MINERVA – Secure collaborative utilization of machine tool data using privacy enhancing technologies
Advancements in digitalization within the machine tool industry enable the collection of increasingly large amounts of data. To exploit this potential for data-driven innovations, the data must be collected and evaluated across companies. The iwb is working with the Fraunhofer Institute for Applied and Integrated Security AISEC (consortium leader), Hufschmied Zerspanungssysteme GmbH, and Siemens AG on the research project "Secure collaborative utilization of machine tool data using privacy-enhancing technologies (MINERVA)". The project researches and develops new technologies for a data infrastructure that guarantees the security of sensitive machine data using modern data protection technologies.
Motivation
The advancing digitalization and connectivity in production offers the machine tool industry the opportunity to make leaps in innovation by collecting and evaluating relevant production data across companies. However, this data contains intellectual property of the companies, which can put these companies at a competitive disadvantage if the data is disclosed. For this reason, machine tool data has so far hardly been exchanged between companies, which means that innovations in the field of big data cannot be transferred to the machine tool industry at present.
Research objective
The use of privacy enhancing technologies (PET) intends to enable machine operators to share their machine data with machine manufacturers without disclosing their intellectual property. The machine manufacturer then collects the anonymized machine operator's machine data. With this large data basis, the machine manufacturer can train generalized machine learning models, which can in turn be employed by the machine operator. The main objective of the iwb's research project is a model for detecting tool wear, which can be trained centrally using anonymized machine tool data and which determines the current tool condition.
Approach
In the first stage of the research project, the iwb is in charge of selecting suitable machine learning models. The focus here is on the condition monitoring of milling tools, which predict the continuous wear of these tools, depending on machine data and external sensor data. Concurrently, a criticality definition for machine data is developed. This allows machine operators to label different data categories, such as the spindle motor current or workpiece geometries, with different levels of criticality. Depending on this criticality assessment, the data is protected with PETs of different levels so that the intellectual property of the machine operators remains protected, but the information content of the data is retained as much as possible. The research project consortium then determines suitable PETs for the use case. These include differential privacy algorithms, trusted execution environments and federated learning strategies.
In addition, the consortium will also implement the necessary software infrastructure for the collection and systematic storage of machine data, the aggregation and evaluation of the data by the machine manufacturer and the final use of condition monitoring on the machines. The software is then integrated into the Siemens Xcelerator (formerly: MindSphere) environment to enable low-cost integration of the system for machine operators. The overall system will be validated by using it on machine tools at the iwb and at the research partner Hufschmied.
Results
With the help of the developed and validated methods within the MINERVA research project, machine operators can improve the efficiency and effectiveness of their production systems through the use of machine learning models. Machine manufacturers can systematically identify potentials for improvement in their products. This first-time systematic use of PETs ensures the sovereignty of machine operators over their machine data and protects their intellectual property. Nevertheless, machine operators will be able to benefit from the advantages of data-driven machine learning processes.
Acknowledgements
MINERVA is funded by the Federal Ministry of Education and Research (BMBF) as part of the program "IoT Security in Smart Home, Production and Sensitive Infrastructures" under the funding code 16KIS1805. We would like to thank the BMBF for the funding and the VDI/VDE Innovation + Technology GmbH for the support and for the good and trusting cooperation.