Cloud–edge–end collaborative data-driven online nondestructive inspection system for yarn manufacturing using machine vision

The combination of machine vision with cloud–edge–end collaboration has emerged as a viable method for non-destructive inspection during yarn manufacturing. However, various challenges have to be resolved to achieve accurate inspection under the collaboration architecture in a large-scale and high-speed spinning scenario. To address these issues, a cloud–edge–end collaborative data-driven online vision inspection system for yarn manufacturing is proposed. In the end, a mobile imaging robot with an automatic calibration mechanism is designed to capture yarn images continuously from spinning lines during manufacturing. In the edge, a combination of image processing algorithms is proposed to obtain a continuous, vertical, and clear yarn image dataset. In the cloud, a deep learning-based recognition model is improved with a multi-level training strategy to output accurate values of yarn quality indexes. Finally, an integrated system is developed and validated in the practical ring spinning workshop. The experimental results, including practical application in the actual ring spinning workshop, demonstrate the effectiveness of the proposed system. The average accuracy of the proposed system for the yarn unevenness inspection exceeds 97.42% within a range of conditions (spindle speed ≤18,000 revolutions per minute, temperature 29.4–31.4℃, and relative humidity 44.5–50.5%).