with complete defect data allows for an optimized textile cut plan, improved downstream processing and quality assurance.