Artificial intelligence methods for in-process high-speed image analysis in laser beam welding of hairpins

Abstract

In the production of modern electric drives for battery electric vehicles, hairpin technology is used to increase the copper fill factor in the stator of a permanently excited synchronous machine. A central process in the production of these stators is the contacting of the hairpin ends by means of laser beam welding. This welding process is characterized by geometric and process-related deviations from previous process steps, which influence the result of the welded joint. It is desirable to find an in-process method for monitoring. As part of the process monitoring of welded joints, high-speed camera images are often used to detect weld spatter. These can be detected by a program based on a static algorithm. For this reason, a feasibility analysis is performed regarding the application of AI for the detection of spatters, in which the methods of semantic segmentation and single-image classification prove to be useful. In a preliminary experiment, three base networks for each of the two methods are evaluated with respect to the best training results. The single-image classification method will then be extended by a subsequent static algorithm, so that a hybrid use of AI and static algorithm will be investigated. The evaluation and final comparison of all evaluation methods is performed using data from a welding experiment. It turns out that the hybrid approach of single-image classification and static algorithm has numerous advantages in the detection of spatter compared to semantic segmentation and the static algorithm.