Method for recognition and positioning of composite three-way pipe fittings based on machine vision

Abstract

To address the problems of low accuracy and slow speed in the recognition and positioning of composite material tee fittings in industrial environments, this paper proposes the YOLOv8-CCS algorithm. The algorithm designs a CGF module to enhance local and contextual feature extraction, introduces a CPS attention mechanism that decomposes 2D global pooling into 1D feature encoding to preserve positional information, and replaces CIoU with the SIoU loss function to accelerate convergence and improve localization accuracy. The 2D coordinates of the fitting center are determined using detection bounding boxes, and 3D coordinates are obtained by combining depth information from a depth camera, with the introduction of a lens distortion correction model. Experimental results show that YOLOv8-CCS achieves an accuracy of 95.5% and 125 FPS on a self-constructed dataset, outperforming mainstream algorithms such as YOLOv8n and Faster-RCNN, with additional comparisons against Transformer-based architectures. Ablation studies validate the effectiveness of each module and report standard deviations. Furthermore, the integration scheme and cost advantages of the vision system with robots and PLCs are analyzed, and the deep integration with the grinding robot’s motion control and force feedback system is explored, envisioning a “perception-decision-execution” integrated intelligent workstation. Photothermal material-based self-cleaning coatings are expected to maintain long-term stable imaging of cameras in dusty environments. In summary, the proposed algorithm meets the positioning accuracy and real-time requirements for automated grinding of composite material tee fittings.