Enabling secure and self-sovereign machine learning model exchange in manufacturing data spaces

Abstract

With the rapid digital transformation of manufacturing, vast amounts of data are being generated and analyzed to uncover valuable patterns in areas such as energy efficiency, predictive maintenance, production scheduling etc. However, much of this data and the intelligence derived from it remain isolated within individual companies. This is strongly influenced by companies reluctance to share data due to concerns over privacy and security associated with the commercially sensitive information. As a result, the potential shared value that can be derived from a richer, larger pool of data and intelligence across multiple companies remains untapped. While solutions such as federated learning exist to address privacy and security issues, strong governance so that the privacy is preserved is crucial to its successful implementation. Currently, there is a lack of software infrastructure that guarantees data sovereignty and governance for data owners in this space. This paper introduces COllaboRative Data Space (CORDS), a framework that enables companies to engage in a machine learning model-sharing ecosystem, providing full control over the access and usage of their data. Aligned with the European Data Space initiative, CORDS aims to foster trusted collaboration by providing a software infrastructure constituting a set of tools for both intra and inter-organization data asset management and ML model exchange. To the best of our knowledge, CORDS is the first minimum viable data space (MVDS) designed to address the broader challenges of sovereignty, interoperability, compliance & governance in cross-party ML model sharing. This paper also highlights the value of data sharing by applying CORDS to a use-case focused on improving energy efficiency in manufacturing. Extensive performance evaluation showcases CORDS’ utility in securely managing data assets and facilitating machine learning model exchanges. CORDS is available as open-source software, supporting further research and practical applications of trusted data spaces in both academia and industry.