Dynamically Distributed Real-Time Systems in Adaptive Manufacturing

The competitiveness of manufacturing companies depends on their ability to rapidly adopt technological innovations while meeting growing customer demands. Emerging technologies such as artificial intelligence, digital twins, and the Industrial Internet of Things require flexible, adaptive infrastructures that still satisfy strict requirements of time-critical industrial applications. Industrial infrastructures have historically grown hardware-centric, whereas Industry 4.0 technologies are driven by software-defined solutions, leading to convergence of information and operational technology. Virtualization and containerization enable hardware-independent software provisioning, and automated orchestration ensures flexibility and scalability. However, these systems do not yet support strict industrial real-time requirements. This dissertation presents a holistic solution for containerized real-time applications in deterministic, mixed-criticality Ethernet networks, introducing the components needed to ensure real-time capability across computation, communication, and orchestration. Regarding computation, deterministic execution of containerized applications on distributed nodes is investigated using real-time Linux technologies. For deterministic communication, Time-Sensitive Networking (TSN) solutions are examined, and a TSN-capable virtual bridge plugin for Kubernetes is implemented to adapt TSN to container communication. A configuration solution is designed and prototypically implemented to configure the plugin at node level and across remaining network devices, following existing TSN standards and integrated into Kubernetes. To enable automated configuration in large-scale networks, microservice-based solutions suitable for seamless Kubernetes integration are evaluated, and configuration capabilities are extended to include virtual bridges. Finally, addressing orchestration of real-time-critical containerized applications, scheduling enhancements are introduced: a scheduling algorithm is developed and integrated into Kubernetes to select suitable deployment nodes, extending its capability to ensure end-to-end latency constraints for interdependent, distributed applications. The implemented components are validated for industrial adaptability by measuring performance and fulfillment of real-time requirements, using a robot-aided laser structuring use case with control loops of varying criticality as the evaluation example.