Abstract

The theory of Cyber-physical systems (CPSs) has applications in critical infrastructures such as smart grids, manufacturing systems, transportation systems, and autonomous systems such as Unmanned Aerial Vehicles(UAVs). In the CPS, there is a coordination between communication, computation, and control. The communication link in CPS can be subjected to malicious cyberattacks. On the other hand, the physical system in CPS can be faced with different faults such as sensor, actuator, and component faults. Therefore, two significant and challenging problems in CPS can be the detection of faults and cyberattacks. These two threats are intrinsically distinctive and need different strategies to deal with when they occur. This research mainly focuses on providing a methodology to detect and isolate faults and cyberattacks.
This work considers false data injection and replay attacks as security threats. Two different adaptive neural network-based detection methods are proposed in this thesis. These adaptive neural networks are able to detect, isolate, and estimate false data injection, and replay attacks.
Another contribution of this thesis is to provide a scheme for isolating faults and cyberattacks (false data injection and replay attacks) by using virtual sensors on the plant side, which makes the simultaneous detection of faults and false data injection cyberattacks possible.
A nonlinear model of a quadrotor is considered the case study, and the performance of the neural network-based schemes is evaluated through various numerical simulation scenarios.