A Deep Learning-Based Cyber-Physical System for the Detection of Defective Capsules Using a Transfer Learning Strategy

Authors

Johannes Breitenbach, University of BayreuthFollow
Jan Gross, Aalen UniversityFollow
Ricardo Buettner, University of BayreuthFollow
Hermann Baumgartl, Aalen UniversityFollow
Samuel Bayerlein, Aalen UniversityFollow
Dennis Flathau, Aalen UniversityFollow
Daniel Sauter, Aalen UniversityFollow

Paper Number

1847

Abstract

This study addresses the relevance of high-quality pharmaceutical products and proposes a cyber-physical system-based quality control approach that differentiates between usable and non-usable capsules. With a balanced accuracy of 95.71 percent, we set a new benchmark in detecting defective capsules (e.g., scratches) in the pharmaceutical domain automatically using image classification with convolutional neural networks. Following the growing need for efficiency and quality in manufactured products, we contribute to the technical evolution within the industrial sector. Our model does not require further data processing, allowing implementation in different production environments. While our approach advances the efficiency of industrial processes such as quality control to identify product defects in real-time, our approach represents a low-cost alternative to comparatively expensive and technical complex inspections (e.g., X-ray) which is of high practical relevance.

Comments

Paper Number 1847

Recommended Citation

Breitenbach, Johannes; Gross, Jan; Buettner, Ricardo; Baumgartl, Hermann; Bayerlein, Samuel; Flathau, Dennis; and Sauter, Daniel, "A Deep Learning-Based Cyber-Physical System for the Detection of Defective Capsules Using a Transfer Learning Strategy" (2022). PACIS 2022 Proceedings. 287.
https://aisel.aisnet.org/pacis2022/287