Investigation of monitoring methods for ultrasonic metal welding

Abstract

Ultrasonic metal welding (UMW) is becoming a more broadly used technology for joining ductile materials, especially in the electric vehicle sector. The process, however, lacks monitoring capabilities that would improve confidence in the repeatability of welded joints. Often only destructive testing is used for quality evaluation. This method of inspection is insufficient for a production line and can lead to high scrap rates and failure to identify all poorly welded components. The work discussed in this paper aims to close the gap of weld quality evaluation for UMW through in-process monitoring. Multiple sensors were installed in line with a linear ultrasonic metal welder. Current, voltage, frequency, shear force, and displacement of the horn both laterally and vertically were monitored throughout welding trials. Parameters for expected overweld, underweld, and acceptable weld qualities were selected through screening trials and design of experiment (DOE) methods. Twenty welds from each of these three quality sets were made with all monitoring tools for several rounds of testing. Destructive analysis was used to confirm the weld quality for each experiment. This included peel testing by hand for qualitative results and mechanical peel testing for quantifiable results, as well as metallographic analysis of the cross section and weld interface. Signal analysis, performed for each set of sensor data, extracted unique features that may be correlated to the input weld parameters and resulting weld quality. Machine learning techniques were applied on these features to classify weld quality based on in-process monitoring data. Algorithms predicted weld quality with over 90% accuracy.