Description
Modern manufacturing increasingly relies on larger datasets to improve the efficiency
and reliability of Tool Condition Monitoring on the shop floor. Sharing data across
company borders could significantly improve the accuracy of predictive models, but
concerns about confidentiality and the risk of Intelectual Property leaks prevent such
collaboration.
This thesis explores the development of a secure data infrastructure that enables
participants to engage in collaborative analytics without compromising their sensitive
industrial data. Building on the concepts of Privacy-Enhancing Technologies, this work
introduces a tailored set of Confidentiality-Protecting Technologies that attempt to
obfuscate industrial machining data while maintaining analytical usability for tasks
like tool wear prediction.
This platform is based on a client-server architecture to simulate real-world data
exchange between participants and a trusted third party aggregator. At the core of
the system is an automated client-side pipeline that allows participants to extract,
transform, visualize and encrypt their CPT-enhanced data before transmission.
This work was evaluated using real industrial machining data, applying a selection of
privacy enhancing methods and analyzing their impact. Results show that while some
configurations preserve the usefulness of the data, others reduce prediction accuracy
or fail to sufficiently protect the original content. Visual comparisons and predictive
model tests helped assess the trade-off between privacy and utility.
This thesis highlights both the potential and the challenges of secure data sharing
in the industrial sector. While our proposed solution provides a functioning starting
point, further refinement is needed to strike a satisfying balance between privacy and
data usability.
|
Thesis topic in co-operation with the Fraunhofer Institute for Applied and Integrated Security AISEC, Garching