Doctoral Defence: Aleksandar MATOVIC

Abstract:

In an era of growing cyber threats, where critical infrastructure such as power grids, healthcare systems, and transportation networks are increasingly targeted by sophisticated attacks, the urgency of designing resilient cyber-physical-systems (CPS) has never been more pressing. Cyber-physical systems form the very backbone of our modern society, and their disruption can have catastrophic con- sequences, ranging from economic losses to threats to human life. Against this background, this thesis addresses two fundamental challenges in the field of CPS: firstly, enhancing resilience against a wide range of threats by leveraging application knowledge to improve on the costs of resilience, ranging from accidental system failures to carefully coordinated cyber-attacks, and secondly, ensuring the adaptability of these systems in the face of dynamic and unpredictable operational environments.

The first challenge addressed is the improvement of system resilience. We introduce a novel Consensual Resilient Control (CRC) method to systematically convert stateful control tasks into statelessly recoverable ones, by leveraging consensually updated shared state introduced in the thesis is central to this challenge. CRC significantly improves the performance of control task replication by exploiting the inherent stability of many systems to tolerate occasional missed control task deadlines. This approach rejuvenates replicas within each control cycle, improving system resilience and operational efficiency. This not only enables rapid recovery but also significantly reduces the overheads associated with traditional replication methods, particularly in environments prone to cold start effects. The effectiveness of CRC is not just theoretical, but demonstrated through practical applications, such as our implementation in the custom-built inverted pendulum system, which demonstrates the robustness of the CRC in unpredictable environments and its ability to efficiently maintain system resilience with fewer resources.

The second challenge is to ensure system adaptability in the face of changing operational conditions. To this end, the thesis presents the AεGIS architecture, a solution that seamlessly integrates dual control systems to optimise performance while maintaining safety. The adaptive nature of AεGIS is particularly beneficial in open environments where CPSs are exposed to a wide range of disturbances. The architecture’s minimal switching overhead and its utility in complex tasks such as environmental monitoring illustrate its practical importance in enhancing system robustness.