Cyber-Physical Fault Detection in Autonomous Electric Vehicle Platoons with Lateral Motion

Health monitoring systems for detecting and clas-sifying cyber-physical faults are essential for connected au-tonomous vehicle (CAV) platoons, given the vulnerabilities associ-ated with their wireless communication and information-sharing capabilities. Various model-based, signal-based, and data-driven methods have been developed for CAV health monitoring in the literature. However, research considering lateral dynamics and associated lateral manoeuvres, such as lane changes, is still limited. This paper discusses the importance of considering lateral dynamics in the context of CAV cyber-physical fault detection and classification, and investigates the use of multi-head attention (MHA) and graph neural network (GNN) models for this purpose. Our results demonstrate that attention-based classifiers outperform other methods in most scenarios, achieving approximately 73% accuracy. In contrast, while GNN-based methods like the graph attention network capture localized dependencies in simplified cyber-attack cases, their performance remains lower at around 65 %. These findings indicate the need for further research on robust attack classifiers and integrated fault detection frameworks for complex cyber-physical fault scenarios in electric vehicle platoon operations.