Abstract

The operation of Autonomous Unmanned Vehicles (AUVs) that is used for environment protection, risk evaluation and plan determination for emergency, are among the most important and challenging problems. An area that has received much attention for use of AUVs is in underwater applications where much work remains to be done to equip AUVs with systems to steer them accurately and reliably in harsh marine environments. Design of control strategies for AUVs is very challenging as compared to other systems due to their operational environment (ocean). Particularly when hydrodynamic parameters uncertainties are to be integrated into both the controller design as well as AUVs nonlinear dynamics. On the other hand, AUVs like all other mechanical systems are prone to faults. Dealing effectively with faulty situations for mechanical systems is an important consideration since faults can result in abnormal operation or even a failure. Hence, fault tolerant and fault-accommodating methods in the controller design are among active research topics for maintaining the reliability of complex AUV control systems.
The objective of this thesis is to develop a nonlinear Model Predictive Control (MPC) that requires solving an online Quadratic Programming (QP) problem by using a Recurrent Neural Network (RNN). Also, an Extended Kalman Filter (EKF) is integrated with the developed scheme to provide the MPC algorithm with the system states estimates as well as a nonlinear prediction. This hybrid control approach utilizes both the mathematical model of the system as well as the adaptive nature of the intelligent technique through neural networks. The reason behind the selection of MPC is to benefit from its main capability in optimization within the current time slots while taking future time slots into consideration. The proposed control method is integrated with EKF which is an appropriate method for state estimation and data reconciliation of nonlinear systems. In order to address the high performance runtime cost of solving the MPC problem (formulated as a quadratic programming problem), an RNN is developed that has a low model complexity as well as good performance in real-time implementation. The proposed method is first developed to control an AUV following a desired trajectory. Since the problem of trajectory tracking and path following of AUVs exhibit nonlinear behavior, the effectiveness of the developed MPC-RNN algorithm is studied in comparison with two other control system methods, namely the linear MPC using Kalman Filter (KF) and the conventional nonlinear MPC using the EKF.
In order to guarantee the fault-tolerant features of our proposed control method when faced with severe actuator faults, the developed MPC-RNN scheme is integrated with a dual Extended Kalman Filter that is used for a combined estimation of AUV states and parameters. The actuator faults are defined as the system parameters that are to be estimated online by the dual-EKF. Therefore, the developed Active Fault-Tolerant Control (AFTC) strategy is then applied to an AUV faced with loss of effectiveness (LOE) actuator fault scenarios while following a trajectory. Analysis and discussions regarding the comparison of the proposed AFTC method with Fault-Tolerant Nonlinear Model Predictive Control (FTNMPC) algorithm are presented in this work. The proposed approach to AFTC exploits the advantages of the MPC-RNN algorithm properties as well as accounting explicitly for severe control actuator faults in the nonlinear AUV model with uncertainties that are formulated by the MPC.