A DATASPACE-DRIVEN EDGE COMPUTING AND FEDERATED LEARNING FRAMEWORK FOR SENSORY TOOLING SYSTEMS

Abstract

Although manufacturing becomes increasingly data-driven, sensor data is typically available only locally, with limited integration across systems or organizations. This lack of available, heterogeneous data prevents the development of robust AI models to predict key machining outcomes such as tool wear and surface roughness. To overcome this challenge, a cooperative approach that leverages Dataspaces and Compute-to-Data is proposed. This enables access to heterogeneous data and the subsequent development of robust algorithms without sharing the actual data, thereby keeping know-how and intellectual property secure. Therfore, an edge computing architecture is proposed, integrating sensory tool holders with machine tools and enabling high-frequency measurements of acceleration during machining. These measurements are transmitted via Bluetooth from the tool to a stationary transceiver unit, labeled and subsequently stored on a local data storage. The data is then made accessible through a Dataspace. To prevent stakeholders from accessing raw data of each other, the AI model training is conducted in Compute-to-Data environments. The approach is illustrated through an exemplary case study, focusing on the prediction of surface roughness during machining. Thus, this work contributes to the cooperative creation of intelligent machine tools. Future research can build on this approach and explore more industrialized implementations.