Identifying welding skills for training and assistance with robot

Abstract This paper aims at identifying the differences between skilled and unskilled welders by analysing the position data obtained with a three-dimensional motion capture system. Using the position data of markers the tip point position of the torch is constructed. The kinematic data of the torch tip obtained from various skilled and unskilled performances is analysed. The analysis reveals that a skilled welder has better control of the torch tip with less variation of the position and speed in the three directions. Six criteria are developed to distinguish the skilled and unskilled performances. First, threshold values are determined for the three position and three speed variations. Then, these threshold values are used to determine alarm levels, which signify the degree of violation of the thresholds. Lastly, the alarm levels are used to construct a degree of skill for each performance. The performance of the six criteria are tested on the training and validation set. The criteria based on the variation of position along the welding line and the height from the welded material are the most successful. In the conclusion the paper discusses how to make use of the results in order to develop a physically interactive welding robot for training and real time assistance purposes.