The number of cells required varies depending on the vehicle type and model, but the assembly process of connecting the anode and cathode is crucial for ensuring a reliable electrical and physical connection for each cell. Cell connections are typically made using ultrasonic or laser welding techniques. In the electric automotive industry. The proportion of laser welding and AI technologies is gradually increasing due to its fast process speed and high reproducibility. However, despite the growing demand for AI technologies to monitor these welding processes and predict welding quality, ensuring consistent quality in high-speed welding, particularly for dissimilar materials, remains challenging.


This study developed and evaluated an algorithm for classifying penetration depths in dissimilar material joining (stainless steel and copper) using one-dimensional convolutional neural network (CNN) models. Spectral sensing data were recorded from a high-speed spectrometer and two photodiodes installed coaxially and off-axially to provide in-situ process information. By analyzing macro-sectional images of the welded specimens, the penetration depths were classified into four categories. The CNN-based algorithm utilizing spectrometer data achieved over 99% accuracy, while the photodiode-based model showed a lower classification performance. This study aimed to compare and analyze the accuracy of optical sensing in high-speed laser welding and to propose the more suitable sensor for dissimilar material applications.