Towards data-driven quality monitoring for advanced metal inert gas welding processes in body-in-white

In recent years, numerous monitoring approaches have been developed in the field of intelligent welding manufacturing to predict quality-related characteristics using process data and artificial intelligence-based techniques. While most investigations have focused on welding steel with conventional gas metal arc welding processes, the welding of aluminum and its alloys using advanced process variants has been less explored. This work addresses this gap by investigating data-driven methods for fault diagnosis and detection in an advanced metal inert gas welding process commonly used in body-in-white manufacturing. To this end, electrical, acoustic, and spectroscopic signals were recorded from numerous welding tests simulating typical fault causes. Various predictive models, ranging from traditional machine learning algorithms to state-of-the-art deep learning techniques, were trained and evaluated for classifying faulty seams and identifying their root causes. The results demonstrate that combining sensor data enhances the performance of predictive models compared to using individual sensors alone. However, a deep learning approach based solely on electrical signals emerged as the best solution for both use cases, considering both the results and practical aspects. Overall, the experiments highlight the significant potential of data-driven techniques to enhance quality monitoring in advanced MIG welding processes, promoting their more widespread adoption in body-in-white manufacturing.