Aspects of Module Placement in Machine Learning Operations for Cyber Physical Systems

The traditional field of industrial manufacturing is in the process of being revolutionized as machines become smart and processes are translated to and perfected by digital systems. The application of Machine Learning (ML) has established itself as a smart technology in the manufacturing industry. The optimal operation and training of ML applications requires a flexible, adaptable, and modular service infrastructure for industry 4.0. The objective of this work is the elaboration of requirements for ML integration in Cyber Physical System (CPS) and the added value of modularization in CPS-related ML tasks. As the paradigm of Machine Learning Operations (MLOps) be-comes firmly established, principles for determining an optimized placement of the different ML pipeline modules in distributed systems are required. A collaborative cross-company use-case management framework KOSMoS is used to discuss selected benefits of using MLOps in industrial scenarios.