Yarn inspection and sorting system using robotic vision and machine learning

The increasing demand for automation in the textile industry, particularly in quality inspection processes, underscores the need for intelligent and cost effective solutions. Conventional methods of yarn classification and sorting remain labor-intensive, time-consuming, and susceptible to human error, resulting in inconsistent quality control. This study introduces an automated system for yarn inspection and sorting that integrates robotic vision, machine learning, and position-based visual servoing (PBVS) for real-time motion control. The proposed system combines Raspberry Pi-based machine learning with computer vision utilizing a 4-degree-of-freedom (4-DOF) robotic manipulator and a webcam, enabling precise pick-and-place operations based on yarn classification into four categories: good, striped, moldy, and dirty. Experimental results validate the system’s effectiveness, achieving an average deviation of 0.375 mm along the x-axis, 0.69 mm along the y-axis, and 0.675 mm along the z-axis, resulting in an overall position error of 0.58 mm. These results demonstrate the system’s robustness and reliability in dynamic industrial environments. The novelty of this research lies in leveraging a low-cost embedded architecture with advanced visual servoing for textile automation, reducing operational errors, improving efficiency, and supporting industry 4.0 adoption.