Federated learning to serve smart manufacturing process improvement while retaining control of confidential data

MUSKETEER, a H2020 funded project, offering an industrial data platform leveraging federated learning and other privacy preserving technologies, has released initial open source libraries enabling stakeholders to share and train AI models without compromising the confidentiality of their data.

The presence of a huge number of machines in industrial automation factories and the elevated cost of downtime, produce large expenses for production line maintenance. Getting a more accurate evaluation of robot performance helps to avoid damaging the production capacity contingently (by 5 to 20% in certain cases)[1]. But collecting data from different factories to build powerful AI tools can raise privacy issues. Data used to train AI models can be sensitive company data but also lead to personal data concerns (e.g. data can include information about operators working at the plant). The MUSKETEER platform offers a solution to tackle these issues mixing privacy preserving technologies (federated machine learning, etc.) while respecting sovereignty of the stakeholders as defined by IDSA standard[2].

The first project release concerns the MUSKETEER client package. It describes how to interact with the IBM MUSKETEER cloud platform for federated machine learning. Developers learn how to design and run federated machine learning algorithms on the platform:

https://github.com/IBM/Musketeer-Client

This package, which is tested in a manufacturing use case, is helping to solve complex challenges for the automotive sector:

  • Improving the welding quality assessment to develop predictive maintenance for robots while increasing product safety at the same time
  • Training a welding quality assessment algorithm on large datasets from multiple factories

To find out more about this story and discover the complete scenario along with user testimonials:

https://musketeer.eu/publications/

MUSKETEER project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 824988.

[1] https://www2.deloitte.com/... and operations/us cons predictivemaintenance.pdf  

[2] https://www.internationaldataspaces.org/… content/uploads/2019/03/IDS Reference Architecture Model 3.0.pdf

Über International Data Spaces e. V.

The International Data Spaces Association (IDSA) has defined a reference architecture and a global standard for creating and operating virtual data spaces. The IDS Architecture is based on commonly recognized data governance models facilitating secure exchange and easy linkage of data within business ecosystems. IDSA is a non-profit association with currently more than 120 members from numerous companies, organizations and research institutes across 20 countries.

Firmenkontakt und Herausgeber der Meldung:

International Data Spaces e. V.
Emil-Figge-Str. 80
44227 Dortmund
Telefon: +49 (231) 70096-501
http://www.internationaldataspaces.org

Ansprechpartner:
Gal Weiss
EU Programs & Partnerships Manager, IBM
Telefon: 353-87-693-6134
E-Mail: wgal@ie.ibm.com
Für die oben stehende Pressemitteilung ist allein der jeweils angegebene Herausgeber (siehe Firmenkontakt oben) verantwortlich. Dieser ist in der Regel auch Urheber des Pressetextes, sowie der angehängten Bild-, Ton-, Video-, Medien- und Informationsmaterialien. Die United News Network GmbH übernimmt keine Haftung für die Korrektheit oder Vollständigkeit der dargestellten Meldung. Auch bei Übertragungsfehlern oder anderen Störungen haftet sie nur im Fall von Vorsatz oder grober Fahrlässigkeit. Die Nutzung von hier archivierten Informationen zur Eigeninformation und redaktionellen Weiterverarbeitung ist in der Regel kostenfrei. Bitte klären Sie vor einer Weiterverwendung urheberrechtliche Fragen mit dem angegebenen Herausgeber. Eine systematische Speicherung dieser Daten sowie die Verwendung auch von Teilen dieses Datenbankwerks sind nur mit schriftlicher Genehmigung durch die United News Network GmbH gestattet.

counterpixel