A guidance system for robotic welding based on an improved YOLOv5 algorithm with a RealSense depth camera

Area of Science:

• Robotics and Automation
• Computer Vision
• Manufacturing Technology

Background:

• Laser vision sensors (LVS) require manual intervention for robotic welding guidance, limiting autonomy and productivity.
• Existing vision-based systems often struggle with precise weld groove detection and positioning.

Purpose of the Study:

• To develop an autonomous robotic welding guidance system using an improved YOLOv5 algorithm and a RealSense Depth Camera.
• To enhance weld groove detection accuracy and improve the overall autonomy and efficiency of robotic welding.

Main Methods:

• An improved YOLOv5 algorithm incorporating a coordinate attention (CA) module for enhanced weld groove detection.
• Integration of a RealSense Depth Camera to acquire depth information for precise weld groove positioning.
• A system combining depth camera data with YOLOv5 predictions to guide the robot and welding torch along the weld centerline.

Main Results:

• Experimental validation demonstrated the feasibility of the proposed robotic welding guidance system.
• Maximum guidance error of 2.9 mm at a 300 mm distance, with percentage errors within 2% for distances ranging from 0.3 to 2 m.
• The system effectively combines the large-field positioning accuracy of depth cameras with the high precision of LVS.

Conclusions:

• The developed system enables autonomous robotic welding by accurately identifying and tracking weld grooves without manual intervention.
• The integration of improved YOLOv5 and depth sensing significantly enhances robotic welding guidance accuracy and efficiency.
• This approach offers a robust solution for automated welding, improving productivity and reducing reliance on human operators.