top of page
검색
  • 작성자 사진RAMO

Demo: ML-based Vision Classification in the Smart Brewing Factory Testbed


Authors: Sujin Kook, Sihun Baek, Hyelin Nam and Seong-Lyun Kim


This work is accepted to ICC demo 2022, yet published.


The main purpose of this paper is to introduce our implementation of 5G MEC on a factory, and make it as a part of 5G smart factory. We adopt the ML (Machine Learning) based vision classification, and implement it using 5G connectivity and MEC. In particular, our focus is how to deploy the so called split and federated learning [1], [2] on a real system to

support URLLC.

We utilized the ML based vision classification to a brewing factory for the first time to the best of our knowledge. Until now, due to fast production line speed, and stability, the brewing factory was difficult to be fully automated. Our target system is to classify recyclable bottles to detailed brands.


  • Bottle Classification

Breweries in Korea reuse recycled glass bottles to make new products. Therefore, empty bottles arrive in the factory, and these should be classified into usable/useless, and in more detail, into 8 different brands.

  • Inference system

To satisfy fast moving process of factory, we split the systmen into two parts with different deep learning models.

- Edge device distinguishes whether the bottle is recyclable or not with a smaller model. This tasks require a real-time service, so it is proper to process in an edge device.

- Cloud server, which has big computation resources distinguishes detailed brands with a bigger model.

  • Training system

We used Split Learning to collaborativel train the distributed neural network models on edge devices and the server. A set of lower layers of neural network model are stored on the edge device, and the remaining upper layers of the model are stored on the centralized server.

  • Testbed

By connecting a factory in Busan, 400 kilometers away, from our Yonsei University server in Seoul, we make a whole smart factory testbed. Commercial 5G has challenging to be used in factories due to limited uplink speed. So, we establish the network connecting which was constructed utilizing the KOrea Advanced Research Network (KOREN), a public institution-managed R&D testbed network infrastructure, for supporting the reliability of the communication.




Reference

[1] Y. Gao, M. Kim, S. Abuadbba, Y. Kim, C. Thapa, K. Kim, S. A. Camtepe, H. Kim, and S. Nepal, “End-to-end evaluation of federated learning and split learning for internet of things,” arXiv preprint arXiv:2003.13376, 2020

[2] C. Thapa, M. A. P. Chamikara, S. Camtepe, and L. Sun, “Splitfed: When federated learning meets split learning,” arXiv preprint arXiv:2004.12088, 2020.

댓글 0개

최근 게시물

전체 보기

[RAMO] 2024년도 한-EU 6G 과제 최종 선정

RAMO 연구실이 2024년도 한-EU 6G 과제에 최종 선정되었다. 본 과제는 연세대학교가 주관으로 연구하며 국내 및 유럽 기관들과 공동으로 연구에 진행한다. 국내에서는 서울대학교, 고려대학교, 한국전자통신연구원(ETRI), 엘지전자(주)가...

Kommentare


bottom of page