Abstract

Ultrasonic metal welding is an efficient, solid-state welding process that is well established in the production of electrical equipment. However, isolated and untraceable failures occasionally occur in industrial production processes. According to the state of the art, these failures are due to a multitude of influencing variables, such as material and surface condition of the joining partners, as well as tool and fixture wear. This study presents quality prediction models based on measurements of mechanical vibrations of the welding machine using laser triangulation sensors, sonotrode penetration depth measurements, and machine internal welding power signals. The acquired data are processed and transformed into a random forest model to estimate the shear strength of the welds. The robustness of the prediction model has been successfully validated by welding experiments with significant external disturbances, such as surface roughness, contamination and material hardness. Within the framework of this study, the development of a stable and industrially applicable concept for process monitoring is demonstrated with a regression model that achieves a mean relative estimation error of 4.30% and a R2 value of 0.964. Furthermore, a classification model that determines the external disturbances for the individual welds was successfully validated, achieving a micro F1 value of 94% and a macro F1 value of 95%.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call