关键词: 机器视觉, 多层感知器, 霍夫圆检测, 自动换筒, 手眼标定

Abstract: In textile production, the replacement of bobbins is an unavoidable key process. Currently, most textile enterprises still employ manual bobbin replacement methods, which poses safety risks and is labor-intensive. The carbon fiber, known as the "black gold" of the 21st century, is a new type of fiber material with a carbon content exceeding 90%. Because of its light weight, high strength, and corrosion resistance, the carbon fiber has been widely used in various fields. In recent years, the government has been actively promoting the development of the carbon fiber industry. Both the “13th Five-Year Plan” and the “14th Five-Year Plan” have explicitly called for the strengthening of research and application of high-performance fibers and composite materials like carbon fibers. In carbon fiber weaving and production, the replacement of carbon fiber bobbins is a critical step. This paper explores methods to achieve automatic bobbin replacement, using carbon fiber bobbin replacement as a case study.
 To achieve the intelligent replacement of carbon fiber yarn bobbins, this paper proposes an automatic bobbin-changing method based on machine vision detection and robotic arm collaborative operation, and establishes a corresponding intelligent bobbin-changing system. The system is mainly divided into hardware and software parts. The hardware part includes an image acquisition module, a yarn rack device module, an upper computer module, and a robotic arm control module. The software part is responsible for recognizing the target object in the image and controlling the robotic arm. This paper mimics the yarn rack design of an actual factory and designs a yarn rack device suitable for laboratory settings. First, the image acquisition module is responsible for capturing and saving images; then, the upper computer module integrates the software programs of the entire system, which are used to monitor and determine the status of the yarn bobbin and transmit information to the robotic arm; finally, the robotic arm control module receives signals from the upper computer and completes the bobbin replacement according to the planned path. The image processing part of the system is based on an optimized Hough circle detection algorithm, incorporating the LM algorithm and monocular distance measurement principles to limit the radius range of the yarn bobbin, and adding a concentric circle detection mechanism to achieve more accurate bobbin positioning. In addition, a multi-layer perceptron (MLP) model is used to complete hand-eye calibration, determining the relationship between the image coordinates and the robotic arm base coordinates, thus obtaining the precise position of the robotic arm's end.
In the experimental tests, this paper addresses the sensitivity to ambient light and background noise by adding Gaussian noise to the captured raw images and adjusting the brightness (with parameter values of -50, 20, and 50). Through these data augmentation operations, it is verified that the optimized Hough circle detection algorithm possesses strong robustness and reliability, maintaining high detection accuracy in complex environments. Compared with the Random Forest and K-nearest Neighbor algorithms, MLP shows the best accuracy on the X/Y/Z axes, with mean square error controlled within 1.77 mm². The results indicate that this study achieves high precision and effectiveness in the collaborative work of machine vision and robotic arms, providing important technical support for the practical application of intelligent replacement systems.

Key words: machine vision, multilayer perceptron, Hough circle detection, automatic bobbin replacement, hand-eye calibration