Development of in-process welding torch position control system using AI technology

The authors have developed a torch position control system for narrow groove automatic TIG welding. This system can detect the feature point (electrode, wire, groove wall, and weld pool) positions in a weld pool image, calculate the relative positions, and move the electrode and wire to the correct positions. In order to identify the wavelength range that is less susceptible to arc light when capturing weld pool images, spectroscopic analysis was performed and a 1000-nm bandpass filter was selected. Since the brightness distribution suitable for detection differs for each feature point, weld pool images were captured with multiple exposure times. In order to accurately detect the feature points of various weld pool images, AI technology (the pose estimation model DarkPose) that can improve detection accuracy by adding training data was used. When the detection models were evaluated, it was found that the electrode, wire, and groove wall were detected with high accuracy. The torch position control system using the developed feature point detection technology was implemented. The system accurately detected the feature point positions and moved the electrode and wire to the correct position when the feature point position was misaligned. Also, the processing speed of the system was sufficient for torch position control of actual automatic TIG welding.