Welding Seam Recognition and Trajectory Planning Based on Deep Learning in Electron Beam Welding

Area of Science:

• Materials Science and Engineering
• Robotics and Automation
• Computer Vision and Image Processing

Background:

• Vacuum electron beam welding presents significant weld recognition challenges due to dark environments and metal reflections.
• Existing methods struggle with accuracy and real-time performance in these complex industrial settings.

Purpose of the Study:

• To develop an improved hybrid algorithm for robust weld recognition in vacuum electron beam welding.
• To enhance the automation of electron beam welding through effective pre-weld visual guidance.

Main Methods:

• A hybrid algorithm combining YOLOv11-seg with adaptive Canny edge detection was proposed.
• Incorporated the UFO-ViT attention mechanism and EIoU loss function for network optimization.
• Utilized the Otsu method for adaptive threshold setting in Canny edge detection.

Main Results:

• The optimized model achieved an average precision (mAP) of 77.4%, a 9% improvement over the baseline.
• The system operates at 20 frames per second (FPS), ensuring real-time processing capabilities.
• Generated welding trajectories showed an average length deviation of less than 3 mm from actual welds.

Conclusions:

• The proposed hybrid algorithm significantly enhances weld recognition performance under complex conditions.
• This approach offers an effective solution for pre-weld visual guidance, advancing electron beam welding automation.
• The real-time performance and high accuracy make it suitable for industrial applications.