Abstract

Autonomous Underwater Vehicles (AUVs) are capable of performing a variety of deepwater marine applications as in multiple mobile robots and cooperative robot reconnaissance. Due to the environment that AUVs operate in, fault detection and isolation as well as the formation control of AUVs are more challenging than other Multi-Agent Systems (MASs). In this thesis, two main challenges are tackled.

We first investigate the formation control and fault accommodation algorithms for AUVs in presence of abnormal events such as faults and communication attacks in any of the team members. These undesirable events can prevent the entire team to achieve a safe,
reliable, and efficient performance while executing underwater mission tasks. For instance, AUVs may face unexpected actuator/sensor faults and the communication between AUVs
can be compromised, and consequently make the entire multi-agent system vulnerable to cyber-attacks. Moreover, a possible deception attack on network system may have a negative
impact on the environment and more importantly the national security. Furthermore, there are certain requirements for speed, position or depth of the AUV team. For this reason, we propose a distributed fault detection scheme that is able to detect and isolate faults in AUVs while maintaining their formation under security constraints. The effects of faults and communication attacks with a control theoretical perspective will be studied.

Another contribution of this thesis is to study a state estimation problem for a linear dynamical system in presence of a Bias Injection Attack (BIA). For this purpose, a Kalman Filter (KF) is used, where we show that the impact of an attack can be analyzed as the solution of a quadratically constrained problem for which the exact solution can be found efficiently. We also introduce a lower bound for the attack impact in terms of the number of compromised actuators and a combination of sensors and actuators. The theoretical findings are accompanied by simulation results and numerical can study examples.