CNN-based classification of the laser assembly process for ultra-small batteries

• Published in: Journal of mechanical science and technology Vol. 37; no. 12; pp. 6181 - 6192
• Main Authors: Kim, Beomjin, Park, Wonshik, Kim, Kihyun, Kim, Hyo-Young
• Format: Journal Article
• Language: English
• Published: Seoul Korean Society of Mechanical Engineers 01.12.2023; Springer Nature B.V; 대한기계학회
• Subjects: Artificial neural networks; Classification; Control; Data augmentation; Dynamical Systems; Engineering; Feature extraction; Image enhancement; Image filters; Image processing; Industrial and Production Engineering; Inspection; Laser beam welding; Lasers; Machine learning; Mechanical Engineering; Microbatteries; Original Article; Vibration; 기계공학; Deep learning; CNN; Lead frame; Image filter; Vision inspect; Ultimate tensile strength; Small battery
• ISSN: 1738-494X; 1976-3824
• DOI: 10.1007/s12206-023-2411-4

In this study, we facilitated the electrode welding of a micro-battery utilizing a laser through the application of a convolutional neural network (CNN) for the classification of micro-battery welding quality, utilizing a dataset comprised of battery-welded images. While prior studies focused on enhancing CNN performance through virtual image generation and conversion, our approach distinguishes itself by optimizing the CNN’s performance through the adjustment of hyperparameters within the feature extraction section and the application of an image filter. To address insufficient image data, data augmentation and image shift techniques were implemented. The investigation delved into the influence of hyperparameters on CNN performance during the inspection of welding images, where the grayscale filter exhibited commendable performance in the context of battery welding images. Evaluation of the classification performance was conducted using a confusion matrix, revealing accurate identification of the two welding conditions. The experiments conducted in this study not only established the viability of the laser welding process but also demonstrated the potential of vision inspection using deep learning, presenting a practical solution for the micro-battery process.