A Scene-Aware Confidence-Guided YOLO Framework for Robust Weld Seam Tracking Under Intense Arc Interference

Area of Science:

• Robotics and Automation
• Computer Vision
• Artificial Intelligence

Background:

• Intelligent robotic welding systems face challenges in accurate weld seam tracking due to severe arc interference, causing radiometric saturation and visual degradation.
• Existing YOLO-based tracking methods often fail under strong arc interference because their assumption of detection confidence reflecting localization reliability breaks down.

Purpose of the Study:

• To propose a robust weld seam tracking framework, Scene-Aware Confidence-Guided YOLO (SACG-YOLO), to overcome limitations of existing methods under severe arc interference.
• To enhance the accuracy, robustness, and continuity of weld seam tracking in complex industrial welding environments.

Main Methods:

• Introduced a scene-aware confidence estimator that assesses welding scene observability by considering structural consistency and photometric variation.
• Fused scene confidence with detector confidence to evaluate measurement reliability, adaptively regulating the Kalman filter update process.
• Implemented a strategy to incorporate reliable observations and reject unreliable measurements, favoring motion prediction for improved tracking continuity.

Main Results:

• The SACG-YOLO framework significantly outperformed conventional YOLO-based tracking methods in industrial welding image sequences.
• Under stable and weak arc interference, average tracking error decreased from 0.492 mm to 0.155 mm, and maximum error reduced from 0.829 mm to 0.354 mm.
• Under severe arc interference, where standalone YOLO tracking failed, SACG-YOLO maintained continuous weld seam localization with a maximum error of 0.406 mm.

Conclusions:

• Explicitly modeling measurement reliability through scene-aware confidence estimation substantially improves weld seam tracking accuracy, robustness, and continuity.
• The SACG-YOLO framework offers a viable solution for reliable robotic welding in challenging industrial conditions with significant visual interference.