Semi-Supervised Training for Positioning of Welding Seams

Area of Science:

• Robotics
• Computer Vision
• Machine Learning

Background:

• Robotic welding relies on vision-based systems for seam placement.
• Traditional machine vision methods struggle with manufacturing and imaging variations.
• Supervised deep learning requires extensive labeled data, which is costly and time-consuming.

Purpose of the Study:

• To develop a semi-supervised learning algorithm for robust key-point detection in robotic welding seam placement.
• To improve accuracy and robustness without extensive manual data labeling.
• To address the limitations of traditional methods in dynamic manufacturing environments.

Main Methods:

• A novel semi-supervised learning algorithm for key-point detection was developed.
• The algorithm was designed to work with a minimal number of labeled images (as few as fifteen).
• Full image resolution was utilized to enhance key-point detection accuracy.

Main Results:

• The proposed semi-supervised approach demonstrated robustness with limited labeled data.
• The method achieved high accuracy in detecting key-points for welding seam placement.
• The algorithm effectively handled variations in manufacturing and imaging conditions.

Conclusions:

• Semi-supervised learning offers a viable solution for accurate and robust robotic welding seam detection.
• The developed algorithm significantly reduces the need for expert data labeling.
• This approach enhances the reliability and efficiency of automated welding processes.