Can computer vision and AI techniques impact the quality control system for textile yarns? (Review)

Abstract

Abstract Quality of textile yarns is very important for determining how well the finished textile product will perform and appear. In the past few years, the use of Computer Vision (CV) and Artificial Intelligence (AI) have changed the way yarn inspections take place, and the industry is currently transitioning from offline laboratory-based inspections to real-time, in-line monitoring systems. This paper provides a detailed overview of traditional yarn inspection approaches (capacitive and photoelectric sensors) as well as new CV and AI-based techniques as they used in detecting yarn defects, estimating yarn diameter, measuring mass irregularity, and evaluating yarn hairiness. Recent developments in deep learning techniques (Convolutional Neural Networks (CNN), object detection architectures various YOLO models, Artificial Neural Network (ANN), and hybrid systems) are highlighted in this review. By comparing the various approaches to yarn inspection based on reported performance metrics, the preferred method for each application, and the potential for an industrially applicable method, this review has identified the main trends, benefits and limitations of each method discussed. Through this comparison, it was determined that deep learning-based inspection methods provided higher levels of accuracy and reliability, yet there are many challenges that continue to exist for data availability and access to real-time data and generalizing results across yarn types through deep learning. Finally, this analysis concluded with a discussion of currently unaddressed research opportunities in developing intelligent and scalable yarn inspection methods.