Date of Award

8-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Automotive Engineering

Committee Chair/Advisor

Ass. Prof Behnaz Papari

Committee Member

Dr Benjamin Lawler

Committee Member

Dr Robert Prucka

Committee Member

Dr Christopher Edrington

Abstract

This dissertation advances the cybersecurity of hybrid tracked vehicles (HTVs) and ship power systems (SPSs) by developing innovative cyber-attack models and corresponding defence frameworks. First, we formulate stealthy false-data-injection attacks (FDIAs) on HTV energy-management systems as a partially observable Markov decision process (POMDP) solved via deep reinforcement learning. A novel sniffing-based reward function guides the attacker to covertly degrade battery capacity and energy efficiency, which we evaluate using custom stealth–impact metrics and a sliding-window anomaly detector (Isolation Forest with Dynamic Time Warping). Additionally, we model sophisticated control-layer attacks in HTVs, including reinforcement-learning-optimised replay attacks and denial-of-service (DoS) attacks targeting generator-speed sensors to disrupt the energy supply. These scenarios, implemented in MATLAB/Simulink, establish challenging benchmarks for developing resilient countermeasures.

To detect and mitigate such threats, we propose a dynamic trust-evaluation scheme grounded in Zero Trust Architecture principles. This framework fuses current and historical sensor behaviour in a multi-attribute model to identify anomalies under stealthy replay and DoS attack vectors in real time. Experimental validation on HTV systems demonstrates a detection accuracy of 96.35% with only 3.2ms latency. We further introduce hard- and soft-trust modes to quantify the trade-off between detection precision and responsiveness.

Finally, we extend our cybersecurity approach to medium-voltage DC ship power systems by designing a federated-learning-based anomaly-detection framework. Leveraging distributed controller data and real-time digital twins, this method detects and localises compromised agents via reconstruction-error analysis. Simulation results using OPAL-RT hardware-in-the-loop test data confirm its effectiveness in detecting the investigated cyberattacks. Collectively, these contributions provide a comprehensive security blueprint for mission-critical land and naval cyber-physical systems, significantly enhancing the resilience of defence and autonomous-vehicle platforms against advanced cyber threats.

Author ORCID Identifier

https://orcid.org/0000-0002-3451-9826

Available for download on Monday, August 31, 2026

Share

COinS