Temperature-based quality analysis in ultrasonic welding of copper sheets with microstructural joint evaluation and machine learning methods

Ultrasonic metal welding (USMW) is a highly attractive joining technology due to high energy efficiency and solid-state joint formation. Various joining solutions for conductor materials can be realized with USMW. Still, a big challenge for complex industrial applications is an adequate process monitoring that allows to cope with inevitable and complex process fluctuations. In this work, the suitability of contactless temperature measurements for process monitoring of copper sheet welding is examined and compared with the suitability of vibration measurements by means of machine learning methods. Different sensor signals acquired during welding on a metrological test rig are used for predicting the tensile shear strength of the joint. Results show that quality predictions based on temperatures exceed the state-of-the-art monitoring based on the welding energy. Yet, solely temperature-based predictions are exceeded by quality predictions based on either welding machine signals or tool vibration measurements. To further explore temperature-based quality analysis, joint microstructure analyses are carried out. These reveal concurring joint formation mechanisms associated with the mechanical and thermal process domains. To finally cover both domains in quality prediction, a sensor-fusion-based regression model is set up relying on vibration and temperature measurements. This fusion model exceeds all previously considered regression models with a mean absolute percentage error of 7.4% on the test data set. These results stress the importance of both process domains and suggest the combination of temperature and vibration measurements as a good starting point for future industrial monitoring of USMW processes.