Weld feature extraction method based on inter-frame constraint and dynamic template.

Area of Science:

• Robotics and Automation
• Intelligent Manufacturing Systems
• Computer Vision for Industrial Applications

Background:

• Intelligent manufacturing relies heavily on welding, with autonomous welding robots becoming prevalent.
• Active vision-based seam tracking systems offer high precision but struggle with environmental interferences like arc splash and workpiece deformation.
• Accurate weld feature extraction is crucial for reliable robotic welding performance.

Purpose of the Study:

• To develop a robust weld feature extraction method for intelligent welding robots.
• To address challenges posed by complex welding environments, including arc splash, object reflection, and local workpiece deformation.
• To improve the accuracy and reliability of seam tracking systems in dynamic industrial settings.

Main Methods:

• A novel weld feature extraction method combining laser line point and line characteristics.
• Utilizing inter-frame constraints and dynamic templates based on correlations and differences in adjacent video frames.
• Implementing a system designed to handle reflections, arc splash, and local deformations during the welding process.

Main Results:

• Achieved a distance error within three pixels for weld feature extraction.
• Demonstrated an average processing time of 9.6 milliseconds per frame.
• Successfully extracted weld features despite challenging conditions such as object reflection, arc splash, and local deformation.

Conclusions:

• The proposed method offers significant robustness for weld feature extraction in intelligent manufacturing.
• The inter-frame constraint and dynamic template approach effectively overcomes common environmental interferences.
• This technique enhances the reliability and applicability of seam tracking systems in complex robotic welding scenarios.