Abstarct: Unmanned Underwater Vehicles are widely used in commercial, scientific, and military missions for various purposes. What makes this technology challenging is the increasing mission duration and unknown environment. Underwater vehicles are equipped with a sensor system to accomplish the specific mission it has been commanded to complete. Faulty sensors may cause process performance degradation or fatal accidents. Therefore, it is essential that a fault detection scheme can be developed so as to be able to detect and identify possible faults in the system as early as possible. In this dissertation, we study the observer-baxsed fault detection and isolation problem for linear time-invariant dynamic systems with an emphasis on robustness. After introducing some basic definitions, the problem of model-baxsed fault detection is introduced. This is followed by a summary of different approaches in generating diagnostic residual signals. Observer-baxsed methods are further studied. Three observer structures are discussed, including Disturbance Observer, Unknown Input Observer and Proportional Integral Observer. Fault detection baxsed on various observer structures is studied. The robustness issues are then defined in connection to fault detection. An important focus is to isolate the effect of disturbance on residual signals for robust fault detection. In this dissertation we present a robust sensor fault detection and isolation scheme for an unmanned underwater vehicle (UUV). An unknown input observer (UIO) is designed for a linearized model of NPS UUV. As an innovation in this field, we introduce a PSO approach to design of optimized UIOs. Simulation results are presented to demonstrate the effectiveness of proposed robust fault detection schemes.