| dc.description.abstract | Cyber-physical systems (CPS) involve the integration of physical hardware with computational processes, requiring the cyber components to align with the underlying physics for system correctness. To improve efficiency and scalability, decentralized CPS, known as distributed cyber-physical systems (DCPS), utilize communication networks to allow edge nodes to combine local processing with information from neighbors. This enables better decision-making and faster achievement of operational goals. Safety-critical CPS must meet stringent performance and safety requirements, even in the event of cyber component failures. Designing fault-tolerant systems for DCPS is challenging due to added complexity, communication coordination, and resource constraints. This research aims to design fault-tolerant DCPS that surpass traditional fault-tolerant computing approaches by considering the physical aspect of the system and distributed algorithms executed by the cyber layer. It employs a bottom-up methodology and explores proactive and reactive fault-tolerance. Proactive fault-tolerance involves anticipating risks during design and adhering to safety specifications, while reactive fault-tolerance employs detection and mitigation measures during runtime.
The research addresses several research questions. First, it develops an automated model generation tool for verifying distributed CPS software using standard modeling techniques and timing and interaction patterns between agents. It uses a software framework called Resilient Information Architecture Platform for Smart Grid (RIAPS) and generates Timed Automata (TA) models compatible with the UPPAAL model checker. A resilient deployment framework is then implemented, considering user-defined constraints to optimize redundancy or cost and resource limits. Experiments with a microgrid energy management application demonstrate the impact of deployment configuration on resilience and resource demands. Additionally, a configurable virtual communication topology framework called Topology Manager for Peer-to-Peer Links (TopLinkMgr) is introduced, allowing users to specify connectivity configurations and deploy peer-to-peer applications. A fault-tolerant self-adaptive virtual topology management algorithm, Bounded Path Dissemination, enhances convergence speed, accuracy, and robustness against node failures while minimizing communication bandwidth. Lastly, the research discusses a hierarchical design philosophy for fault tolerance in distributed CPS, utilizing the RIAPS framework. It demonstrates fault detection and mitigation mechanisms and their integration into application-specific functions. A distributed load shedding algorithm is used as an example, and its performance is evaluated under various fault scenarios.
By bridging fault tolerance in computing systems and physics-based systems like the power grid, this research highlights the interconnections and feedback effects between these areas. It emphasizes the importance of combining proactive and reactive design philosophies to achieve highly resilient and dependable CPS. | |
