Abstract:
Unmanned Aerial Vehicles (UAVs) are becoming an integral part of cyberphysicalsystems (CPS), but their increasing reliance on interconnectedcommunication makes them highly vulnerable to both system faults and cyberintrusions. Previous research has largely focused on either fault detection orintrusion detection in isolation, leaving a gap for an integrated frameworkcapable of handling both threats simultaneously while distinguishing normalUAV operations. In this paper, we introduce Secure-AI-FDD, a hybrid deeplearning framework that unifies fault detection, intrusion detection, and normaloperation classification. The novelty of our approach lies in leveragingcontrastive learning using TabNet as a backbone to structure UAV data into threesuper-classes: fault, intrusion, and normal operation. Within each super-class,we apply a nested pairing technique to generate subclass pairs (e.g., specificintrusion types grouped under intrusion, fault categories under fault, etc.),enabling the model to learn fine- grained distinctions. To enable experimentation,we curated a synthetic UAV dataset, carefully formatted to support contrastivetraining. Experimental evaluation demonstrates that Secure-AI-FDD achieves anoverall accuracy of 78%, effectively detecting and classifying UAV states acrossfaults, intrusions, and normal operational subclasses. This work contributes asignificant advancement toward building robust, AI- driven, hybrid anomalydetection systems for UAV-CPS environments, enhancing their reliability,resilience, and trustworthiness.
Page(s):
109-109
DOI:
DOI not available
Published:
Journal: 4th International Conference of Sciences “Revamped Scientific Outlook of 21st Century, 2025” , November 12,2025, Volume: 1, Issue: 1, Year: 2025
Keywords:
intrusion detection
,
Cyberphysical systems
,
Fault Detection
,
Contrastive learning
,
UAV security