Potentials of few-shot learning for quality monitoring in laser welding of hairpin windings

Raffin T, Mayr A, Baader M, Laube N, Kühl A, Franke J (2023)

Publication Type: Journal article

Publication year: 2023

Journal

Book Volume: 118

Pages Range: 901-906

DOI: 10.1016/j.procir.2023.06.155

Abstract

For the analysis of high-dimensional data in electromechanical manufacturing, data-driven techniques such as deep learning show great potential. However, the vast amount of required data poses a major barrier to the industrial adoption of said techniques. In the production of electric traction drives, this situation is further exacerbated by short innovation cycles. One of these innovations is hairpin technology, which is increasingly being employed by automotive manufacturers and suppliers. In contrast to classical deep learning approaches, few-shot learning aims at generalizing well despite limited data availability. This makes it ideal for the ramp-up phase of new production technologies for which no large legacy datasets exist yet. Therefore, this paper presents an empirical evaluation of few-shot learning techniques in the context of hairpin stator production. In doing so, we evaluate and benchmark several few-shot learning architectures and traditional approaches for quality monitoring of laser-welded hairpin windings based on image data.

Authors with CRIS profile

How to cite

APA:

Raffin, T., Mayr, A., Baader, M., Laube, N., Kühl, A., & Franke, J. (2023). Potentials of few-shot learning for quality monitoring in laser welding of hairpin windings. Procedia CIRP, 118, 901-906. https://dx.doi.org/10.1016/j.procir.2023.06.155

MLA:

Raffin, Tim, et al. "Potentials of few-shot learning for quality monitoring in laser welding of hairpin windings." Procedia CIRP 118 (2023): 901-906.

BibTeX: Download