Abstract

Cyber-physical systems (CPSs) have revolutionized various domains, including smart grids, manufacturing, transportation systems, and autonomous vehicles such as Unmanned Aerial Vehicles (UAVs). While CPS configurations offer numerous advantages, they are particularly vulnerable to stealthy cyber-attacks due to their specific structure, in which the command and control center is far from the plant and operation side, making security a critical concern. Among these attacks, covert and replay attacks pose significant challenges for detection, particularly in UAV applications. This research focuses on detecting such stealthy attacks using two approaches: a model-based scenario where the mathematical model of the system is available and a data-driven scenario where only data is accessible, making it more practical for high-fidelity systems.
In the model-based approach, the research develops a coding design to enhance the detection functionality and expand its scope to meet security requirements. This framework strengthens the system’s ability to counter stealthy attacks effectively. On the other hand, for scenarios where only data is available and inherently vulnerable to cyber-attacks, an effective algorithm is designed
to detect and identify covert and replay attacks by assigning appropriate labels. This algorithm leverages neural network (NN) models for training and evaluation, ensuring high accuracy and
proficiency in attack detection.
To optimize computational efficiency, the algorithm employs feature selection techniques during the data preprocessing stage, minimizing the reliance on complex NN models. This not only
reduces computational resource consumption but also enhances the accuracy of the detection model. The effectiveness of the proposed methodologies is validated through simulations on a 6-degree-of-freedom quadrotor, a critical application highly susceptible to cyber-attacks. The results demonstrate the efficiency and reliability of the contributions in detecting and mitigating stealthy cyberattacks in CPS configurations.
This research provides a framework for improving the security of CPSs in both model-based and data-driven scenarios, contributing to safer and more resilient systems in critical applications.